3 - Jacobs University
3 - Jacobs University
3 - Jacobs University
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
Second-Party Quality Auditing<br />
in the Automotive Industry<br />
In-depth Methodology Analysis<br />
by<br />
Borislav Hadzhiev<br />
(borislav.hadzhiev@volkswagen.de)<br />
a thesis for conferral of a Doctor of Philosophy in<br />
Communications, Systems and Electronics<br />
Dissertation Committee<br />
Prof. Dr. Werner Bergholz • <strong>Jacobs</strong> <strong>University</strong><br />
Dr. Mathias Bode • <strong>Jacobs</strong> <strong>University</strong><br />
Prof. Dr. Roland Jochem • TU Berlin<br />
Roberto Lotz • Volkswagen AG<br />
Roman Mogalle-Kiebler • Volkswagen AG<br />
Chair of the Dissertation Committee: Prof. Dr. Werner Bergholz<br />
Date of Defense: September 21, 2012<br />
School of Engineering and Science<br />
<strong>Jacobs</strong> <strong>University</strong> Bremen • Campus Ring 1 • 28759 Bremen • Germany<br />
Group Quality Assurance Supply Parts Chemistry<br />
International Coordination Supplier Audit<br />
Volkswagen AG • P.O. Box 011/14670 • 38436 Wolfsburg • Germany
Disclaimer<br />
April 23, 2013<br />
To whom it may concern,<br />
I, Borislav Hadzhiev, hereby declare that this Ph.D. Thesis is an independent work that has not<br />
been submitted elsewhere for conferral of a degree.<br />
Publications about the content of this work require the written consent of Volkswagen AG.<br />
The results, opinions and conclusions expressed in this thesis are not necessarily those of<br />
Volkswagen AG.<br />
Borislav Hadzhiev<br />
i
Abstract<br />
Today automotive producers operate on a global market with very strong competition and vast<br />
variety of customers, each with their own preferences. Under these conditions automotive companies<br />
need to provide products of excellent quality in order to stay in business. However, managing<br />
quality in the automotive production is a particularly demanding task due to the fact that automotive<br />
Original Equipment Manufacturers (OEM) have some of the most intricate production networks<br />
which exist. They are involved in only about 20% to 40% of the actual production process and the<br />
trend is that in the future this share will keep on shrinking. Therefore OEMs should make sure that<br />
they cooperate only with reliable and quality capable business partners.<br />
Supplier auditing is a very important quality management tool, which is used to assess the capability<br />
of a particular supplier to deliver products of high quality and therefore its suitability as a<br />
business partner. Quality auditing is employed before awarding contracts and during the series production<br />
to assess the quality risks down the supply chain. This work studies the ability of supplier<br />
quality auditing in the automotive industry to provide reliable process capability evaluations. The<br />
case study presented here was carried out in cooperation with Volkswagen AG and evaluates the<br />
effectiveness of the supplier auditing process at the German automotive manufacturer with respect<br />
to the challenges on the global automotive market.<br />
This paper employs a research approach for assessment of the quality auditing process, which<br />
draws similarities to sampling – an important measurement method in the field of Electrical Engineering.<br />
The discussion points out important aspects for the technical implementation of the<br />
quality audit in the automotive industry as well as critical points regarding the information management<br />
of audit evaluation records. Even though the focus of this work is on the automotive<br />
industry, the analytical approach and the statistical methods used here can be used to assess the<br />
effectiveness of supplier quality auditing also in other industry sectors.<br />
ii
Contents<br />
1 Introduction 1<br />
2 Research Question and Research Motivation 3<br />
3 Research Approach 8<br />
3.1 General Strategy for Employing the Quality Audit . . . . . . . . . . . . . . . . . . 13<br />
3.2 Effectiveness of Implementation of a Single Quality Audit . . . . . . . . . . . . . 14<br />
4 Specifics of the Automotive Industry 17<br />
4.1 Internationalization of the Automotive Operations . . . . . . . . . . . . . . . . . . 17<br />
4.2 Rising Development Costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21<br />
4.3 The Automotive Supply Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24<br />
4.4 Implications for the General Automotive Quality Auditing Strategy . . . . . . . . . 28<br />
5 Fundamental Principles of the Quality Audit 32<br />
5.1 Audit Planning, Initiation and Preparation . . . . . . . . . . . . . . . . . . . . . . 34<br />
5.2 On-site Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37<br />
5.3 Reporting and Closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39<br />
6 Empirical Data 39<br />
6.1 Quality Capability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40<br />
6.2 Quality Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48<br />
6.3 Automation of the Data Processing . . . . . . . . . . . . . . . . . . . . . . . . . . 54<br />
6.3.1 Matching Quality Capability and Quality Performance . . . . . . . . . . . 56<br />
6.3.2 Supplier Identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57<br />
7 Results and Discussion 58<br />
7.1 State and Quality of the Available Empirical Data . . . . . . . . . . . . . . . . . . 58<br />
7.2 Possible Sources of Data Bias . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70<br />
7.2.1 Revisions of Volkswagen’s Quality Auditing Process . . . . . . . . . . . . 70<br />
7.2.2 Period of Relevance of Quality Audit Results . . . . . . . . . . . . . . . . 84<br />
7.2.3 The Factor ”Human” and People Calibration . . . . . . . . . . . . . . . . 86<br />
7.2.3.1 Calibration of Supplier Quality Auditors . . . . . . . . . . . . . 86<br />
7.2.3.2 Calibration of Production Quality Assessment . . . . . . . . . . 88<br />
7.3 Matching Quality Capability and Quality Performance . . . . . . . . . . . . . . . 95<br />
7.4 Relation Between Number of Defective Parts and Delivery Amount . . . . . . . . 102<br />
8 Conclusion 121<br />
8.1 General Auditing Strategy (Sampling Frequency) . . . . . . . . . . . . . . . . . . 123<br />
8.2 Effectiveness of Individual Audits (Accuracy of Samples) . . . . . . . . . . . . . . 124<br />
8.3 Answers of the Research Questions . . . . . . . . . . . . . . . . . . . . . . . . . 129<br />
Appendices 133<br />
A Results from the analysis presented in Section 7.2.1 133<br />
B Results from the analysis presented in Section 7.2.3.2 155<br />
B.1 List of Material Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161<br />
References 162<br />
iii
Acknowledgements<br />
I hereby want to express my sincere gratitude to the numerous people, who were involved in<br />
this project and whose constructive input made it possible to achieve the results presented in this<br />
paper. Even though the list is very long and I am very thankful to every single person on it, there<br />
are a few names, to which I want to bring attention.<br />
I would like to express my great appreciation to my supervisor Prof. Dr. Werner Bergholz for<br />
his high commitment and incessant support throughout every single stage of the project, to Roberto<br />
Lotz, who with his more than 30 years of experience in the automotive sector brought valuable<br />
practical background in the discussions, to Dr. Mathias Bode for his constructive questions and<br />
insightful remarks, to Prof. Dr. Roland Jochem for his sincere and objective input, to Roman<br />
Mogalle-Kiebler for his committed and cooperative conduct. I want to also thank Roland Assmann<br />
for his interest and help in initiating the project.<br />
Furthermore, I am thankful to Carlo Boettger, Wolfgang Hering, Tanja Brueggemann, Osman<br />
Meneses and every further colleague at Volkswagen, who contributed to the productive and comfortable<br />
working atmosphere at any time.<br />
And last but not least I want to express my gratefulness also to Silviya Nikolova for her patience<br />
and strong moral support throughout the entire project.<br />
Sincerely,<br />
Borislav Hadzhiev<br />
iv
A customer is the most important visitor on our premises. He is not dependent<br />
on us. We are dependent on him. He is not an interruption on our<br />
work. He is the purpose of it. He is not an outsider on our business. He is<br />
a part of it. We are not doing him a favour by serving him. He is doing us<br />
a favour by giving us an opportunity to do so.<br />
Mahatma Gandhi<br />
1 Introduction<br />
The Cambridge Dictionaries Online (2012) define business as ”the activity of buying and selling<br />
goods and services, or a particular company that does this, or work that you do to earn money”<br />
1 . The target of each commercial organization is to maximize its long-term profits and to assure<br />
sustainability and continuity of its operations by building a set of customers which are going to<br />
repeat business with the organization in the future. Under free market conditions organizations can<br />
only achieve this, if their products meet the expectations of their target customers.<br />
The term ”quality” is defined by the international standard ISO 9000 (2005) as the ”degree to<br />
which a set of inherent characteristics fulfils requirements” (p. 18). Quality plays a central role<br />
in today’s business environment. If an organization manages to identify and fulfill its customers’<br />
requirements through its products to a high degree (achieve good quality), its products will deliver<br />
high value to its customers who will be willing to do business with the company in the future.<br />
Thus, the good reputation of the organization and the perceived value of its products will grow.<br />
In his work The One Number You Need to Grow, Reichheld (2003) argues that company growth<br />
correlates very well with its customers’ readiness to recommend company’s products to other potential<br />
customers, such as e.g. their friends or colleagues. When satisfied customers share their<br />
good experiences with other potential customers, the customer base of the organization expands<br />
generating business growth and larger long-term profits. If, on the other hand, an organization fails<br />
to meet the expectations of its customers (i.e. provides poor quality), it is very likely that the latter<br />
will not be willing to do business with the organization in the future. The pool of customers of the<br />
1 In this paper the term product is used to denote both a physical object and/or a service.<br />
1
organization will therefore shrink and the organization might not be able to sustain its business in<br />
the long run.<br />
The importance of quality is further intensified in a competitive environment. Most of the time<br />
a number of companies (business competitors) are competing for a limited number of customers.<br />
The company that offers the best quality will be able to capture the largest share of the customer<br />
pool and therefore be the most profitable. For that reason quality has become a cornerstone of the<br />
business strategies in every major industry today. It should be noted, however, that in a competitive<br />
business environment price is also an important factor along with quality.<br />
To remain profitable in the longer term companies need to sustain and continuously improve<br />
quality. This vision is reflected in international standards such as the ISO 9001 (2008, p. 9). A key<br />
concept in business management is the EFQM Excellence Model (2012) of the European Foundation<br />
for Quality Management (EFQM), which defines a framework for assessment of company<br />
progress, ideas communication, and achieving company’s business targets. However, modern business<br />
world is defined by complexity and to achieve quality of your products is not always straight<br />
forward. Definitely the ability to identify trends in customer demand and quickly and adequately<br />
respond to these changes plays a crucial role. That is the reason why companies spend millions on<br />
market research. Even though very important, identifying customer needs alone is not sufficient.<br />
It is equally challenging to translate the identified customer expectations into specific companyinternal<br />
objectives. For example, a chocolatier might know that her customers prefer the chocolate<br />
mousse neither too sweet nor too bitter. But in order to make high quality chocolate mousse the<br />
task of the chocolatier would be to first understand what ”neither too sweet nor too bitter” means<br />
to her customers and then to translate it into a specific proportion of sugar and cocoa in her recipe.<br />
The task becomes even more difficult if we move to more complex products such as a television<br />
set or a smart phone. But probably the most complex every-day product most of us would ever<br />
use is a car. An automobile has to meet a very broad set of requirements ranging from extremely<br />
diverse design and performance preferences, through tax, safety and environmental regulations, to<br />
better price for value expectations.<br />
Companies, which offer complex products, rarely produce these entirely by themselves. Very<br />
often one or rather several external companies are involved in the production process. In the race<br />
for market share business competitors are put under pressure to offer superior products at lower<br />
prices. Outsourcing part of their processes to external partners allows companies to reduce costs<br />
and concentrate valuable resources on their most important capabilities, thus maximizing the profit<br />
2
margins and strengthening their market position (Abele, Meyer, Näher, Strube, & Sykes, 2008). In<br />
a number of market sectors outsourcing has become so important that doing business without it is<br />
unthinkable.<br />
Along with its benefits, however, outsourcing turns quality assurance into a significant challenge.<br />
Increasing the number of parties involved in the production process raises the complexity<br />
of the communication flow, the number of process interfaces grows (practice shows that most of<br />
the quality related problems arise at such interfaces) and thus companies run higher risk of miscommunication<br />
of their business objectives. Nonetheless, it is the responsibility of the outsourcing<br />
company to make sure that their products meet customers’ expectations, which in turn means that<br />
quality requirements must be fulfilled in the entire supply chain. van Weele (2010) writes: ”quality<br />
of the finished product is determined to a large extent by the quality of raw materials and<br />
components” (p. 241). Therefore, it is very important for a company to carefully select its business<br />
partners. For the latter understanding the generally accepted quality practices and being able to<br />
properly manage quality are a basic requirement.<br />
This paper studies quality assurance down the supply chain of the automotive industry – a<br />
business sector with one of the most intricate mass production supply chains which exist 2 . More<br />
precisely, the aim of the subsequent discussion is to present and evaluate the effectiveness of supplier<br />
quality auditing. Supplier quality auditing is a widely used quality assurance tool, which<br />
evaluates the quality capability of production processes down the supply chain and ultimately provides<br />
essential input for decision making in the outsourcing process.<br />
2 Research Question and Research Motivation<br />
Due to the highly competitive environment on the international automotive market, the growing<br />
diversity of customer preferences, and the fact that on average 60 to 80 percent of the global automotive<br />
production is outsourced (Veloso & Kumar, 2002), competing in the automotive business<br />
is particularly demanding. Good management of the entire production chain and more concretely<br />
the use of effective quality management tools is therefore of key importance for staying in the<br />
sector. This work examines the effectiveness of second-party quality auditing in the automotive<br />
2 The supply chains of the aircraft and ship industries are of comparable complexity as that of the automotive sector.<br />
Their production scale is substantially smaller, however, which is the reason why QM tools in the two industries have<br />
not been developed and implemented to the same extent as in most other mass production sectors, what represents an<br />
additional challenge for the quality management process in those two sectors.<br />
3
industry and especially its strategic importance as an integral part of the supplier management and<br />
supply chain management processes. The purpose of this work is formulated within the following<br />
research question: Can quality performance of automotive suppliers in the mass production be anticipated<br />
based on quality evaluation of the capability of their production processes? Additionally,<br />
the analysis presented here tries to give an answer to a second, more subjective question: What<br />
is the quality of the quality auditing process at Volkswagen Group and what is its improvement<br />
potential?<br />
Naturally the quality capability of a production process is reflected in the extent to which the<br />
delivered components comply with the relevant quality specifications. It is therefore possible to<br />
infer the quality capability of a particular production process based on the quality of supplier’s<br />
products. If supplier’s products are of high quality, which is stable over time, one can conclude<br />
with relatively high level of certainty that its process has high quality capability (Figure 1). In many<br />
cases the quality of a particular part is not only expressed in its physical characteristics but also in<br />
the quality of services which the respective supplier offers. Thus for example, measures such as<br />
the on-time delivery, having sufficient capacity to keep up with the customer demand, as well as<br />
good responsiveness in cases of potential problems are quite important for the customer-supplier<br />
relationship.<br />
Supplier<br />
Production<br />
Process<br />
Product<br />
Customer<br />
(OEM)<br />
Process Quality<br />
Capability<br />
reactive<br />
Product<br />
Quality<br />
Second-Party<br />
Quality Audit<br />
proactive<br />
Figure 1: Inferring the quality capability of a supplier and ultimately its suitability as a business<br />
partner based on the quality of its final product is relatively straight forward, but rather reactive<br />
(only possible after the contracts are in place). Therefore companies try to proactively infer the<br />
quality of the final products based on evaluations (second-party quality audits) of the quality capability<br />
of the relevant production processes, which is a significantly more demanding task. The<br />
question is how successful is quality auditing in achieving this.<br />
4
Certainly companies want to work only with suppliers which are quality capable. Thus for example<br />
the Volkswagen quality criteria for contract awards described in Formel Q-capability (2008)<br />
include the following requirement: ”In order to ensure the quality of the components/modules/systems<br />
the companies of the VOLKSWAGEN GROUP [require] the supplier to [produce] the components/modules/systems<br />
in an A-rated [(quality capable)] production site” (p. 9). Even though<br />
quality of final products and services is a straightforward way to measure quality capability of production<br />
processes, making conclusions about the quality capability of a particular supplier based<br />
on its quality performance, however, is rather reactive. Quality performance can be measured only<br />
after the start of mass production. Something, what companies definitely want to avoid, is to receive<br />
parts with low quality during the series production of a particular product. Such situations<br />
would unbalance the overall business process and are therefore quite undesirable. For that reason,<br />
it is important for companies to be able to assess the quality capability of their suppliers before<br />
signing a contract. To do that, however, they need an alternative method, which allows them to<br />
determine the quality capability of a particular supplier in a proactive manner.<br />
To this end one particularly important aspect of business relationships is standardization. Given<br />
the complexity and size of the automotive supplier network, coping with the challenges posed by<br />
modern-day production requires establishing and maintaining well-structured systems for quality<br />
management (QMS) and quality engineering. The latter were originally developed by companies<br />
in an individual and informal manner, e.g. Ford’s Q101, GM’s Supplier Performance and Evaluation<br />
Report (SPEAR) (Hoyle, 2005, p. 100). It was not before the appearance of the first quality<br />
standards that the development and maintenance of quality management systems was formalized<br />
and the individual implementation steps were well-documented.<br />
Nowadays a number of international quality standards provide guidelines for the effective implementation<br />
of a quality management system (e.g. ISO 9000, 2005; ISO/TS 16949, 2009), which<br />
assures the quality of an organization’s products. Certification according to quality standards such<br />
as the ISO 9001 or ISO/TS 16949 (which builds upon ISO 9001 and is of particular importance<br />
for the automotive sector) serves to build trust in the business relationships. Companies use such<br />
certification to show that they are quality aware, to provide proof of the effective implementation<br />
of a quality management system, and to assure their clients that they work on the continuous<br />
improvement of their products. This makes certified companies preferred business partners. Moreover,<br />
even though the use of standards is of recommending character, nowadays to enter business<br />
without a quality certification is almost impossible in a vast number of industries.<br />
5
Quality standards build upon the Total Quality Management (TQM) principle, which postulates<br />
that business should be sustained through continuous improvement of its products and processes.<br />
The ISO 9000 (2005) standard defines eight quality management principles, which are fundamental<br />
for the successful operation of an organization:<br />
a) Customer focus<br />
Organizations depend on their customers and therefore should<br />
understand current and future customer needs, should meet customer<br />
requirements and strive to exceed customer expectations.<br />
b) Leadership<br />
Leaders establish unity of purpose and direction of the<br />
organization.<br />
They should create and maintain the internal<br />
environment in which people can become fully involved in achieving<br />
the organization’s objectives.<br />
c) Involvement of people<br />
People at all levels are the essence of an organization and<br />
their full involvement enables their abilities to be used for<br />
the organization’s benefit.<br />
d) Process approach<br />
A desired result is achieved more efficiently when activities and<br />
related resources are managed as a process.<br />
e) System approach to management<br />
Identifying, understanding and managing interrelated processes<br />
as a system contributes to the organization’s effectiveness and<br />
efficiency in achieving its objectives.<br />
f) Continual improvement<br />
Continual improvement of the organization’s overall performance<br />
6
should be a permanent objective of the organization.<br />
g) Factual approach to decision making<br />
Effective decisions are based on the analysis of data and<br />
information.<br />
h) Mutually beneficial supplier relationships<br />
An organization and its suppliers are interdependent and a<br />
mutually beneficial relationship enhances the ability of both<br />
to create value. (pp. 5–6)<br />
Furthermore, Hendricks and Singhal (2000) provide statistical evidence that companies, which<br />
have embraced the TQM philosophy and have successfully incorporated the TQM principles in<br />
their organization show convincing long term performance. Their financial stability and good market<br />
standing makes such companies reliable and preferred business partners. For that reason in the<br />
automotive sector certification to the international quality standards such as the ISO 9000 family<br />
and especially ISO/TS 16949 has become a fundamental requirement for entering business. Certification<br />
attests the conformity of the implemented quality management system to the respective<br />
standards.<br />
Gropp (2009) argues, however, that possessing a valid certificate does not necessarily imply<br />
high product quality. He points out that for a number of companies the major incentives for certification<br />
are the desire to maintain a good corporate image as well as the practical compulsion<br />
for certification itself imposed by the outsourcing companies. Gropp (2009) writes further that in<br />
many cases certified organizations do not follow the TQM philosophy and show high deficits in<br />
the implementation of the standard requirements especially on the process level, and sees this as<br />
the main reason why automotive companies resort to additional quality control of their certified<br />
business partners especially on the process and product levels.<br />
A standard business practice to proactively assess the quality capability of your suppliers’ production<br />
processes is through quality auditing (Figure 1). Such type of evaluation is based on the<br />
assumption that process parameters are especially important and have a direct influence on the<br />
quality of the process outputs (VDA 6.3, 2010; ISO 19011, 2011; The EFQM Excellence Model,<br />
2012). The quality audit focuses therefore on critical processes and compares their conformity to<br />
7
generally accepted benchmarks. Quality auditing is particularly useful for supplier evaluation, because<br />
it additionally offers the opportunity to estimate the levels of risk associated with the quality<br />
capability of the individual processing steps. The degree to which the according processes conform<br />
with the state-of-the-art practices is then used to make a statement about their overall quality capability<br />
and its sustainability in the long term. Having the focus of the quality audit on the correct<br />
process parameters is therefore essential for its effectiveness.<br />
There is limited scientific literature (Stroescu-Dabu, 2008; Hadzhiev, 2009), which addresses<br />
the question whether second-party quality auditing in the automotive industry is indeed effectively<br />
implemented in practice and whether it succeeds in providing a reliable foresight of the quality of<br />
purchased components. On the other hand, answering such a question is very important, because<br />
supplier process evaluations have a direct impact on a company’s sourcing decisions and ultimately<br />
affect its market performance. The current paper was motivated by these circumstances and aims<br />
at providing more insights on the topic. The following section presents in detail the research<br />
approach. Section 4 provides a general description of the automotive business sector. Special<br />
attention is paid to the trends of development of the sector and its major driving factors. The<br />
specifics of automotive production are then put in quality perspective in order to outline the quality<br />
risks with special focus on problems originating in the supply chain and their implications for<br />
OEMs’ general quality auditing strategies. The rest of the paper focuses on the effective practical<br />
implementation of quality auditing and analyses empirical data resulting from the supplier auditing<br />
process at one of the largest automotive producers in the world – Volkswagen Group.<br />
3 Research Approach<br />
To address the research questions described above this study takes an empirical approach and<br />
analyzes real-world operational business data. The analysis presented here was carried out in cooperation<br />
with Volkswagen Group and in particular its corporate department for Quality Assurance<br />
of Purchased Parts (from German Konzern Qualitätssicherung Kaufteile), which among others is<br />
responsible for the international coordination of Volkswagen’s global quality auditing units. Given<br />
the scale of the organization, the provided empirical data is particularly useful to assess challenges<br />
faced by automotive OEMs on the major automotive markets worldwide. Volkswagen Group is<br />
the largest European automotive manufacturer and currently among the three largest in the world,<br />
alongside the Japanese Toyota and the American General Motors. Volkswagen Group operates<br />
8
more than 60 production facilities on five continents and its global operations are supported by<br />
more than 9000 direct business partners (1st-tier suppliers) 3 . Because of the large number of<br />
suppliers, second-party auditing has received a central role in Volkswagen’s quality philosophy.<br />
Furthermore, the scale of its global operations and the size of its supplier base provide a broad set<br />
of empirical data, and therefore the opportunity for an in-depth analysis of the topic at hand. For<br />
these reasons, Volkswagen Group is an ideal research object.<br />
The essence of quality auditing at Volkswagen Group lies within regular evaluations of the<br />
quality capability of suppliers’ production processes. As will be explained in detail later in this<br />
paper a particular supplier quality audit lasts just several days and provides simply an assessment<br />
of the momentary quality capability state of the evaluated production processes. On the other<br />
hand, the long-term development of the quality capability of a particular production process can<br />
be obtained only by a succession of quality audits at different points of time. Interestingly this<br />
approach draws a lot of similarities with an important technique in the signal processing called<br />
sampling. Sampling is used to convert analog to digital signals and is fundamental for areas such as<br />
digital communications (Karrenberg, 2002; D’Antona & Ferrero, 2006). The idea behind sampling<br />
is rather simple: ”the input signals are converted into a sequence of sampled values by means of a<br />
sampling operation performed at given time instants, with a constant sampling period”, (D’Antona<br />
& Ferrero, 2006, p. 33) . In practical terms this means that an analog (continuous in time) signal<br />
(Figure 2a), such as speech for example, passes through an analog to digital (A/D) converter, which<br />
converts it into a digitalized (discrete in time) copy of the original signal (Figure 2b). Proper<br />
sampling reduces the amount of information contained within the original signal, but preserves the<br />
main characteristics of the signal or system under study.<br />
Most of the time analog signals contain a significant amount of redundant information, which<br />
could be omitted and the core message be still transmitted. This less relevant information introduces<br />
a computing overhead during the data processing cycle. Through sampling the amount of<br />
information in the original signal is reduced and therefore its processing is simplified. A major<br />
consideration in sampling, however, is to use enough samples in order to be able to completely<br />
retrieve the essential information in the original signal from the sampled sequence (Karrenberg,<br />
2002; D’Antona & Ferrero, 2006). Depending on the rate of variation of the individual continuoustime<br />
signals, the frequency of the sampling measurements should be adjusted accordingly. If the<br />
3 Source: personal communication with Volkswagen AG employees.<br />
9
sampling frequency is not high enough, this will lead to the improper representation of the original<br />
signal and therefore loss of information (Figure 3b (II.)). In the theory of signal processing this<br />
concept is formalized by the sampling theorem (Karrenberg, 2002; D’Antona & Ferrero, 2006),<br />
which states that an analog signal can only be fully reconstructed from its time-discrete representation,<br />
if the sampling frequency f S is at least two times larger than the highest frequency f MAX<br />
in the original signal. Here the mathematics behind this statement are omitted. For a formal mathematical<br />
derivation of the theorem and the necessary preconditions for its validity the reader is<br />
referred to e.g. D’Antona and Ferrero (2006, pp.35–40). What is important to understand here,<br />
however, is that in order to get an idea about the true characteristics of a particular continuous<br />
signal, the latter needs to be sampled with frequency sufficiently larger than the rate of its inherent<br />
variations. Thus, signals with particularly high amount of variations need to be measured<br />
more frequently than signals with lower amount of variations in order to preserve their essential<br />
characteristics.<br />
At this point one could ask what exactly the term signal means. Sinha (2010) defines a signal as<br />
”any time-varying physical quantity” (p. 9). In this line of thoughts, when talking about sampling<br />
signals, instead of referring to quantities commonly used in the field of Electrical Engineering<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ● ●<br />
●<br />
●<br />
● ●●<br />
●●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ● ●<br />
(a) Continuous time signal<br />
(b) Sampled signal<br />
Figure 2: Analog to digital (AD) conversion of an analog (continuous in time) sinc signal – (a) –<br />
to a digital (discrete in time) signal – (b) – using sampling with a constant frequency.<br />
10
I.<br />
f MAX = 1Hz<br />
I.<br />
f MAX = 6Hz<br />
II.<br />
●<br />
●<br />
●<br />
f S = 12Hz<br />
II.<br />
●<br />
●<br />
●<br />
f S = 12Hz<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
III.<br />
● ● ● ● ● ● ● ● ● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
f S = 48Hz<br />
III.<br />
● ● ●<br />
● ● ●<br />
● ● ●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
● ● ●<br />
●<br />
f S = 48Hz<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ● ● ● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ● ● ● ● ● ● ● ●<br />
● ●<br />
● ● ●<br />
● ●<br />
●<br />
● ● ●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
● ●<br />
(a) Signal y(t) = sin (2πf 1 t) sampled at 12Hz and<br />
48Hz respectively.<br />
(b) Signal y(t) = sin (2πf 1 t) + 0.25 sin (2πf 2 t)<br />
sampled at 12Hz and 48Hz respectively.<br />
Figure 3: Sampling of two analog signals. In each of the two figures the uppermost plot presents<br />
the continuous time signal (I.). Each signal was sampled using f S = 12Hz (II.) and f S = 48Hz<br />
(III.). The signal on the left has one frequency component f 1 = f MAX = 1Hz, while the signal on<br />
the right has two frequency components of f 1 = 1Hz and f 2 = f MAX = 6Hz respectively.<br />
such as voltage or current, one may as well refer to any other quantity varying over time such<br />
as the amount of money in a particular bank account measured in EUR or the level to which a<br />
particular supplier production process fulfills the quality requirements of the outsourcing company<br />
(process quality capability) measured in percent. In the discussion presented in this paper, we are<br />
of course interested in the latter. Naturally such signals would exhibit variations on significantly<br />
different time scale than electrical signals. Nevertheless, it is still possible to measure them on a<br />
periodic basis to gain an idea about their overall behavior.<br />
Usually a company would want to assess the quality capability of its suppliers and its variation<br />
over time. The ability to identify any negative trends in the development of the quality capability<br />
of their suppliers allows companies to detect and eliminate potential problems in a timely manner.<br />
In this regard the quality audits could be deemed as the individual sampling measurements and<br />
the sequence of audits at a particular supplier as the overall sampling signal. The average length<br />
of the time intervals between individual quality audits would be therefore the ”sampling period”<br />
T S = 1/f S .<br />
11
Two fundamental factors influence the accuracy of such assessments. As already mentioned<br />
above, one is the frequency with which the individual audits are performed. Too frequent auditing<br />
of suppliers with rather stable production processes is not a problem from the quality point<br />
of view, but on the other hand would make the auditing process rather inefficient in terms of resources.<br />
However, if suppliers experience high variation of their internal processes, it can be rather<br />
misleading to expect that their overall quality capability will remain constant over long time periods.<br />
Moreover, quite probably it will be subject to persistent changes. Therefore if audits at such<br />
suppliers are not performed frequently enough, conclusions about supplier’s quality capability may<br />
rather quickly become out-of-date and not correspond to the current state of the respective production<br />
processes. For these reasons, to assure the effectiveness of the quality auditing as a quality<br />
management tool, companies need to consider the specifics of the individual links down the supply<br />
chain and accordingly adjust the auditing (”sampling”) frequency.<br />
The second important factor, which is usually omitted in the classic discussions on sampling,<br />
is the assumption that throughout the sampling process the individual measurements are accurate<br />
enough to detect the correct level of the input analog signal. In the field of Electrical Engineering<br />
such a consideration is probably rather insignificant, due to the fact that measurement of quantities<br />
such as voltage or current is very well defined and can be achieved with very high precision. When<br />
we talk about measuring quality capability, however, it is not always obvious that this condition<br />
will be fulfilled. There are a number of factors which influence the correct assessment of supplier’s<br />
quality capability during a single audit. These need to be comprehensively examined in order to<br />
make sure that the individual quality evaluations will adequately evaluate the level to which a<br />
particular production process fulfills customer’s quality requirements.<br />
Consequently, in order to guarantee the effectiveness of quality auditing as a quality management<br />
tool, companies need to address the problem from two different perspectives. On the one<br />
hand, it is important individual quality audits to be able to adequately assess the capability of a<br />
particular production process to provide the necessary quality, while on the other hand the general<br />
strategy for applying and the frequency of use of this quality management tool needs to correspond<br />
to the developments and the specifics of the according industry sector. To answer the research questions<br />
this paper pursues a similar approach and deals with the topics from two different viewpoints.<br />
12
3.1 General Strategy for Employing the Quality Audit<br />
It is quite difficult to find scientific studies, which deal with the effectiveness of the quality audit<br />
from a strategic viewpoint, especially in the automotive industry. The lack of literature is probably<br />
partially due to the fact that if there were such studies at all, these were performed internally and<br />
their results are kept unpublished because of their perceived value for the respective OEMs and<br />
the potential competitive advantage they offer. On the other hand, most of the available literature<br />
which deals with the topic of quality auditing concentrates primarily on the specifics of the implementation<br />
and individual steps of a single audit (Arter, 1989; Parsowith, 1995; Green, 1997;<br />
Wealleans, 2005) and very few works discuss also the strategic importance of effective quality auditing<br />
in the overall market strategy of the corporation (Wealleans, 2005). Even when there is such<br />
a discussion it is somewhat superficial. Thus for instance, even though works in the field of supply<br />
chain management recognize the strategic importance of supplier evaluation (van Weele, 2010),<br />
their discussion usually concentrates on the overall supply management process and the topic of<br />
quality auditing is not covered in sufficient depth.<br />
A strategy for the effective deployment of quality auditing as a major quality management tool<br />
in the automotive industry has to address several important aspects of the automotive market and<br />
to adequately respond to the challenges, which OEMs face in the different regions. First, it is very<br />
important to start with the basics and consider the implications which assembly line manufacturing<br />
alone has on the overall business process. Here it is particularly important to account for major<br />
sources of uncertainty such as changing customer demand, inherent variations of the production<br />
process as well as variations originating in the supply chain such as supply interruptions due to<br />
delivery delays or inconsistent quality of the delivered components. A second major point in this<br />
discussion are the dynamics of the global market and in particular the specifics of the developed<br />
western markets, i.e. the Triad – USA, Japan and Western Europe, versus the emerging markets,<br />
including among others the BRIC countries (Brazil, Russia, India, and China), South East Asia,<br />
and Eastern Europe. An important topic in this regard is the increasing need for diverse product<br />
portfolios, which need to be able to answer the growing diversity of customer preferences. A third<br />
important factor, which needs to be considered, are rising development costs and the trend towards<br />
outsourcing major portion of the production process. It is especially pressing to account for the<br />
growing complexity not only of the supplier network but also of the sourced components, as well<br />
as to consider topics such as sustainability of the production process quality capability.<br />
All these factors have a strong influence on the market performance of automotive OEMs since<br />
13
they put the quality of the final products to the challenge and are therefore relevant aspects to<br />
be taken into account in a study about the effectiveness of quality audits. The individual points<br />
are discussed in detail in Sections 4 and 4.4, as the discussion aims to provide the reader with<br />
background for commonly accepted business strategies in the automotive sector. Of particular<br />
interest are the therewith associated quality risks and the consequences these have on the overall<br />
quality auditing strategy. As a particular example the quality auditing policy of Volkswagen Group<br />
is presented in detail with respect to the presented quality challenges in the automotive sector.<br />
3.2 Effectiveness of Implementation of a Single Quality Audit<br />
The second part of this work deals with the ability of a single quality audit to accurately infer the<br />
actual quality capability of the respective supplier at the time of the audit. If one assumes that<br />
the quality audit is effectively implemented and can adequately infer the quality capability of a<br />
supplier, one should expect that audit evaluation results would correspond to the actual quality<br />
performance of the respective suppliers at the time of the audit. In the end an audit evaluation<br />
score and the quality performance of a particular supplier are just two different ways to evaluate<br />
the quality capability of the same supplier production process.<br />
This assumption is confirmed in an empirical study by Wittmann and Bergholz (2006), who<br />
analyzed the benefits of quality auditing at Infineon Technologies AG (a producer of electronics).<br />
They found that there is ”at least qualitative” correlation between the quality performance<br />
of individual Infineon suppliers and their audit scores (Wittmann & Bergholz, 2006). Suppliers<br />
with higher quality evaluation scores would cause lower number of quality related problems in the<br />
overall production process. Furthermore, the results of the analysis by Wittmann and Bergholz<br />
(2006) also show that quality auditing has a positive effect on the long term quality capability<br />
of suppliers’ production processes. The authors observed the tendency that suppliers’ evaluation<br />
scores continuously improve during the subsequent audits (second and third audit). On the other<br />
hand, the results of their study show that the quality capability of suppliers of wafer substrates was<br />
rated significantly higher than that of suppliers of process materials such as gases and chemicals<br />
(Wittmann & Bergholz, 2006). The authors do not provide information about the relation between<br />
the evaluation scores of the latter and their actual quality performance. Nevertheless, such findings<br />
suggest important differences within the individual parts of the supplier network, which have to<br />
be studied carefully as they would eventually identify potential adjustments of the overall quality<br />
auditing procedure specific to the respective type of industry.<br />
14
The work of Wittmann and Bergholz (2006) provides an important method for assessment of<br />
the effectiveness of quality auditing. An evaluation of the consistency between quality performance<br />
of Volkswagen suppliers and their respective quality evaluation scores is therefore a good starting<br />
point to assess the effectiveness of the implementation of quality auditing at Volkswagen Group in<br />
particular and in the automotive industry in general. Consequently, a substantial part of this paper<br />
is devoted to a set of studies, which investigate this relationship.<br />
Performance information of Volkswagen Group suppliers was acquired from production plant<br />
records and is compared to the perceived quality capability of the respective supplier production<br />
processes provided by their quality auditing scores measured in percent. At this point the initial<br />
hypothesis of this part of the analysis can be summarized in that suppliers which obtained higher<br />
scores during the audit evaluation are expected to have fewer problems throughout the series delivery<br />
and therefore better quality performance, and vice versa. Any discrepancies between the<br />
evaluation scores of suppliers and their actual quality performance imply potential deficiencies of<br />
the quality auditing routine and therefore identify particular areas which need closer examination.<br />
On the other hand, best practice audit processes can be derived from cases which show good parity<br />
between the two investigated quantities.<br />
Two recent studies (Stroescu-Dabu, 2008; Hadzhiev, 2009) evaluate the effectiveness of<br />
Volkswagen Group’s quality audit and employ the same analytical approach. These works already<br />
outline important findings, which serve as a basis for the analysis presented here. The first<br />
important point is the fact that the automotive supply network is characterized by a high level of<br />
diversity. This is not surprising, provided its enormous scale and the great variety of individual<br />
components, which are assembled in a car. Stroescu-Dabu (2008) uses a general supplier categorization<br />
based on the type of industry in which each company operates. Suppliers are divided into<br />
metal, chemical, and electrical suppliers. The analysis of Stroescu-Dabu (2008) shows that the<br />
distribution of evaluation scores of electrical suppliers is statistically different than the evaluation<br />
distributions of metal and chemical suppliers. The latter, on the other hand, have similar evaluation<br />
score distributions. Furthermore, while in the electric industry auditing scores show relatively<br />
good correspondence to the quality performance of the respective suppliers, for suppliers which<br />
operate in the metal and chemical industries results do not provide any evidence that suppliers with<br />
better audit scores perform better in terms of quality of the delivered components. This is a highly<br />
undesired state of affairs and needs to be clarified.<br />
In an attempt to narrow down the causes of these results Hadzhiev (2009) extends the findings<br />
15
of Stroescu-Dabu (2008) and uses a more detailed data categorization, which takes into account<br />
not only industry of operation, but also the type of delivered components as well as the processes<br />
involved in their production. The general trends of the overall data, observed by Stroescu-Dabu<br />
(2008), are confirmed also by the analysis of Hadzhiev (2009). Nevertheless, the results of the<br />
second work reveal additional subgroup differences within the individual industry sectors. In<br />
particular, certain supplier subgroups in both electric and metal industries, such as suppliers of<br />
lighting components or metal profiles, show significantly better consistency between their quality<br />
evaluations and actual quality performance than the rest of the suppliers in the respective industry.<br />
Meanwhile, industry sub-groups, such as suppliers of electro-mechanic controls or cast metal<br />
components, show a negative relation between their quality capability evaluation scores and their<br />
quality performance. For such suppliers a higher quality evaluation score corresponds to a larger<br />
number of quality related problems in the series production.<br />
The studies by Stroescu-Dabu (2008) and Hadzhiev (2009) suggest potential shortcomings of<br />
the quality auditing practice at Volkswagen Group. Even though some of their results are on the<br />
verge of statistical significance, the findings of the two studies emphasize the need for further research<br />
on the topic in order to understand the underlying reasons for these observations. There are a<br />
number of factors, which are not accounted for in the two papers presented above, but which could<br />
potentially bias the data and therefore strongly influence the analyses. One particularly important<br />
point is consistency of the evaluated quality performance data. The supplier quality performance<br />
indicator used in both works is ppm (defective parts per million). At the first stage of this project<br />
it was found, however, that there are significant differences in the way ppm are collected for simple<br />
components as opposed to more sophisticated assembly units (for a detailed discussion of this<br />
point see the following sections). These characteristics of the data make comparisons between<br />
records, based on ppm-values only, difficult and sometimes even meaningless. It is even possible<br />
that, partially due to heterogeneous quality performance data, sectors such as metal and chemical<br />
industry do not show the desired consistency between the quality capability scores and the actual<br />
quality performance of the respective suppliers. In certain cases it is more meaningful not to rely<br />
on ppm records alone but to use additional indicators to describe supplier’s quality performance.<br />
Later on the author discusses also a number of additional biasing factors such as the validity<br />
of a particular audit evaluation over time, the proper definition of product groups used for categorization<br />
of audit evaluation scores, as well as the ”human” factor in the data including auditor and<br />
quality inspector ”calibration”. All these influences could potentially affect the consistency of the<br />
analyzed data and must be considered in order to assure the conclusiveness of the analysis. How-<br />
16
ever, to improve the structure of the presented argumentation, it is meaningful to first introduce the<br />
specifics of the automotive business sector and their implications on quality auditing.<br />
4 Specifics of the Automotive Industry<br />
Strategic decision making in any organization is a complex process which has multiple dimensions<br />
besides the quality aspects (especially important are the financial incentives). It is therefore the<br />
responsibility of quality managers to understand corporate strategies, to identify the quality challenges<br />
resulting from them and accordingly to counteract these challenges in order to guarantee<br />
the effectiveness of their quality management tools and eventually the high quality of final products.<br />
For these reasons the current section introduces the major aspects of the automotive sector, its<br />
driving factors and draws conclusions upon particularly important aspects relevant for the quality<br />
auditing process.<br />
4.1 Internationalization of the Automotive Operations<br />
Until the 1980s major automotive players had their operations concentrated mainly on the local<br />
markets and international presence, even though existing, was quite limited (Veloso & Kumar,<br />
2002). Moreover, most of the automotive sales were generated in the Triad – USA, Western Europe,<br />
and Japan. American automotive producers dominated the American market, Japanese car<br />
manufacturers dominated the Japanese market and the European players were present mostly in<br />
Europe. However, automotive producers had been facing a mature market in the Triad for decades<br />
with limited annual market growth and were looking for ways to expand their customer base. This<br />
is the reason why in the recent a few decades the international presence of automotive producers<br />
increased, which intensified the competition for market share. Local brands were highly affected<br />
by this global expansion. American producers for example lost about 20 percent of their local<br />
market share to Japanese car-makers (Veloso & Kumar, 2002). Similar trends were observed in<br />
Europe as well, though on smaller scale due to stricter trade regulations (Veloso & Kumar, 2002)<br />
and higher quality standard of the European manufacturers.<br />
The increase of international operations, however, was not limited only to the Triad. In the late<br />
1980s and the 1990s a number of new markets outside the Triad opened and offered opportunities<br />
to the automotive players to increase their market share. Governments in South America and Asia<br />
17
Million Units<br />
80<br />
60<br />
40<br />
Worldwide Automotive Production<br />
15,8 14,1 16,4 18,3 17,9 19,6 21,5 25,1 27,3 30,2 34,2<br />
35,3<br />
35,8<br />
46,5<br />
20<br />
0<br />
100%<br />
39,2 39,4 39,8 40,0 38,4 39,4 39,2 39,4 39,2 39,1 39,1 35,4<br />
25,9 31,1<br />
1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010<br />
75%<br />
29% 26% 29% 31% 32% 33% 35% 39% 41% 44% 47% 50% 58% 60%<br />
50%<br />
25%<br />
0%<br />
71% 74% 71% 69% 68% 67% 65% 61% 59% 56% 53%<br />
Rest of the world<br />
50% 42% 40%<br />
USA + Western Europe + Japan<br />
Figure 4: Shift of the automotive production capacity from the Triad (USA, Western Europe, and<br />
Japan) to other more cost competitive production locations. (Data Source: OICA, 1997 – 2010)<br />
alleviated strict governmental trade policies in their attempts to attract Foreign Direct Investments<br />
(FDI), (Veloso & Kumar, 2002). At the same time, the fall of communist regimes in the early<br />
1990s opened Eastern European markets and granted Western companies access to new customers<br />
also in this region.<br />
In recent years developing countries have been the motor of growth of global sales with regions<br />
such as China, South East Asia, Eastern Europe, South America, and India seeing the major<br />
growth. In the 1990s almost 80% of the automotive sales were still generated in North America,<br />
Europe and Japan (Veloso & Kumar, 2002). In the last decade, however, markets in Asia and<br />
South America soared and their growth was the major driving factor for an overall increase of<br />
global sales with about 55% to surpass 70 million vehicles (Veloso & Kumar, 2002). About half<br />
of the global sales now take place outside the Triad. The more favorable conditions on the newly<br />
opened markets prompted major automotive players to invest in these new regions and to relocate<br />
significant part of their production close to their new customers. In 1997 a mere 29% of all vehicles<br />
worldwide were produced outside the Triad (see Figure 4). In less than 15 years this share<br />
more than doubled, and 60% of the global automotive production output in 2010 came from the<br />
18
developing markets (Figure 4).<br />
Due to the extreme geographic dispersion<br />
of final customers nowadays it is very important<br />
for the automotive producers to run global<br />
operations. Usually an automotive OEM would<br />
run a number of production facilities strategically<br />
located within or close to its most important<br />
markets, with the majority of its key suppliers<br />
situated in close proximity. This not only<br />
reduces the uncertainty of deliveries to OEMs’<br />
final customers, but also significantly reduces<br />
transportation costs (transportation of voluminous<br />
products such as automobiles on long distances<br />
is quite expensive). Therefore, carmakers’<br />
decisions on where to position their production<br />
facilities are governed by the development<br />
of global markets.<br />
Million Units<br />
20<br />
15<br />
10<br />
5<br />
0<br />
Growth of Automotive Production<br />
in Developing Countries<br />
USA<br />
China<br />
Eastern and Central Europe<br />
South America<br />
South Korea<br />
India<br />
1997<br />
1998<br />
1999<br />
2000<br />
2001<br />
2002<br />
2003<br />
2004<br />
2005<br />
2006<br />
2007<br />
2008<br />
2009<br />
2010<br />
Figure 5: Growth of the automotive production<br />
in the developing countries<br />
(Data Source: OICA, 1997 – 2010).<br />
All these recent changes made today’s automotive<br />
business environment much more complex<br />
as compared to the one just a few decades ago. To defend their market positions major players<br />
need to adjust their strategies to the new market environment. On the one hand, market leaders<br />
need to defend their strong positions on the developed western markets and fend off new business<br />
rivals. Here competition does not come only from producers from other developed regions.<br />
Automotive players with recently gained momentum, such as the Korean Hyundai-Kia, currently<br />
the fourth largest automotive manufacturer worldwide, also strive to set strong foot on Western<br />
markets. Meanwhile, stagnant customer demand and limited sales in these markets exacerbate<br />
the price competition. On the other hand, automotive OEMs need to keep up with the headlong<br />
growth of developing markets and capture as large a market share as possible on these markets.<br />
Doing business in the newly opened regions in many cases can be particularly challenging due to<br />
specific governmental regulations which favor local producers. Thus for example in China foreign<br />
automotive manufacturers may only produce through local joint ventures with their participation<br />
limited to no more than 50 percent (Economist Intelligence Unit, 2011). Moreover, to develop key<br />
competences such as the production of smaller and cheaper cars, which correspond to the lower<br />
19
udget levels in these regions, plays a key role in securing a share of the sales. To support this<br />
business restructuring the supplier quality auditing process needs to be adjusted accordingly.<br />
Most of the leading automotive producers rely on supplier networks with long history of cooperation.<br />
The longer two parties work together, the better they know each other. Lasting business<br />
relationships are a sign of trust and this is the reason why OEMs prefer to work with old partners on<br />
new projects rather than enter in business relationships with new and unknown suppliers (Veloso &<br />
Kumar, 2002). However, investment in new plants in growing markets is a cornerstone in OEMs’<br />
business strategies for international operations. Particularly challenging for the expansion to developing<br />
markets turns out to be finding the proper local business partners, which are going to support<br />
their operations. Building lasting relationships with key suppliers in the new regions is therefore<br />
vital. As already pointed out, in many cases governmental regulations set particular requirements<br />
for the share of local content in the final product. In such cases it is necessary to source from the<br />
local market and accordingly work with local suppliers, with which the respective OEMs have no<br />
previous business partnerships. To mitigate the risks associated with starting up new facilities in<br />
developing markets major players invite their key suppliers to the new regions and try to replicate<br />
already existing supply structures (Veloso & Kumar, 2002). Automotive suppliers usually have to<br />
choose between opening an own production facility in the new region or transferring their knowhow<br />
to a local supplier (Veloso & Kumar, 2002). Such business moves are associated with great<br />
deal of risk and are therefore critical for the success of the outsourcing OEM on the respective<br />
market. This puts a particularly high responsibility on the quality audit.<br />
However, there are a number of common factors which make quality management in developing<br />
regions quite challenging. The high growth rates of these markets are a significant consideration.<br />
Production facilities should not only be up and running, but also project plans need to<br />
make room for capacity expansions and volume increases. Even when there are such plans, very<br />
often local suppliers are overwhelmed with new projects and fail to keep the quality requirements<br />
of their customers in focus. This leads very often to lower quality of the supplied components,<br />
increased part rejection rates at OEM’s side and jeopardizes the integrity of the entire production<br />
process. Avoiding such situations is crucial for OEMs in strengthening their positions on the new<br />
markets and this needs to receive a particularly high attention in the quality auditing process.<br />
Studies in the field of Human Resource Management show that there are several major problems<br />
faced by multinational corporations in developing countries also on the level of personnel<br />
management (Napier & Vu, 1998; Scullion, Collings, & Gunnigle, 2007). Scullion et al. (2007)<br />
20
write that ”countries such as India and China face shortages of suitably qualified and skilled employees<br />
for MNCs [multinational corporations] and local enterprises alike” (p. 311). For that<br />
reason it is difficult to recruit locally the necessary managerial force, which has the qualification<br />
to operate in these new environments (Scullion et al., 2007). Scullion et al. (2007) argue that<br />
such deficiencies generate ”a growing demand for expatriate employees” (p. 314) in the emerging<br />
markets. Furthermore, the authors write that it is difficult to persuade experts to transfer to these<br />
regions, and even when this happens, very often due to cultural differences these experts do not<br />
necessarily possess the required skills to manage in the new region (Scullion et al., 2007). At the<br />
same time high growth in emerging economies and the specifics of the market situation offer a lot<br />
of opportunities for career jumps. Very often companies in these markets face exceptionally high<br />
levels of employee turnover, which in turn inhibits the corporate learning process and makes proper<br />
knowledge management difficult to achieve. This leads to the conclusion that during quality audits<br />
in such regions it is especially important to put special emphasis on the personnel qualification and<br />
knowledge management of the suppliers.<br />
All these factors have negative influence on the quality capability of the individual suppliers<br />
which operate in these regions and especially on its sustainability. A number of re-qualification<br />
quality audits conducted by Volkswagen Group in 2010 in regions such as India and China revealed<br />
that the quality capability of their local business partners can drastically change for the<br />
worse just within several months 4 . In many of the cases the principal causes for the rapid decline<br />
in suppliers’ quality capability were related to production volume increases and high personnel<br />
fluctuations. Given the fact that emerging markets are especially important for the growth of automotive<br />
corporations and for sustaining their business, the quality assurance policies and especially<br />
the employment of quality management tools such as supplier quality auditing need to pay special<br />
attention and adequately respond to risks specific to growing markets.<br />
4.2 Rising Development Costs<br />
The next aspect of the automotive business which is relevant for quality auditing are rising development<br />
costs and the production approaches which automotive producers employ to reduce them.<br />
To cope with business challenges specific for the different regions it is essential for automotive<br />
producers to establish a product portfolio, which adequately captures the abundance of customer<br />
requirements. It is no longer sufficient to offer just a few models rather it is important to have a<br />
4 Source: personal communication with Volkswagen AG employees.<br />
21
wide product range. The number of market segments has significantly increased in recent years<br />
and the separation between the different market niches has lost its sharpness. To better respond<br />
to their customers’ requirements, OEMs constantly introduce new models, which are quite often<br />
tailored exclusively for a specific market. Such a strategy allows automotive OEMs to have access<br />
to a larger customer pool and therefore increase the amount of their sales. Furthermore, to<br />
spark additional sales the generally accepted standard in automotive business is to renew the model<br />
ranges in ever smaller time cycles which are nowadays in the order of just a few years.<br />
Developing a larger number of models much more frequently than in the past, however, is quite<br />
expensive. Meanwhile, even though currently rising – global automotive sales are nevertheless<br />
limited and the growing diversity of car models translates in fewer sales per model. Thus for example<br />
in the US the number of models offered on the market almost doubled from 550 in 1980<br />
to 1050 in 1999, while in the same time period the average number of yearly sales per model decreased<br />
from more than 20,000 units in 1980 to less than 15,000 sales per model in 1999 (Veloso<br />
& Kumar, 2002) despite an increase of the overall market volume of more than 40%. The smaller<br />
scale raises accordingly the development costs per sold unit. Given the high competitiveness characteristic<br />
for the automotive sector, providing good price for value is crucial in the race for market<br />
share. The rising development costs put the drive towards less expensive products to a significant<br />
challenge. A common practice for automotive OEMs is therefore to base a number of models on<br />
a single standardized platform such as Volkswagen’s ”A”, Fiat’s ”178” or General Motor’s ”Mid-<br />
Range” platforms (Veloso & Kumar, 2002). Platforms are a cornerstone of the automotive business<br />
operations for a number of reasons.<br />
On the one hand, the use of platforms offers a number of financial benefits, which have positive<br />
effect on a company’s market performance. Platforms allow car-makers to split development costs<br />
over several models and therefore achieve higher economies of scale. Lowering the development<br />
costs per sold unit respectively reduces the price of the individual products and has direct impact<br />
on customers’ satisfaction through better price for value. Furthermore, platforms offer a lot of<br />
flexibility and make it possible for the automotive OEMs to develop and sell a particular model<br />
practically worldwide with small modifications defined by the specifics of the individual markets.<br />
The use of platforms is also especially useful for the quick integration of technical advancements<br />
in the entire product range, which in turn gives the respective OEM a competitive advantage over<br />
its business rivals.<br />
On the other hand, the employment of standardized platforms affects positively a number of<br />
22
internal processes and increases organization’s operational efficiency and therefore its profitability.<br />
Platforms reduce the number of components and therefore simplify OEMs’ inventory management<br />
(van Weele, 2010). Meanwhile, the larger volumes per component accelerate the learning process<br />
in production, which eventually results in improved quality. An important advantage of standardized<br />
platforms is also risk-pooling. In cases of shortages of components for a particular model, the<br />
OEM can quickly counteract by filling in the inventories with the same components stock-piled<br />
for other models (Veloso & Kumar, 2002; van Weele, 2010). This is particularly useful to offset<br />
misguided forecasts of customer demand such as unexpectedly high interest in particular products<br />
and lack of interest in others (van Weele, 2010).<br />
Due to the significant advantages which the platform use offers, very often the whole product<br />
portfolio of a particular automotive group is based on just a few standardized platforms. Quite<br />
common are also cases in which automotive business competitors would form alliances among<br />
one another and develop joint platforms. Recently, more daring projects aim at the even further<br />
integration of individual models in OEMs’ product ranges and therefore larger cost reductions.<br />
One such example is Volkswagen’s MQB strategy (from German Modularer Querbaukasten or<br />
modular toolkit). The target of Volkswagen’s new strategy is to develop a universal platform for<br />
the construction of transverse, front-engine, front-wheel-drive vehicles, which is intended to be<br />
highly standardized, but at the same time flexible enough to be incorporated in a very large number<br />
of different models in several different market segments 5 . The use of the MQB is an important<br />
pillar in Volkswagen’s business strategy. It is expected that the introduction of the MQB platform<br />
would reduce the one-off expenditure and the unit costs by up to 20% each, and the engineered<br />
hours per vehicle by up to 30% (Pötsch, 2011). MQB is also particularly useful for significant<br />
weight reductions and respectively lower emissions. The first mass production models based on<br />
the MQB concept are expected to roll off the Volkswagen production lines already in 2012.<br />
Despite their immense advantages, standardized platforms represent still another significant<br />
challenge for the quality management process and in particular for quality auditing. The fact<br />
that they offer a way to suppress rising development costs, and at the same time allow OEMs to<br />
maintain product portfolios with a large number of models, gives platforms a central position in<br />
OEMs’ production strategies. However, potential defects in the individual standardized production<br />
components immediately affect all models using the platform and respectively a very large amount<br />
of vehicles. Such undesired situations not only have serious financial consequences, but also are<br />
especially harmful for the image of the respective company. In the end of 2009 and the beginning<br />
5 Source: personal communication with Volkswagen AG employees.<br />
23
of 2010 there was a major recall campaign of Toyota models. One of the reported causes for the<br />
Toyota problems was a malfunctioning accelerator pedal. Only in the USA a total of 12 different<br />
Toyota models and accordingly almost 7 million vehicles were affected by just this single problem<br />
(NHTSA Recall Report, 2011). The strikingly large number of vehicles, which Toyota recalled,<br />
is mainly due to the fact that all affected models used the same standardized component. The<br />
financial consequences of the recalls were massive. Toyota models in the US alone affected by the<br />
acceleration problems accounted for 57% of sales in 2009 (New York Times, 2010). Even worse –<br />
models of other manufacturers such as General Motors’ ”Pontiac Vibe [were] among the vehicles<br />
recalled because [they were] built through a joint venture with Toyota”, (New York Times, 2010).<br />
The Toyota recalls are a vivid example of how quickly a potential defect can propagate throughout<br />
a major portion of the product portfolio. Therefore, the growing standardization of components in<br />
the automotive production needs especially high attention in the quality management process and<br />
the quality auditing strategy in particular.<br />
4.3 The Automotive Supply Chain<br />
Another important development of the automotive sector affecting the quality auditing process is<br />
the growing complexity of the automotive supply chain. With the progressively larger number of<br />
similar products and intensifying competition, the ability to properly position your products on<br />
the heterogeneous automotive market is a key competence in the race for market share. Considering<br />
the large diversity of customer requirements and the therewith associated differences between<br />
individual markets, however, the proper management of the product portfolio requires a lot of resources.<br />
That is the reason why nowadays automotive producers get continuously less involved<br />
in manufacturing and concentrate significant portion of their capacities on enhancing the value<br />
of intangible concepts such as their brand image, which play an increasingly central role in the<br />
automotive business (Veloso & Kumar, 2002) OEMs can successfully restructure their core competences,<br />
however, only given that they cooperate with a number of carefully selected business<br />
partners, which overtake a major portion of the manufacturing process.<br />
Outsourcing is not a new concept to automotive manufacturers, as it dates back as early as the<br />
first years of automotive production. Nevertheless, outsourcing in the past would usually be limited<br />
only to individual components, which companies assembled into their final products. In recent<br />
years, however, due to the globalization of sales and the need to concentrate resources on other important<br />
competences, the amount of outsourced automotive production has drastically increased.<br />
24
Moreover, today OEMs tend to supply entire systems rather than individual components. As a<br />
result suppliers of assembly units gain increasing importance and a lot of the suppliers, which produce<br />
the individual assembly components and used to deliver these directly to the OEMs, operate<br />
now in the 2nd or higher tiers (see Figure 6). Due to such business restructuring, in the past a few<br />
decades the automotive supplier network has evolved into a very complex, highly interconnected,<br />
multi-layer, global structure. Nowadays between 60 and 80 percent of automotive production is<br />
outsourced (Veloso & Kumar, 2002). The number of just the 1st-tier suppliers the biggest automotive<br />
producers have is in the order of several hundred up to a few thousand, while the supplier<br />
base is growing ever bigger as the portion of outsourced production expands. In 2010 Volkswagen<br />
Group ran several pilots on sub-supplier management, which revealed that the average Volkswagen<br />
Group 1st-tier supplier has from 5 to 10 critical sub-suppliers, which are involved in a substantial<br />
part of its production process 6 . The results of this evaluation show that 1st-tier suppliers represent<br />
just the tip of the iceberg (reaching in the best case up to 10 to 20 percent) of the enormous<br />
pool of delegated manufacturing responsibilities and underscore once again how complex the real<br />
automotive supply chain is.<br />
Outsourcing plays a particularly important role in price reduction as well. OEMs strive to reduce<br />
the number of their business partners and at the same time increase the share of business<br />
for individual suppliers. The larger order quantities allow the outsourcing companies to negotiate<br />
better prices. However, there are particular differences between the approaches of individual<br />
automotive players in how far they would go with reduction of the number of their direct suppliers.<br />
Companies such as Renault and Volkswagen pursue a more conservative strategy and use the<br />
”two-plus-one” formula (Veloso & Kumar, 2002). This means that for every component there are<br />
two major suppliers with a third one, closely following behind. The third supplier delivers smaller<br />
quantities but meanwhile receives enough business so that in case one of the two primary suppliers<br />
cannot keep up with the deliveries the third one will step in and take over the orders. Ford on<br />
the other hand pursues a more aggressive strategy and wants to have a single supplier for each<br />
component or system (Veloso & Kumar, 2002). Currently Ford has a comprehensive catalogue of<br />
the prices of individual components and therefore an idea about what a whole system should cost.<br />
This gives Ford advantage in the negotiation process. In the future, however, Ford would know no<br />
more the prices of individual components but only the prices of the whole systems, which would<br />
give its suppliers a lot of power in the negotiations. One further advantage of the ”two-plus-one”<br />
strategy is that working with and accordingly auditing two or three suppliers provides a much more<br />
6 Source: personal communication with Volkswagen AG employees.<br />
25
comprehensive overview of the the state-of-the-art production and technology of the product, than<br />
when a company has an overview of the processes (and challenges) of a single supplier. In this<br />
regard the monetary aspect and the aspect of security of supply are complemented through the win<br />
of knowledge through auditing.<br />
Even though in most of the cases production alone is cheaper at the outsourcing company<br />
than at suppliers, companies still prefer to outsource a major portion of their production process<br />
(van Weele, 2010). The reason for that is that development of the components is particularly cost<br />
intensive. A major requirement for automotive suppliers is therefore that they possess the necessary<br />
development capabilities and take over not only the production process but the development of the<br />
respective components as well (Veloso & Kumar, 2002). Automotive OEMs expect that their<br />
suppliers will be able to offer lower prices by splitting the development costs over several of their<br />
customers. Furthermore, instead of sourcing individual components, the number of purchased<br />
complete systems increases. Usually this would require from suppliers to invest a great deal in<br />
development and be ready to first wait for several years before they can make any profit (Veloso &<br />
Kumar, 2002). Only suppliers which have the necessary development know-how and the required<br />
resources are therefore able to compete on the highest levels of the supply chain. This is the reason<br />
why in recent years the automotive supply market is dominated by large international suppliers.<br />
Suppliers with smaller financial resources cannot compete on global terms and usually restructure<br />
their business to position themselves in the lower tiers of the supply chain as local component<br />
suppliers of the large automotive system integrators (Veloso & Kumar, 2002, see also Figure 6).<br />
This increase of the depth of the automotive supply chain (lower tiers appear) means that quality<br />
auditing can achieve a representative overview of the quality risks down the supply chain only if<br />
second party audits concentrate not only on the first-tier suppliers but also on the lower tiers.<br />
One negative side of outsourcing is that it increases the quality risks and therefore influences<br />
the quality management process. As already explained in detail, nowadays companies are able<br />
to outsource significant parts of their production to external partners. Due to the high levels of<br />
outsourcing OEMs and the quality of their products are increasingly dependent on their suppliers,<br />
which gain an ever growing importance in the overall business process (van Weele, 2010). Any<br />
supply shortages or poor quality of the delivered components could lead to halts of the production<br />
and thus result into out-of-stock situations. The fact that today every automotive mass producer deploys<br />
an assembly line, makes the individual production steps interdependent and a disturbance at<br />
26
Tier 1<br />
Supplier 1<br />
Tier 2 (Tier n)<br />
Supplier 1<br />
Tier 1<br />
Supplier 2<br />
Supplier 2<br />
Supplier 3<br />
OEM<br />
Supplier 3<br />
OEM<br />
…<br />
Supplier N<br />
…<br />
Supplier N<br />
Trend of Development<br />
of the Automotive Supply Chain<br />
Figure 6: Increasing complexity of the automotive supply chain.<br />
Suppliers of simpler components and subsystems tend to position their businesses in the second and<br />
lower tiers of the automotive supply chain. The first tier is dominated by large system integrators<br />
(Veloso & Kumar, 2002).<br />
any process step has a large impact on the entire production chain. Considering the extremely competitive<br />
business environment such process variations can have significant financial repercussions<br />
leading to lost sales opportunities and consequently decrease of market share, therefore degrading<br />
the overall financial performance of the respective organization.<br />
At this point it is necessary to keep in mind that suppliers are independent business units and<br />
determine their business strategies autonomously. Very often the quality capability of suppliers<br />
is influenced by factors, which the automotive OEMs cannot control. Different managerial decisions<br />
within the individual suppliers such as changes of the workflow and process optimizations,<br />
investment in new technologies, restructurings of suppliers’ supply chains and sometimes even the<br />
relocation of entire production facilities influence the stability of suppliers’ production processes.<br />
Nowadays such modifications happen rather frequently in many business sectors and target continuous<br />
improvement and price reductions. Very often such changes degrade the quality of the<br />
delivered products and endanger the continuity of OEMs’ workflow. It is therefore very important<br />
for the outsourcing company to follow closely major changes down the supply chain and take<br />
these factors into account in its general supply management policy and the auditing frequency in<br />
particular.<br />
27
The concentration of a large portion of business in a few suppliers transforms the structure of<br />
their business operations. As indicated before, most of the 1st-tier automotive suppliers nowadays<br />
deliver entire integrated systems, rather than single components. To develop and produce these<br />
systems suppliers need a great deal of specific technical knowledge and therefore they, too, resort to<br />
outsourcing. The number of business partners of the largest suppliers on the automotive market can<br />
easily reach a few hundred. Increasingly important for such corporations is the ability to manage<br />
the quality risks down their own supply chain. Volkswagen experience shows that a significant<br />
portion of the problems originating in the supply chain are due to issues in the production processes<br />
of sub-suppliers 7 . This is the reason why quality audits at such suppliers must put special emphasis<br />
on their supplier management capabilities.<br />
4.4 Implications for the General Automotive Quality Auditing Strategy<br />
The previous several sections presented the specifics of the automotive business and already outlined<br />
a number of quality related challenges, which need to be addressed by the general strategy<br />
for quality auditing. The current section summarizes the insights gained by the discussion above<br />
and organizes the aspects relevant for the quality auditing process, which result from the specifics<br />
of the automotive sector. Figure 7 provides a structured summary of the said in the previous a<br />
few sections. Based on this summary an idealized priority listing is derived, which provides a<br />
guideline for defining the general quality auditing plan, i.e. addressing the ”sampling frequency”<br />
aspect of quality auditing (see Section 3). At the end of the section the quality auditing process of<br />
Volkswagen is compared to the results of the discussion.<br />
Several important conclusions stem from the automotive discussion above. A key point is<br />
that outsourcing has received a central role in the automotive business today and the viability<br />
of car-makers’ operations is highly dependent on the ability of their suppliers to support OEMs’<br />
core business processes. Quality is a determining factor for success in the extremely competitive<br />
automotive business environment and therefore quality assurance in the entire production chain<br />
has to be top priority for every single market player. Due to the specifics of the sector and the<br />
increased complexity of the manufacturing process in recent years, however, assuring quality in<br />
the automotive industry and especially in the supply chain has become particularly challenging.<br />
Therefore suppliers’ quality capability should be followed regularly with the help of quality audits.<br />
7 Source: personal communication with Volkswagen AG employees.<br />
28
In this respect evaluation of the quality capability of suppliers’ production processes should be<br />
carried out not only before awarding a particular contract but also regularly throughout the series<br />
production. To maximize the effectiveness of quality auditing it is important that individual quality<br />
Internationalization of Automotive Sales<br />
Business Challenges Strategic Solutions Quality Challenges<br />
• Mature Western market with<br />
local presence only<br />
• Limited opportunities for<br />
expansion of the customer<br />
base<br />
• Internationalization of the<br />
operations<br />
• High growth in emerging<br />
markets (opportunities for<br />
organizational growth and<br />
increase of market share)<br />
• Production shifts to cost<br />
competitive countries<br />
• Closer to new customers<br />
• Too high growth in emerging<br />
markets (negative impact on<br />
quality)<br />
• Governmental regulations for<br />
local content (work with new<br />
and unknown suppliers,<br />
increased quality risk)<br />
• Lack of qualified personnel,<br />
high personnel fluctuations<br />
(inhibits corporate learning,<br />
unstable production<br />
processes)<br />
Rising Development Costs<br />
Business Challenges Strategic Solutions Quality Challenges<br />
• Large variety of customers<br />
requires large product portfolio<br />
with many models<br />
• Renew model ranges in<br />
shorter cycles (shorter time for<br />
return on investment)<br />
• Fewer sales per model (larger<br />
development cost per sold<br />
unit)<br />
• Strong competition (pricing<br />
pressures)<br />
Standardized platforms<br />
• Split development costs over<br />
several models (lower<br />
development cost per unit)<br />
• Market flexibility (same model<br />
worldwide with market specific<br />
modifications)<br />
• Quick integration of technical<br />
advancements<br />
• Simplify inventory mgmt.<br />
• Shorter learning cycles<br />
• Risk pooling<br />
• Potential defects immediately<br />
affect all models based on the<br />
platform<br />
• Quality risk concentration in<br />
very high volumes<br />
• Quality problems have very<br />
high financial consequences<br />
and can damage company<br />
image (Toyota)<br />
The Automotive Supply Chain<br />
Business Challenges Strategic Solutions Quality Challenges<br />
• Very strong competition<br />
• Product portfolio and brand<br />
image management require<br />
resources (determine market<br />
success)<br />
• Rising costs of operation<br />
• Outsourcing gains growing<br />
importance (60% to 80% and<br />
growing)<br />
• Source entire systems (free<br />
resources)<br />
• Fewer suppliers with larger<br />
order quantities (price<br />
reduction)<br />
• Suppliers overtake also<br />
development costs<br />
• Growing complexity of the<br />
supply chain (1 st -tiers<br />
potentially only 10% to 20% of<br />
all suppliers)<br />
• OEM dependent on suppliers<br />
(vulnerable to supply<br />
shortages and poor quality)<br />
• Suppliers autonomously<br />
determine their business<br />
strategies (affect OEMs<br />
processes anyways)<br />
• Suppliers’ ability to manage<br />
their suppliers is critical for the<br />
entire manufacturing process<br />
Figure 7: Quality challenges resulting from the market strategies of automotive producers.<br />
29
audits are properly managed and are incorporated in a well-structured quality auditing plan. It is<br />
essential that the strategy for the employment of quality auditing adequately reflects the market<br />
situation and responds to the individual quality challenges specific to it.<br />
While the quality audit should address problems in the entire supply chain, parts of the business<br />
process which are characterized with high levels of variation and increased quality risks need<br />
particularly high attention. The latter should be audited more frequently than the rest and companies<br />
need to concentrate more resources on early problem detection in these problematic areas (see<br />
also the discussion on sampling in Section 3). One especially critical step is the start of business<br />
relationships with new and unknown suppliers. These need to be audited more frequently in order<br />
to make sure that the new business partners understand the quality requirements of the outsourcing<br />
organization and reduce the quality risks associated with the implementation of the respective<br />
projects. As the two business parties gain trust in each other and their business relationship stabilizes<br />
the audit frequency can be reduced (see Section 3). This consideration is particularly relevant<br />
for developing regions where OEMs have to work with a number of new business partners, while<br />
at the same time rapid growth and high employee turnover have negative impact on quality sustainability<br />
in these regions (Figure 7, see also Section 4.1). Due to the high concentration of quality<br />
related issues, audits in emerging markets should be rather frequent.<br />
Another particularly important point, which needs to be considered in the general quality auditing<br />
plan, are the increasingly integrated product ranges (see Section 4.2). The use of standardized<br />
platforms increases the volumes of certain components and therefore also the cost of potential<br />
problems (Figure 7). Cases in the past have shown the detrimental effects, which poor quality of<br />
standardized components has on the entire business of an organization (the Toyota recalls from<br />
2009 and 2010 are mentioned above). For that reason suppliers which deliver standardized platform<br />
components or entire integrated systems should be handled with particular care and be audited<br />
quite frequently as well. The frequency of the audits should be determined by the complexity of<br />
the delivered components, industry sector, maturity of the according technology, order volumes as<br />
well as the specifics of the region in which the respective production facilities are located. Due to<br />
the increase in deliveries of entire systems lately and the therewith associated increased complexity<br />
of the sub-supplier network (see Section 4.3), it is also important to pay special attention in the<br />
quality audit planning to sub-supplier management capabilities of the respective suppliers. This<br />
concerns particularly suppliers with high number of sub-suppliers.<br />
30
Finally, the quality auditing process has to encompass the entire supplier pool, meaning that<br />
suppliers without any quality related problems and non-critical production processes should still be<br />
subject to regular quality auditing, however, not with the same intensity as in the previous cases.<br />
All the points mentioned above can be used to prioritize quality audits and serve as a basis for<br />
the definition of a quality auditing plan, which accounts for the specifics of the automotive sector.<br />
Figure 8 presents an idealized priority list for planning second-party quality audits.<br />
High Priority<br />
(Very Frequent Auditing)<br />
• Suppliers with critical quality problems<br />
• New and unknown suppliers<br />
• Suppliers in fast developing regions<br />
(focus on personnel qualification and knowledge management)<br />
• Suppliers of components for standardized platforms<br />
Medium Priority (Frequent Auditing)<br />
• Suppliers with large supply networks such as suppliers of complex<br />
automotive systems (focus on sub-supplier management)<br />
• Suppliers with minor quality problems<br />
• Suppliers with critical production processes<br />
Low Priority<br />
(Regular re-Auditing)<br />
• Regular requalification of suppliers without quality problems<br />
Figure 8: Idealized priority list for planning second-party quality audits in the automotive sector.<br />
Figure 9 presents the average age of the audit evaluations performed by Volkswagen auditors in<br />
different regions worldwide. The empirical data shows a lot of similarities with the recommendations<br />
for the audit planning presented in Figure 8. Thus for example, in developed regions such as<br />
North America and Germany the average age of the performed audits is considerably larger than<br />
the age of audits in developing regions such as China, India, or Russia. In the light of the discussion<br />
in Section 4.1 it is reasonable to expect that the company faces indeed a lot more quality challenges<br />
in the latter three regions and therefore needs to evaluate its suppliers there more frequently.<br />
Volkswagen’s auditing strategy accounts also for the growing complexity of the supplier pool<br />
(see Section 4.3). In recent years the capability of Volkswagen’s first tier suppliers to manage<br />
their own supply chains has received increasing importance and the number of the conducted subsupplier<br />
audits has been steadily increasing 8 . One particular point, which definitely has to be<br />
8 Source: personal communication with Volkswagen AG employees.<br />
31
Average Time Period Since the Last Audit<br />
North America<br />
Germany<br />
England<br />
Spain<br />
Eastern Europe<br />
Mexico<br />
Brazil<br />
China (VGC)<br />
Argentina<br />
India<br />
South Africa<br />
Russia<br />
Time<br />
Figure 9: Average ”age” of the audit evaluations of Volkswagen Group suppliers sorted by region.<br />
Note: VGC = Volkswagen Group of China<br />
considered next in Volkswagen’s quality audit planning, is the start of production of vehicles based<br />
on the new MQB platform, which is planned for 2012. Due to the increased weight of potential<br />
problems originating from MQB components, the respective suppliers will have to be audited<br />
rather frequently.<br />
These results show that the general auditing strategy of Volkswagen adequately responds to the<br />
quality challenges in the automotive sector (at least qualitatively). On the other hand, to determine<br />
whether the absolute quality auditing frequency is suitable for the specifics of the respective regions,<br />
it is required a more comprehensive analysis of the specifics of the according region, which<br />
is not subject of this paper.<br />
5 Fundamental Principles of the Quality Audit<br />
After Section 4 introduced the specifics of the automotive industry and summarized the insights<br />
resulting from this discussion with respect to the general strategy for employing quality auditing,<br />
the current section introduces the technical aspects of the implementation of a single quality audit.<br />
32
This information is particularly helpful for the clarity of the discussion presented in the remaining<br />
part of this paper.<br />
There are three types of quality audits – first-, second- and third-party. All three types assess<br />
the capability of particular processes to provide high-quality products or services. There are certain<br />
implementation differences, however, which distinguish the individual types of audits from<br />
one another. A first-party, also called an internal audit, is conducted within an organization on<br />
internal demand by internal auditors. The purpose of this type of audit is to evaluate the own<br />
quality management system and to identify possible improvements in the work of the organization<br />
(Parsowith, 1995; Wealleans, 2005). Second-party quality auditing is performed by an organization<br />
to its first-tier (sometimes second- and even higher-tier) suppliers in order to ”prove the<br />
’technical’ capability of the [supplier] to provide the product or service”, ”check that [the supplier]<br />
has sufficient capacity and resources to cope with the customer demands”, and ”assess the rigor of<br />
[supplier’s] operational processes” (Wealleans, 2005, p. 5). As already discussed in the previous<br />
sections, second-party auditing in the higher tiers of the supply chain is of increasing importance<br />
for automotive manufacturers due to the growing complexity of their supplier network. Sometimes<br />
audits are performed by an external independent party, whose purpose is to preserve objectivity<br />
– third-party auditing. This type of audit could be performed by a company hired by the main<br />
contractor to review the work of its subcontractors or more commonly a company would hire an<br />
external auditor to review its own quality management system and its conformance to the generally<br />
accepted quality management standards (Parsowith, 1995; Green, 1997; Wealleans, 2005).<br />
Extending the production footprint by outsourcing could be regarded as an expansion of manufacturer’s<br />
factory. Therefore the employment of first- and second-party quality audits together<br />
assure control over the entire production process from the development phase of a certain component<br />
down the production chain to its placement in the final product. One could assume that<br />
third-party quality audits could have a similar function as second-party auditing. The parties involved<br />
in a second-party quality audit, however, are not so much interested in systems rather in the<br />
technical and commercial specifics of their relationship (Wealleans, 2005). This makes third-party<br />
audits more suitable for quality assurance in the sense of certifying and making sure that particular<br />
production and business processes comply with general quality standards.<br />
Even though all three types of quality auditing are an important part of the overall quality<br />
assurance process, the focus of this work falls upon second-party quality audits. Second-party<br />
quality auditing is a process, which can be divided into several individual implementation stages.<br />
33
The audit stages include audit planning, initiation and preparation, on-site evaluation, reporting<br />
and closure, (Arter, 1989; Parsowith, 1995; Wealleans, 2005).<br />
5.1 Audit Planning, Initiation and Preparation<br />
The first thing that could make the quality audit ineffective is its inadequate integration into the<br />
quality management system in terms of organization and technical content. This makes general<br />
audit planning a vital part of company’s quality assurance program and often it could be regarded as<br />
a separate phase of the auditing process (Wealleans, 2005). Audit planning, in this sense, deals with<br />
the overall quality assurance strategy. It covers scheduling quality audits within the entire planning<br />
horizon of an organization and putting the audit into strategic and general business context. These<br />
aspects were already discussed in the context of the automotive industry in Section 4. Naturally,<br />
areas, which represent high risk or contain serious quality related problems, as well as areas, which<br />
are prone to significant production process variation (e.g. suppliers in new regions with high<br />
personnel fluctuation), will be subject to more frequent auditing (Wealleans, 2005).<br />
In addition to scheduling, an important aspect of audit planning is scoping. Scoping identifies<br />
the areas which will be covered by each individual audit and usually involves identifying the<br />
facilities, processes, products, activities and quality assurance systems which will be reviewed,<br />
(Parsowith, 1995). Defining the appropriate scope of an audit can have a strong impact on the<br />
validity and relevance of the audit results. Making the scope too broad hides the risk of making the<br />
audit too superficial forcing the auditors to rush through the topics under the pressure of limited<br />
time. A narrow scope of the audit, on the other hand, allows for in-depth analysis of a particular<br />
problem, but also hides the risk of inefficient resource allocation (Parsowith, 1995; Wealleans,<br />
2005).<br />
Once the general audit planning is finalized, a lead auditor will be assigned to each individual<br />
audit and he has to take care of the rest of the preparation. His first task will usually be to select<br />
the rest of the team members based on their competence in the audited area. Depending on the<br />
width and depth of the audit the audit team size may vary. Most of the available sources agree<br />
that using more than six auditors on the team will make the audit very difficult to manage (Arter,<br />
1989; Parsowith, 1995; Wealleans, 2005). On the other hand, there are different opinions about<br />
the minimal size of an audit team. While Parsowith (1995) recommends that the team should<br />
comprise of at least two auditors, Wealleans (2005) does not exclude the cases in which a single<br />
auditor performs the process quality evaluation and even argues that in some particular cases this<br />
34
could be more efficient than having a larger team size. Given the usually limited resources in<br />
the real business world and the large amount of work, such practice increases the efficiency of<br />
the auditing units and saves costs. Volkswagen auditors for example perform supplier quality<br />
evaluations individually and sometimes in teams of two. The audit teams rarely exceed two people.<br />
By contrast, quality audits of Porsche are conducted by larger teams reaching in some cases up to<br />
five or six people.<br />
Furthermore, Parsowith (1995) suggests that in order to keep the objectivity of the audit and<br />
avoid biasing of conclusions, consequent audits within a certain area to be performed by different<br />
auditors. Wealleans (2005), by contrast, argues that assigning the same auditor for subsequent<br />
audits saves time, since the auditor will already be acquainted with the audited area and will need<br />
less time for preparation. Nevertheless, he acknowledges that if a particular auditor runs audits<br />
within the same area or organization too frequently the auditing procedure would become stale.<br />
Volkswagen pursues the strategy to build strategic audit competence, meaning that Volkswagen<br />
quality auditors develops specific process know-how and each auditor performs audits mainly in<br />
his area of competence (usually no more than two or three audits in a row at a particular supplier<br />
are performed by the same auditor). This approach enables Volkswagen to have a large number of<br />
experts possessing profound knowledge of the state-of-the-art of strategic production processes.<br />
After the audit team has been assembled, its members will have to gather all required information<br />
and get acquainted with the areas and processes, which have to be evaluated (Parsowith, 1995;<br />
Green, 1997; Wealleans, 2005). The audit team has to collect information about the delivered products,<br />
what contracts are currently in place, what the customer-supplier relationship at hand is, how<br />
important the supplier is for the overall business of the customer, and what information is there<br />
about past performance of the supplier (getting this information for new suppliers may not be fully<br />
applicable), (Wealleans, 2005). Among the documents a Volkswagen auditor prepares for an audit<br />
are technical drawings of the audited parts, specific testing requirements for the components (e.g.<br />
flammability test (TL 1010) for interior components), applicable legal regulations and standards,<br />
supplier’s quality performance record and current quality problems, etc. Wealleans (2005) recommends<br />
further that even though gathering the information may start quite early in the preparation<br />
process, the actual preparation to happen as close as possible to the audit itself. Going through the<br />
relevant documents would help identify spots which need improvement, clarification, or in-depth<br />
study. Preparation should result into obtaining detailed notes on the topics which need to be addressed,<br />
which later on could be used to prepare audit checklists (Wealleans, 2005). Wealleans<br />
(2005) does not recommend the use of standardized checklists, however, since auditors tend not to<br />
35
prepare for an audit and rely on the predefined questions, rather than on their investigation skills.<br />
Instead he recommends the use of the so-called process-based approach where individual steps of a<br />
process are described on a sheet of paper together with the relevant inputs and outputs (Figure 10).<br />
During the on-site evaluation information is gathered about all supporting elements of the process.<br />
Enablers:<br />
Materials<br />
Methods<br />
Manpower<br />
Machines<br />
Environment<br />
• …<br />
• …<br />
• …<br />
• …<br />
• …<br />
• …<br />
• …<br />
• …<br />
• …<br />
• …<br />
Measurement and Monitoring<br />
• …<br />
• …<br />
• …<br />
Inputs<br />
Process<br />
Outputs<br />
• …<br />
• …<br />
• …<br />
• …<br />
• …<br />
• …<br />
• …<br />
• …<br />
• …<br />
• …<br />
Figure 10: Process preparation sheet. Note: The figure was prepared according to Figure 9.2 on<br />
page 182 in (Welleans, 2005).<br />
Due to the limited time available for the actual audit, effective time management is a must.<br />
Thus an important part of the preparation for the audit is designing a detailed plan of action for<br />
the on-site evaluation (Arter, 1989; Parsowith, 1995; Wealleans, 2005). It should provide the time<br />
frame for the individual evaluation steps. Scheduling the supplier assessment process in advance<br />
will help team members to focus on their tasks and serve them as a reference to check on their<br />
progress with respect to the remaining time till the end of the evaluation. This part of the audit<br />
preparation is extremely important since good initial planning would ease the actual evaluation and<br />
increase the efficiency of the audit and its effectiveness in identifying issues critical to quality.<br />
Once the plan of action has been finalized a formalized copy of the audit action plan has to<br />
be approved by customer’s management and an official contact with the supplier would follow.<br />
36
The audit plan together with the names of all audit team members will be sent to the supplier and<br />
the dates of the on-site evaluation will be fixed. Sending the initial contact should happen well in<br />
advance in order to give the supplier enough time to prepare for the upcoming evaluation. Usually<br />
a Volkswagen quality audit is scheduled from several weeks up to two or three months in advance.<br />
It is supplier’s responsibility to organize the logistics of the on-site evaluation such as to allocate<br />
sufficient rooms for the meetings, assign guides for the audit team members, provide transportation<br />
between different facilities if necessary, etc.<br />
5.2 On-site Evaluation<br />
This is the most important part of the quality audit and covers the time span between the arrival<br />
of the audit team at supplier’s site and their departure (Parsowith, 1995; Wealleans, 2005). This<br />
part has a great direct impact not only on the validity of a particular process quality capability<br />
evaluation, but also on the work of the entire customer organization. Conclusions based on the<br />
information gathered during the on-site evaluation are decisive for the future of the manufacturersupplier<br />
relationship and the ’health’ of the overall production process. Positive audit evaluations<br />
are a necessary prerequisite for contract awarding in the Corporate Sourcing Committee (CSC) of<br />
Volkswagen (see also Figure 12 in Section 6.1). The structure of a particular audit as well as the<br />
methods used to collect data may vary depending on factors such as type of industry, scope of the<br />
audit, available resources, etc. (Arter, 1989; Parsowith, 1995; Wealleans, 2005).<br />
Every on-site evaluation starts with an opening meeting which includes the entire audit team,<br />
supplier’s management, and the according staff responsible for the areas which fall within the<br />
scope of the audit. The meeting is used to make the necessary introductions and to restate the<br />
aims of the upcoming evaluation (Wealleans, 2005). The auditors should address topics such as<br />
the scope of the audit as well as quality and product specifications to which the company is being<br />
audited (Parsowith, 1995). Additionally, the logistics of the audit as well as a tentative plan of<br />
action are negotiated taking into account supplier’s operational hours and resource availability.<br />
Subsequently the audit team collects objective data, which is used to assess the quality capability<br />
of supplier’s production processes (the guideline used by Volkswagen auditors for collecting<br />
objective evidence during an audit – Formel Q-Fähigkeit (2009) – is introduced in detail in Section<br />
6.1 below). Auditors pay particular attention to critical process steps and address known<br />
production problems, which have been identified during the audit preparation. Among the reviewed<br />
areas are process control tools and parameters such as SPC charts and cpk-values, results<br />
37
from different control tests and part release protocols at various process stages, documentation<br />
subject to legal regulations for storage (duty of documentation, TLD), emergency plans, etc. If<br />
during the on-site evaluation particular quality-relevant weaknesses of the production processes<br />
are identified, the supplier is responsible to define and implement the according corrective actions.<br />
During the audit auditors also need to set aside sufficient time to discuss previous evaluations and<br />
determine the progress of particular corrective actions defined in the past. At this stage it is very<br />
important that the auditors are well acquainted with the state-of-the-art of the evaluated process.<br />
This knowledge helps them to identify potential problems easier in the limited amount of time<br />
and therefore increases the quality of the evaluation. As mentioned above Volkswagen pursues the<br />
strategy to develop its auditors in particular areas of specialization, which gives the company the<br />
advantage to possess profound knowledge about strategic production processes and technologies<br />
in the automotive industry.<br />
Several auditing approaches for data collection are employed during a particular quality audit.<br />
The quality of the evaluation is influenced by the proper selection of an auditing approach, whose<br />
suitability is determined by the specifics of the situation and the type of evaluated information. The<br />
most common approaches used in the practice are tracing, corroboration, and sampling (Parsowith,<br />
1995; Arter, 1989; Wealleans, 2005). Tracing, for instance, has several variances as the most<br />
commonly used in second-party auditing are contract tracing and flow-of-work tracing (Wealleans,<br />
2005). When tracing a contract the auditor would investigate the quality assurance system and<br />
its impact on the contract of interest. On the other hand, monitoring the flow of work allows<br />
the process interfaces to be investigated in detail. This is of particular interest due to the fact<br />
that most quality related problems usually appear at these interfaces. This type of auditing is<br />
beneficial primarily for the auditing organization as it provides a high degree of assurance that<br />
the contract related job is taken proper care of (Parsowith, 1995). Volkswagen audits are usually<br />
structured in a way that they follow the flow of work from the input of raw materials through<br />
the process stages down to the final product (the Volkswagen auditing questionnaire is described<br />
in detail in Section 6.1 below). Another important auditing approach is corroboration. In this<br />
approach data sources are cross-referenced as this allows for making sure that collected data are<br />
accurate. The facts must agree based on at least two different auditors, two different records, two<br />
different interviews, or any combination of these (Parsowith, 1995). Finally, sampling is used for<br />
the collection of physical data. This technique is particularly useful for controlling large sets of<br />
data and calibration. This is the approach which is usually used by Volkswagen auditors to carry<br />
out product audits as part of the quality audit (see Section 6.1).<br />
38
Audit data collected during the on-site evaluation can be divided according to four main categories:<br />
physical evidence, sensory observation, comparisons and trends, and interviews and questioning<br />
(Parsowith, 1995). Generally speaking recorded data could be qualitative or quantitative<br />
as both types have their advantages. The main advantage of qualitative data is that it allows for<br />
the in-depth study of selected issues without the necessity to fit data into certain predetermined<br />
categories (Patton, 1987). On the other hand, quantitative data is very useful when great amounts<br />
of data have to be processed. This type of evidence offers ”statistical aggregation of the data” and<br />
”gives a broad, generalizable set of findings” (Patton, 1987, p. 9).<br />
5.3 Reporting and Closure<br />
The information gathered during the on-site evaluation is used as a basis for shaping an opinion<br />
about supplier’s capability to meet the particular quality and commercial requirements of the<br />
outsourcing organization. Once the on-site evaluation has come to an end, the audit team has to<br />
prepare an official report based on the factual observations gathered during the audit. All identified<br />
problems have to be supported with substantial evidence. The report should be submitted to the<br />
supplier shortly after the completion of the on-site evaluation. Based on the report the auditing<br />
organization may assign follow-up audits in order to revisit areas which need improvement. The<br />
supplier, on the other hand, should prepare a corrective action plan which describes the measures<br />
that will be taken to remove the observed problems together with the respective deadlines and perform<br />
the necessary quality improvements. The important aspects of the discussion presented in<br />
this section are summarized in Figure 11.<br />
6 Empirical Data<br />
The preceding sections treated the conceptual and general aspects of quality auditing. Section 4<br />
elaborated on the broad context of this discussion and derived the consequences for quality auditing<br />
resulting from the specifics of the automotive industry, while Section 5 presented in detail the<br />
technical structure of the second-party quality audit. Now that the general framework has been<br />
introduced, the remaining part of the paper deals with the analysis of Volkswagen specific empirical<br />
data and concentrates on the research questions defined in Section 2. The aim of the current section<br />
39
Audit planning<br />
• incorporate quality audit in the general quality assurance strategy<br />
• prepare an audit schedule<br />
• identify problematic areas for frequent auditing<br />
• define audit scoping<br />
Multiple Audits<br />
Initiation and preparation<br />
• determine lead auditor and the audit team<br />
• gather initial information (technical documents, legal and quality<br />
requirements, quality performance of the supplier)<br />
• prepare audit plan<br />
• schedule the quality audit with the supplier<br />
On-site evaluation<br />
• opening meeting<br />
• collect objective data on-site (tracing, corroboration, sampling)<br />
• identify problematic areas for improvement<br />
Reporting and Closure<br />
• evaluate collected evidence<br />
• prepare and present the audit report<br />
• agree upon corrective action plan<br />
• assign follow up audits<br />
Single Audit<br />
Figure 11: Stages of the second-party quality auditing process.<br />
is to introduce the types of information, which are going to be evaluated in Section 7, and to<br />
describe the processes, which generate these data.<br />
6.1 Quality Capability<br />
The quality capability scores of Volkswagen suppliers result from on-site evaluations of their production<br />
processes, i.e. second-party audits. Volkswagen quality audits are carried out according to<br />
a product group classification, which accounts for the specifics of the individual production processes<br />
and is maintained by the Volkswagen Quality Assurance. According to this classification,<br />
a particular product group organizes a number of components based on similarities in their production<br />
(e.g. light-metal alloy parts, welded parts, sensors, etc.). Due to such production process<br />
similarities, Volkswagen assumes that if a supplier has the know-how to produce a component in<br />
a specific product group, it is also capable to produce other components from the same product<br />
group. Therefore, a particular Volkswagen audit result is valid for the entire range of products<br />
within a particular product group. This means that, if a supplier is nominated for the delivery of a<br />
new component from a product group, for which it has already been audited, no new audit will be<br />
required. If the new component is part of a non-audited product group, however, the nomination<br />
will not be considered unless the respective supplier achieves a satisfactory result in a quality audit<br />
40
of the respective product group(s). The latter is very common for new suppliers, which have no<br />
previous business relationships with Volkswagen, as well as for existing Volkswagen suppliers,<br />
which extend their product ranges to new production areas.<br />
The product group classification is particularly important also for this work. It is used to divide<br />
the empirical data into subsets and account for the differences between the individual production<br />
processes, and potentially any negative influences such process differences might have on the final<br />
results of the analysis. In the general case, the effective definition of the product group classification<br />
is subject to several important considerations. A too narrow classification would require a<br />
significantly larger amount of audits and thus unnecessarily would increase the demand for auditing<br />
manpower, due to the fact that similar production processes would be part of different product<br />
groups. In case a certain product group is defined too generally, however, there is the risk that<br />
suppliers, which are good at making one of the products, will get certified to deliver other products<br />
in the same product group, but lack the necessary capability to produce them. In the latter case<br />
a supplier would be awarded a contract for new delivery without being re-audited, even though it<br />
may lack the necessary process know-how. Such situations are a potential reason to observe poor<br />
correlation between a supplier’s quality evaluation and its actual quality performance.<br />
The product group approach provides a convenient way to classify audit evaluation data and is<br />
extensively used in the analysis presented here. Nevertheless, there are still slight differences in<br />
the production cycles even for similar components in the same product group. In the end the implementation<br />
of a particular production process is company specific. For that reason Volkswagen<br />
regards the individual production processes as a combination of processing steps, e.g. painting,<br />
electric welding, injection moulding (plastics). The latter are evaluated separately during a quality<br />
audit and the overall quality capability score summarizes the individual process step evaluations.<br />
Depending on the complexity of the components (and accordingly the associated production processes)<br />
two audit evaluations of the same product group may contain different number of process<br />
steps.<br />
The quality capability metric used in this analysis is generated in a clearly defined algorithm<br />
described in Volkswagen’s Formel Q-Fähigkeit (2009). Formel Q-Fähigkeit is one of a set of documents<br />
describing the quality policy of Volkswagen, and contains among others a comprehensive<br />
supplier evaluation questionnaire, which provides a guideline for conducting quality audits. Due<br />
to the fact that Volkswagen cooperates closely with a number of national and international quality<br />
organizations such as VDA (Verband der Automobilindustrie) and IATF (International Automo-<br />
41
tive Task Force) its supplier evaluation questionnaire draws a lot of similarities with established<br />
automotive standards such as VDA 6.3 or ISO/TS 16949. The questions in this questionnaire<br />
are divided into three primary blocks, which address aspects, regarded as vital for maintaining a<br />
stable, quality-capable production process at supplier’s site. Particular areas of interest during a<br />
Volkswagen quality audit are suppliers’ capability to manage its own supply chain, the effective<br />
application of quality management throughout the entire production process, and finally customer<br />
care and service (Formel Q-Fähigkeit, 2009). In the general case the use of such a guideline is particularly<br />
useful for any organization, since it enables the organization to maintain the same quality<br />
auditing standard on an international level and helps in communicating organization’s quality requirements<br />
to its business partners.<br />
The focus of the first block of questions in the Volkswagen specific questionnaire falls upon<br />
topics regarding the supply chain management capabilities of the respective supplier. Volkswagen<br />
requires from its suppliers to regularly evaluate their own subcontractors and make sure that only<br />
certified and approved business partners are engaged in the production processes. This part of the<br />
questionnaire is very important, especially for suppliers with large amount of purchased components.<br />
As already mentioned in the discussion in Section 4.3 these are usually suppliers of complex<br />
automotive systems, which also rely on plenty of outsourcing to sustain their businesses just like<br />
the automotive OEMs themselves. Quality of the incoming materials needs to be continuously<br />
monitored and evaluated, and in case of deviations from the quality specifications the according<br />
Volkswagen supplier needs to agree upon improvement actions with its own suppliers and attend to<br />
their implementation. For that reason suppliers are required to have the necessary laboratory and<br />
measurement facilities for incoming material inspection. Volkswagen suppliers bear the responsibility<br />
to make sure that all purchased materials are sampled and released before they enter the<br />
series production. Furthermore, this part of the supplier evaluation deals with material handling<br />
and warehousing of raw materials at supplier’s site. Particularly important is to store the incoming<br />
materials under conditions, which preserve their physical properties. The materials have to be<br />
properly labeled to assure traceability and be used in a first-in-first-out manner (FIFO).<br />
The second block of questions deals with the individual production steps. Focus of this part of<br />
the supplier evaluation is personnel qualification, suitability of the processing and testing facilities<br />
and continuous improvement, as well as the appropriateness of the in-house transportation and<br />
warehousing for the specifics of the particular components (Formel Q-Fähigkeit, 2009). In this<br />
regard the responsibilities of the individual operators (process owners) must be clearly defined.<br />
Volkswagen regards it as especially important, that suppliers’ employees possess the necessary<br />
42
qualifications to perform the tasks assigned to them. This aspect is especially critical for the stable<br />
business operations in developing regions which are subject to high personnel turnover (see Section<br />
4.1). Personnel need to be regularly trained and should possess sufficient know-how about<br />
suppliers’ products, frequently occurring problems, as well as the impact of the process on the<br />
final quality. The facilities involved in the production should also be regularly controlled and their<br />
ability to achieve the required quality level of the final products should be continuously monitored<br />
and guaranteed. This is achieved through regular calibration of the machines and statistical control<br />
of the most important process parameters, e.g. temperature, pressure, moisture, time, speed, chemical<br />
composition, etc. Suppliers are also required to have the necessary testing equipment in place<br />
and make sure during the series production that the respective quality requirements are achieved.<br />
Moreover, material transportation between the individual processing stations on the work floor<br />
should be implemented in an ergonomic way and should guarantee a continuous material flow.<br />
Parts must be transported in suitable containers, which do not have negative impact on the quality<br />
of the products, i.e. prevent the transported goods from pollution and damage. To assure traceability<br />
each component has to be properly labeled. Volkswagen holds it for very important to have<br />
easily recognizable identification of components used for reference or calibration, defective parts<br />
as well as components, which need post processing. Such labeling is vital for preventing defective<br />
parts to reach the final customer and for the subsequent analysis of the capability of the overall<br />
production process. Suppliers are required to analyze the yield of the individual production steps<br />
and to define appropriate actions to minimize the scrap rates. This is an important step towards<br />
achieving continuous improvement of their overall process and accordingly lower prices of their<br />
final products.<br />
The third and final block of questions in the Volkswagen auditing questionnaire deals with<br />
customer care and customer satisfaction. Volkswagen expects from its suppliers to be responsive<br />
and react to customer complaints on short notice. In case of customer complaints the respective<br />
problems have to be analyzed as quickly as possible and corrective actions have to be defined and<br />
implemented accordingly. Suppliers have to define emergency plans for alternative ways of supply,<br />
production, and transportation, which assure the continuous delivery of high-quality components<br />
to their final customer even in the case of unexpected situations. These aspects are an essential<br />
precondition for a healthy customer-supplier business relationship which has to be sustained on<br />
the long term by any organization.<br />
During a quality audit a supplier has to provide evidence that it has considered and adequately<br />
implemented all customer requirements part of the audit questionnaire and thus assure Volkswagen<br />
43
Table 1: Evaluation points for a single question.<br />
(Source: Formel Q-Capability, 2009, p. 31)<br />
that its production processes are quality capable. Each of the questions in the individual blocks is<br />
graded on a scale from 0 to 10 points respectively (see Table 1), depending on the level to which<br />
the according production process fulfills the particular Volkswagen Group requirement. The first<br />
and third blocks of the auditing questionnaire (Sub-Contractors / Purchased Material and Customer<br />
Care / Customer Satisfaction (Service) respectively) refer to the overall production process, while<br />
the evaluation of the second block of questions (Production) is carried out separately for each<br />
individual processing step. The overall grade of each evaluation block is the percentage, which<br />
the sum of the points achieved in the individual questions represent from all possible points. For<br />
production processes which are comprised of several processing steps the evaluation result for the<br />
second block of the questionnaire (E PG ) is obtained as an average of the evaluation scores of the<br />
individual processing steps (E 1 , E 2 , . . ., E n ) (Formel Q-Fähigkeit, 2009):<br />
E PG [%] = E 1 + E 2 + . . . + E n<br />
[%] (1)<br />
n<br />
Here n is the number of processing steps. The process evaluation score of a quality audit is<br />
given in percent and is computed as an average of the evaluations obtained in the first (E Z ), second<br />
(E PG ), and third (E K ) questionnaire blocks (Formel Q-Fähigkeit, (2009)):<br />
E P [%] = E Z + E PG + E K<br />
[%] (2)<br />
3<br />
44
Apart from E P Volkswagen has defined four additional quality capability measures, calculated<br />
based on the evaluations of the questions in the block Production. These are namely: E U1 (Personnel<br />
/ Personnel Qualification), E U2 (Machinery / Equipment), E U3 (Transport / Parts Handling<br />
/ Storage / Packaging), and E U4 (Failure Analysis / Corrective Measures / Continuous Improvements).<br />
Along with the process evaluation described above a Volkswagen Group quality audit includes<br />
also a product-audit. The product audit assesses the degree of compliance of supplier’s final products<br />
with Volkswagen Group customer specific requirements and characteristics. Auditors measure<br />
on-site the key characteristics of ready-for-shipment products and any deviations from the<br />
customer requirements which are identified during the product audit have influence on the final<br />
rating of the respective supplier. Volkswagen has defined three major classifications for deviations<br />
detected during the product audit (Table 2).<br />
Each quality audit should provide a statement about the quality capability of a particular supplier.<br />
Due to the fact that this type of quality capability evaluations pursue a holistic approach<br />
and evaluate the entire production chain at a supplier’s location, it is reasonable to expect that this<br />
Table 2: Fault categorization relevant for a Volkswagen product audit.<br />
(Source: Formel Q-Capability, 2009, p. 27)<br />
45
quality management tool (if carried out properly) is able to detect the individual quality risks down<br />
the production chain and therefore give a good predictor for supplier’s quality.<br />
Volkswagen suppliers receive a letter-encoded quality rating – A, B, or C – depending on the<br />
results of their process and product evaluations. These ratings are then used in the corporate sourcing<br />
process as input for sourcing decisions. The general rules for supplier rating and the meanings<br />
of the individual rating categories are listed in Table 3. Quality rating is governed by the so-called<br />
hurdle principle (from the German Hürdenprinzip), however, and a supplier could receive a worse<br />
quality rating, even though its process evaluation score covers the basic requirements for a higher<br />
rating. A number of the requirements listed in the audit questionnaire are regarded by Volkswagen<br />
as especially critical for the overall quality capability of the production process and therefore, in<br />
case a supplier fails to adequately fulfill them, its overall quality rating is downgraded. This concept<br />
makes the evaluation procedure especially flexible in the cases of serious deviations from<br />
the quality requirements, as it gives the auditors the possibility to intervene and raise awareness<br />
about potential problems by downgrading the supplier production process. The according questions<br />
which evaluate the critical quality requirements are marked as *-questions (star-questions<br />
Table 3: Rules for supplier letter-encoded rating.<br />
(Source: Formel Q-Capability, 2009, p. 35)<br />
46
from the German Sternchenfragen). Thus for example, a supplier could get a B quality rating, even<br />
though its process evaluation score is greater than 91% (see Table 3). Common reasons for such<br />
type of downgrading are deficiencies identified during the process audit such as a missing certification<br />
of supplier’s quality management system according to the VDA 6.1 or ISO/TS 16949 quality<br />
standards, a question in the audit questionnaire evaluated with 0 points or a * question evaluated<br />
with 4 points, as well as an evaluation score for one of the individual components (E Z , E PG , or E K )<br />
smaller than 80%. Reason for downgrading a supplier’s quality rating from A to B could also be<br />
a B-type or a systematic C-type deviation identified during the product audit (Formel Q-Fähigkeit,<br />
2009). That is why there is difference between an A-rated supplier and a B-rated supplier with an<br />
A process evaluation score (E P ≥ 92%).<br />
For more serious deviations from the quality requirements a supplier’s quality rating would be<br />
downgraded to C despite a process quality evaluation score greater than 81%. Such are cases in<br />
which a supplier is not able to meet certain project deadlines and implement the defined improvement<br />
actions before start of series production (SOP). A C downgrading would follow also in cases<br />
in which one or more of the individual evaluation components (E Z , a single processing step E 1 ,<br />
E 2 , . . . , E n , E PG , E K , E U1 , E U2 , or E U3 ) are smaller than 70% or a *-question is evaluated with 0<br />
points (Formel Q-Fähigkeit, 2009). Furthermore, if during the product audit an A-type or a systematic<br />
B-type deviation is detected, the supplier will be C-downgraded, too. Mismatches between<br />
the quality evaluation score and the quality rating of downgraded suppliers (B with E P ≥ 92% or<br />
C with E P ≥ 82%) can influence negatively the correlation between quality capability and quality<br />
performance and were therefore regarded in the calculations in the later stages of the project.<br />
In the general case Volkswagen awards delivery contracts only to suppliers with an A or a B<br />
quality rating, as suppliers with an A rating are usually preferred (Figure 12). Suppliers with a<br />
C quality rating are not considered in the sourcing process. In the empirical data analyzed here,<br />
however, there are a few cases of C-rated suppliers which deliver to Volkswagen. Such cases result<br />
due to the fact that suppliers’ rating can be downgraded to C also post factum based on poor quality<br />
performance in the series delivery. Subsequently the status of the Volkswagen business relationship<br />
with such suppliers is set to ”business-on-hold”. This means that the latter can still deliver parts<br />
for the current projects, but will not be considered for any new projects unless they manage to<br />
improve their production process and receive a positive audit evaluation. A part of the contractual<br />
agreement between Volkswagen and its business partners requires from every Volkswagen Group<br />
47
supplier to achieve an A rating (from German ”Ziel-A Vereinbarung” or ”Target A Agreement”).<br />
B- and respectively C-rated suppliers will be therefore subsequently audited, until they achieve this<br />
target. Very often in practice due to limited auditing resources, however, A-rated and sometimes<br />
even B rated suppliers will not be audited for considerably long periods, as long as their products<br />
aren’t subject to any serious quality related problems. In light of the sampling discussion presented<br />
in Section 3 this is quite undesirable. Failing to follow the developments of the quality capability<br />
of the production processes could lead eventually to the untimely detection of serious production<br />
problems down the supply chain. On the other hand, with respect to the analysis presented in this<br />
paper – old evaluations, which do not correspond to the actual quality capability of the respective<br />
suppliers, pose a significant challenge. This is an especially important factor for the subsequent<br />
calculations and is therefore discussed once again in greater detail in Section 7.2.2 below.<br />
6.2 Quality Performance<br />
After Section 6.1 described the quality capability records, the current section introduces suppliers’<br />
quality performance information – the second important type of empirical data, which is necessary<br />
in order to perform the correlation studies suggested in Section 3 as a possible way to give an<br />
answer to the primary research question of this paper. Quality performance data describes supplier-<br />
Controlling<br />
Procurement<br />
Presenting<br />
Legal Entity<br />
Buyer<br />
Head of<br />
Procurement<br />
(CSC)<br />
Commodity<br />
Manager<br />
(VW Group<br />
Procurement)<br />
Quality<br />
Assurance<br />
(Supplier Audit)<br />
Logistics<br />
Volkswagen<br />
Corporate Sourcing<br />
Committee (CSC)<br />
collective decision<br />
Global<br />
Sourcing VW<br />
Forward<br />
Sourcing VW<br />
Research &<br />
Development<br />
In-house<br />
Production<br />
Local<br />
Purchasing<br />
Teams<br />
Procurement of<br />
the Brands<br />
Right to veto<br />
Figure 12: Participants in Volkswagen’s Corporate Sourcing Committee (CSC). An unsatisfactory<br />
supplier evaluation in a quality audit results into a veto in the CSC process.<br />
(Source: Volkswagen Internal Reports, 2012)<br />
48
induced problems originating in the production halls of Volkswagen during series production. The<br />
empirical quality performance records are organized according to a categorization, which will<br />
hereafter be referred to as material groups (from the German term Werkstoffgruppen). The material<br />
group classification is maintained by Volkswagen Procurement and is used not only for recording<br />
quality performance of the respective Volkswagen suppliers, but also in the sourcing process to<br />
classify information regarding their contracts with Volkswagen.<br />
Unlike the product group classification, the purpose of material groups is not to account for<br />
the specifics of parts’ manufacturing processes. Rather this type of classification is applicationoriented,<br />
i.e. material groups include products which have similar application, even though in<br />
certain cases they might be manufactured through significantly different production processes.<br />
This type of classification is particularly useful for the sourcing process. One good example for<br />
the differences between the two types of classifications is the material group of fuel tanks (material<br />
group – 0094). Components, part of this material group, have two major variants – synthetic<br />
material fuel tank assembly and metal fuel tank assembly (Figure 13). Despite the obvious differences<br />
between the two types of production processes involved in parts’ manufacturing (one deals<br />
with plastic components, and the other with metal components), the products are categorized in<br />
the same material group, because they are used for the same purpose – to store fuel. From the<br />
viewpoint of the product group classification, however, the two variants of fuel tanks fall in two<br />
completely different product groups – Assembly Blow Moulded Parts (Zsb. Kunststoffblasteile –<br />
2122) and Assembly Metal Parts (welded) (Zsb. Metallteile (geschweisst) – 1361) respectively.<br />
Audits of suppliers of fuel tanks are therefore carried out according to the latter two classifications<br />
(by auditors from different audit divisions), while their quality performance data is collected under<br />
the same material group – 0094.<br />
Product Group: 1361<br />
Assy. Metal Parts<br />
(welded)<br />
Product Group: 2122<br />
Assy. Blow Moulded<br />
Parts<br />
Material Group: 0094<br />
Fuel Tanks<br />
Product: Fuel Tank<br />
Material: Metal<br />
Product group: 1361<br />
Material group: 0094<br />
Product: Fuel Tank<br />
Material: Synthetic<br />
Product group: 2122<br />
Material group: 0094<br />
Figure 13: Products with the same application (fuel tank) are classified in the same material<br />
group. If the production processes of the respective products differ (metal welded part vs. blow<br />
moulded part), they are classified in different product groups.<br />
49
There are a total of four quality performance indicators available in the analyzed empirical<br />
data. These are namely ppm-values, production-line interferences/failures or also HSF (from the<br />
German Hallenstörfälle), direct quality performance (direkte Leistung) and indirect performance<br />
(indirekte Leistung). The ppm quality indicator (defective parts per million) was already mentioned<br />
above and is one of the most commonly used quality performance measures not only in the<br />
automotive industry but also in many other production industries. It describes deviations of the<br />
characteristics of the delivered components from the according technical specifications (physical,<br />
haptic, performance, etc. properties). This is probably the type of quality performance information<br />
which is most closely related to the manufacturing quality capability of a particular production<br />
process and is also (the only one) used in the analyses by Stroescu-Dabu (2008) and Hadzhiev<br />
(2009). The definition of ppm is as follows:<br />
ppm =<br />
number of delivered defective components<br />
total number of delivered components<br />
× 1.000.000 (3)<br />
HSF is another relevant quality performance metric. It is of more general character than ppm<br />
and describes incidents in which a supplier’s products or services have negative influence on the<br />
overall production process. HSF records result from deliveries of defective components (in which<br />
cases the respective supplier would receive ppm as well), but can also be induced by incidents<br />
which are not necessarily related to any defects of the delivered components. The latter situations<br />
would increase supplier’s HSF record, but would not generate any ppm. Examples of HSF, which<br />
do not result in ppm, are cases in which a supplier delivers the wrong type of components or the<br />
delivered components are not properly labeled. Such cases trigger sorting campaigns and disturb<br />
the continuity of Volkswagen’s internal processes. The time overhead which the processing of a<br />
particular supplier-caused problem incurs for Volkswagen is measured by suppliers’ direct quality<br />
performance in minutes. The direct performance of the suppliers together with precisely defined<br />
costs per unit time is then used by Volkswagen to calculate the total incurred costs of a particular<br />
problem. In case the respective problem has also a monetary dimension the according amount<br />
enters the records as the indirect performance of the supplier in EUR. This quality performance<br />
metric is probably not very suitable for measuring supplier performance, however, since it is dependent<br />
on the present price levels and its consistency is subject to variation over time influenced<br />
by varying prices, exchange rates and inflation. This supplier performance metric is used primarily<br />
for charge-back purposes.<br />
50
While ppm and HSF are two indicators which are consistently recorded in the empirical data,<br />
very few of the available supplier records contain information also about their direct and indirect<br />
performance (around 15% and 2% respectively of all available records for the time period considered<br />
in the analysis). The lack of such information is surprising, since for every single HSF the<br />
supplier receives recourse with the costs required by Volkswagen to process the according problem,<br />
i.e. for every HSF the supplier should also have a direct performance record in minutes. At<br />
this point, it is probably important to mention that the empirical data analyzed here was not obtained<br />
directly from the system, which is used to collect the respective performance information<br />
(KPM-Halle), rather from a secondary SAP-based database (Business Objects), which is used for<br />
evaluation purposes and provides the available information in a format which is convenient for<br />
analysis with statistical software. The reasons why these metrics are not consistently present in the<br />
second system probably lie in the fact that they are almost exclusively used for supplier recourse<br />
purposes, and are hardly used in internal Volkswagen quality reports for tracking the development<br />
of suppliers’ quality performance over time. The fragmented information available about suppliers’<br />
direct performance and indirect performance makes these two quality performance metrics<br />
unreliable for the purposes of this analysis and therefore they are excluded from the subsequent<br />
discussion.<br />
The process of generating the quality performance metrics used in this analysis is significantly<br />
more complex than the way quality capability scores are obtained. This is due to the large amount<br />
of individuals involved in the assessment of suppliers’ quality performance. Even though part<br />
of the delivered components is regularly sampled for defects through incoming inspection, a major<br />
portion of the defect detection at Volkswagen happens at its assembly lines. Practically any<br />
Volkswagen employee involved in the assembly process can detect and report a problem. The<br />
detected problem is documented and the affected batch of components is sent for analysis to the<br />
line rejects handling center (in German Regresszentrum) of the affected production facility. There<br />
the parts are sorted out and all components, which are not subject to the encountered problem,<br />
are then returned to the production line for assembly. This is also where the data collection takes<br />
place. Quality performance records enter Volkswagen’s database KPM-Halle. Every instance in<br />
KPM-Halle represents a single HSF and includes a detailed description of the reported problem,<br />
identification codes of the affected parts, the type and number of the affected components (not provided<br />
for problems unrelated to deviations from the technical specifications of the components),<br />
general identification information of the respective supplier, as well as the time overhead for recourse<br />
purposes.<br />
51
Currently, Volkswagen works on a new concept called ”Quality Filter for Purchased Components”<br />
(from German Warenfilter), which aims at improving the work flow in Volkswagen’s<br />
production facilities. A particular disadvantage of the new concept from the analytical point of<br />
view, however, is that neither ppm nor HSF will enter the quality records of suppliers included in<br />
the quality filter program. Quality filters are external warehouses, which are located in close proximity<br />
to Volkswagen’s production locations and whose function is to perform a 100% incoming<br />
inspection for components with a very large amount of defects. The idea behind the quality filter<br />
is that in the future suppliers which have poor quality performance in the series production will<br />
not be allowed to deliver their components directly to the respective Volkswagen plants. Instead<br />
100% of their deliveries will be first examined in the quality filters and only functional parts will<br />
be then transported to the Volkswagen plants. Suppliers will then have to deliver through the quality<br />
filter until their quality performance reaches satisfactory levels and they are allowed to deliver<br />
again directly to Volkswagen’s production facilities. The employment of quality filters will have a<br />
positive impact on the stability of Volkswagen’s production process and will reduce the number of<br />
line rejects, due to the fact that less defective components will reach the assembly lines.<br />
Even though the process for recording suppliers’ quality performance metrics is clearly structured,<br />
as already mentioned in the previous sections, the meaning of the individual quality performance<br />
indicators is variable and is influenced by the complexity of the delivered components as<br />
well as the according delivery amounts. If a certain problem is caused by a sub-supplier, for example,<br />
no ppm information is included in the quality performance record of the direct Volkswagen<br />
supplier. Rather the latter is held liable only for the incurred costs (the direct supplier receives a<br />
HSF nevertheless). For that reason suppliers of relatively complex components with several subsuppliers<br />
will usually have smaller amounts of ppm than suppliers of simpler parts. Further, in<br />
certain cases there is such a large variety of a particular type of assembly products that instead<br />
of using a separate part number for the individual component variants, Volkswagen describes the<br />
delivered units in its systems as an aggregation of the part numbers of the sub-components used in<br />
their assembly. Good examples for such types of components are axles and seats. Whenever there<br />
is a problem with such products, however, only the part numbers of the defective assembly subcomponents<br />
enter the quality performance record of the respective supplier. Thus the according<br />
ppm values are artificially reduced. For these reasons ppm is not always suitable for performance<br />
comparison especially between suppliers which deliver products with significantly different complexity.<br />
Such comparisons would not necessarily adequately depict the real situation and could<br />
potentially obscure important trends in the empirical data. In the analysis a possible workaround<br />
52
for this problem is to define classes of complexity and look at correlations only within the respective<br />
classes. A good starting point is to use the material group classification as basis for comparison<br />
and avoid comparing records, which are classified into different material groups.<br />
The specifics of Volkswagen ppm records described here are not considered in the analytical<br />
studies by Stroescu-Dabu (2008) and Hadzhiev (2009). Therefore, a potential bias of their results<br />
is not excluded, due to the heterogeneity of the actual meaning of the ppm quality performance<br />
indicator. This holds true especially for the more general, industry-based classification of empirical<br />
data. Such bias could even be a possible reason, which partially explains the unsatisfactory<br />
analytical results for the chemical and metal industries observed in these studies. This statement is<br />
supported by the results observed by Hadzhiev (2009). His analysis shows that for a more detailed<br />
supplier classification, which would be expected to increase the consistency between the individual<br />
ppm records within a certain analytical supplier group, particular supplier groups also from the<br />
metal industry such as suppliers of metal profiles indeed show better consistency between the actual<br />
quality performance and the evaluation scores of Volkswagen suppliers. At this point it is also<br />
important to mention, that assembly units from the electric industry such as navigation systems,<br />
radios, and control electronics, are described with their own part numbers. Therefore ppm records<br />
for electric suppliers are expected to be with relatively good consistency.<br />
The other quality performance metric in the empirical data available for this analysis is HSF.<br />
HSFs are of summarizing character and their definition is independent of the total number of<br />
rejected parts in a particular defective batch. Thus two independent production line rejects can<br />
concern significantly different amount of defective parts, but they will be equal in terms of HSF.<br />
Companies, which deliver components in large batches such as screws or nuts, normally would<br />
have a relatively low number of HSFs even in cases when their ppm records are high. On the other<br />
hand, companies which deliver parts in smaller batches, such as assembly units like seats, are more<br />
likely to have a relatively large number of HSFs, even if their ppm record remains low. In other<br />
words, if the same relative number of defective parts is delivered in several smaller batches the<br />
resulting number of HSFs will be significantly higher as compared to the case, when components<br />
are delivered in fewer, larger batches.<br />
One way to avoid the pitfalls posed by the diverse character of empirical data – both quality<br />
performance records and quality capability evaluation scores – is to split records into a number<br />
of analytical groups, which contain only suppliers with comparable quality indicators. Such classification<br />
should not merely take into account the more general industry-based classification of<br />
53
products (electrical, metal, and chemical parts), but has to also take into account the size of deliveries,<br />
the manufacturing processes involved, as well as the product complexity. Thus, the quality<br />
performance results of suppliers within a certain analytical group will be comparable to one another<br />
based on ppm or HSF accordingly as well as on suppliers’ quality evaluation scores. On the<br />
other hand, comparisons between the records of suppliers in different analytical groups have to be<br />
performed with caution due to the potential differences in the character of the according empirical<br />
data described above.<br />
6.3 Automation of the Data Processing<br />
In order to conduct the analysis according to such detailed criteria and reach any generalizable<br />
conclusions, it is particularly important to consider a representative data sample. Inevitably, given<br />
the large amounts of empirical data, a significant part of the data processing has to be performed<br />
in an automated manner. A significant obstacle for automation of the analysis turned out to be the<br />
structure of the internal Volkswagen databases, which are optimized for the needs of the individual<br />
departments (Quality Assurance and Procurement respectively), and therefore do not contain the<br />
according information in the format required for the current analysis. In most of the cases each<br />
Volkswagen department uses a separate system which is specifically tailored for its own needs and<br />
in many cases does not provide the data in a format suitable for analysis, other than the particular<br />
demands of the respective department. The fact that data required for the analysis presented here<br />
stem from several different systems is therefore related to a number of compatibility issues, which<br />
obstruct the automated data processing.<br />
The challenges faced during the project are expressed within the following. Suppliers’ quality<br />
capability information is stored in the database ISQAD maintained by the Volkswagen Quality<br />
Assurance department. The backbone of ISQAD is an Oracle database accessed via a browserbased<br />
user interface. The system provides access to the records of individual suppliers and all<br />
relevant documents (which is what is required in auditors’ daily routine), but the only summarizing<br />
information, which is accessible over the user interface, is a list with the latest audit evaluations<br />
of the suppliers – supplier catalogue (from the German Lieferantenkatalog). The list includes<br />
the overall supplier rating (A, B, or C) for each of the evaluated product groups, together with the<br />
overall quality evaluation scores (E P ) of the individual suppliers in percent. The supplier catalogue<br />
does not provide any information about previous audit results at a particular location, nor any<br />
details about the individual components of a particular evaluation such as the evaluated process<br />
54
steps as well as the resulting partial evaluations (E 1 , E 2 , . . ., E n ). In general such information is<br />
essential in providing an idea about the complexity of the evaluated processes as well as about the<br />
strengths and weaknesses of suppliers’ production.<br />
At this point it is important to mention, however, that such information is available after all,<br />
but is not systematized in a structured list, and is rather dispersed in the individual audit reports.<br />
Due to the specifics of the Volkswagen internal process, audit reports are saved in the system in<br />
the form of an attachment, which in most of the cases is a scanned copy of the original report,<br />
stored in paper form in Volkswagen’s archive . This form of storage is a particular obstacle for<br />
an automated data processing and results into a serious manual processing overhead. Due to the<br />
fact that a great deal of the information had to be extracted manually from the records, it was<br />
not possible to include in the analysis a substantial amount of the available more detailed quality<br />
evaluation data in a sufficiently large scale.<br />
Based on these experiences already here the first important conclusion from this work can be<br />
made. The supplier evaluation process at Volkswagen generates considerable amount of field information<br />
which is an especially valuable input for statistical analysis of Volkswagen’s supplier<br />
management process. Currently this information is stored in a form, which makes it particularly<br />
difficult to access for large generalizable studies (audit reports in paper form with scanned<br />
copies in the main electronic database). However, making the generated information available in<br />
an electronic form may be achieved relatively straight forward. A positive aspect of the problem<br />
is the fact that the original audit reports are created in an electronic form (Microsoft Excel forms),<br />
which means that ideally the relevant information can be very easily exported into the main Oracle<br />
database (ISQAD). The transfer can be completely automated and will require minimal amount<br />
of time. This improvement will increase the efficiency of the old procedure without compromising<br />
the quality of the available audit records. In addition to the currently available data, the new<br />
procedure would make the rest of important information in the audit reports easy to access and<br />
analyze.<br />
One further challenge is posed by the fact that even though quality performance data are accessible<br />
over an SAP software tool (Business Objects), used for quarterly and yearly reports, a<br />
substantial amount of automated post processing was required here as well. The information included<br />
in the analysis was processed using a Microsoft Excel-based VBA 9 software tool, developed<br />
by the author especially for this purpose.<br />
9 Visual Basic for Applications (a programming language)<br />
55
In addition to the data processing obstacles described above, the following two sections describe<br />
additional challenges and their possible solutions, which came up during the data analysis.<br />
6.3.1 Matching Quality Capability and Quality Performance<br />
In addition to the more trivial problems of data accessibility introduced above there is a more<br />
profound problem: the use of two different categorization types to organize quality capability<br />
(process-oriented product groups) and quality performance (application-oriented material groups)<br />
data. The two categorization types were already introduced earlier in this chapter. This particular<br />
difficulty is rather on the conceptual level than on the practical implementation of the supporting<br />
databases. Usually product groups are more generally defined than material groups, since a particular<br />
manufacturing process could be used to produce components with quite different applications<br />
- e.g. moulded plastic parts for interior and exterior. Therefore, in the general case a product group<br />
is expected to encompass components from several material groups, while a single material group<br />
would encompass products from a single product group. However, this is not always the case. The<br />
manufacturing of a particular component could include a number of very diverse processing steps.<br />
One particular example are electric components such as e.g. the material group 0043 – Exterior<br />
Lights (Außenleuchten), comprised of metal, plastic and electrical subcomponents part of several<br />
different product groups.<br />
Due to the different logic of the two categorizations there is no clearly defined relations matrix<br />
between them. In other words, in a great part of the cases one cannot instantly distinguish<br />
which one of the audit results for a certain supplier refers to the manufacturing process used to<br />
produce a particular type of components present on supplier’s quality performance record. The<br />
lack of product-group-material-group relation matrix makes it difficult to distinguish whether the<br />
processes used to manufacture two components from the same material group are similar or not<br />
and represents a significant challenge for the automation of the subsequent analysis. The only alternative<br />
to perform the desired correlations was to manually match quality performance and quality<br />
capability records, which due to the high processing overhead substantially limited the scale of the<br />
analysis.<br />
At this point it is necessary to mention that a tentative solution to the problem has already been<br />
implemented at Volkswagen and is currently being used to upload new audit data. However, due to<br />
the limited time window not enough empirical were data generated with the new procedure till the<br />
end of this project, which data could serve as a sufficient base for statistical analysis. Nevertheless,<br />
56
this improvement is of particular benefit for any future analyses on the topic.<br />
6.3.2 Supplier Identification<br />
The problem described in the previous section was partially mitigated by the help of the supplier<br />
identification systems in use at Volkswagen, which served as reference to narrow down the manual<br />
data queries. Volkswagen Group uses two code systems to identify its suppliers. These are namely<br />
the Volkswagen internal KRIAS identification system (from the German Kreditoren-Informationsund<br />
Abrechnungssystem or system for supplier identification and accounting) and Dun & Bradstreet’s<br />
international numbering system – D-U-N-S R○ (Data Universal Numbering System).<br />
The KRIAS identification system is primarily used for billing purposes and is managed internally<br />
by Volkswagen. KRIAS numbers are assigned to Volkswagen suppliers individually for every<br />
supplier production location. This means that if a given supplier manufactures its products at two<br />
separate manufacturing sites and delivers its production to the same Volkswagen Group receiving<br />
plant, each of supplier’s production facilities will be assigned with a distinct KRIAS number. The<br />
major disadvantage of KRIAS numbers, however, is the fact that they are assigned autonomously<br />
by the individual companies, members of the Volkswagen Group. Thus, the same supplier location<br />
(and respectively the same production process) is very often associated with several KRIAS<br />
numbers in the cases, in which the same supplier delivers components to different subsidiaries of<br />
the Volkswagen Group. This makes the KRIAS number not very suitable for record identification.<br />
Nevertheless, some of the supplier evaluation records are still associated to KRIAS numbers only,<br />
due to the fact that the according suppliers do not possess a D-U-N-S R○ number. Such records<br />
were excluded from the subsequent analysis.<br />
A more suitable alternative to the KRIAS number is the second identification system in use at<br />
Volkswagen – the D-U-N-S R○ . It was developed by Dun & Bradstreet (D&B) and first introduced<br />
in 1962 (Dun & Bradstreet, 2012). The D-U-N-S R○ number is a freely distributed number for<br />
unique global company identification and is currently used by more than 100 million companies<br />
worldwide (Dun & Bradstreet, 2012). The major advantage of the D-U-N-S R○ identification in<br />
comparison to KRIAS is that it is business unit specific, meaning that every single legal business<br />
unit has its own, unique D-U-N-S R○ number. Generally speaking a single D-U-N-S R○ number is<br />
usually associated with one or several KRIAS numbers. On the other hand, a KRIAS number is<br />
most commonly associated with a single D-U-N-S R○ number. The fact that a D-U-N-S R○ number<br />
provides unique business identification (a single number used worldwide), makes it more suitable<br />
57
than the KRIAS number for managing operational data. This is also the reason why for the parts<br />
of the analysis presented within the following sections the D-U-N-S R○ number was the preferred<br />
identification and was used to more easily determine which quality capability record is relevant for<br />
a particular quality performance entry and vice versa.<br />
At this point it is important to mention, however, that even though D-U-N-S R○ numbers offer<br />
better global identification, the use of the D-U-N-S R○ identification system is associated with<br />
one particular drawback. D-U-N-S R○ numbers, unlike KRIAS numbers, are not managed by<br />
Volkswagen itself, rather by the Dun & Bradstreet organization, making it difficult for Volkswagen<br />
to track changes of the D-U-N-S R○ record of a certain supplier location. This state of affairs poses<br />
serious challenges regarding information consistency within the individual Volkswagen databases.<br />
Despite this disadvantage, D-U-N-S R○ is still the preferred identification for the purposes of this<br />
analysis.<br />
7 Results and Discussion<br />
Section 6 introduced the different types of empirical data, which were subject to a number of<br />
statistical evaluations aiming to contribute to the answer of the research questions introduced in<br />
Section 2. After part of the technical aspects of working with the data were also covered in the<br />
previous section, now it is time to present the results of the conducted evaluations and discuss the<br />
insights gained from the observations. This is the aim of the current chapter.<br />
7.1 State and Quality of the Available Empirical Data<br />
The part of the project presented in this section provides a comprehensive assessment of the composition<br />
of the empirical data analyzed later in this paper. This step was motivated by the fact<br />
that the relevance for quality auditing of any insights gained from the subsequent data analysis are<br />
strongly influenced by how accurately and how thoroughly the recorded data describe the actual<br />
processes. Partial records do not allow to explore the process interactions to their full extent and<br />
therefore limit the depth of the analysis. This part of the analysis has also high practical value. The<br />
findings of this section can help improve Volkswagen’s data management. Objective operational<br />
data is essential for decision making and completeness of the records is especially important for the<br />
management. Any incomplete data records are of little use and are merely a source of additional<br />
58
expenses for the organization (costs to obtain and preserve the data, locked resources and time,<br />
financial consequences of misguided decisions etc.). Therefore it is important on the one hand to<br />
identify and eliminate such records and on the other to improve data collection.<br />
This section includes a set of cross-sectional as well as longitudinal evaluations, whose goal<br />
is to identify incomplete records as well as to assess the behavior of data quality over time and<br />
space. Due to the large overhead to compensate for any incomplete records (use other ways to<br />
get the missing information) and since this is not the main emphasis of this work, all incomplete<br />
records were excluded from the subsequent phases of this study. For quality capability data these<br />
are records, which lack e.g. a D-U-N-S R○ number or have an inconsistent quality capability score<br />
(e.g. an A-supplier with a fulfillment level E P < 90% or 92% respectively). On the other hand,<br />
incomplete quality performance records lack one or more important statistical quantities such as<br />
the amount of delivered components, the respective material group classification, or a D-U-N-S R○<br />
number.<br />
The result of the analysis of quality capability data is presented in Figure 14a. Each quality<br />
capability entry represents the quality audit evaluation of a single product group at a particular<br />
supplier production facility. Thus, a quality audit of a certain supplier, which manufactures components<br />
from two different product groups at the same production site, would generate two separate<br />
entries in the statistic – the according evaluations of each of the product groups at this location.<br />
The results of the evaluation show that 9% of the records have an incomplete quality capability<br />
evaluation score (Figure 14a). In these cases the percent score of the according records is subject<br />
to system input errors and is not entered correctly in the database. In most of the cases it is simply<br />
missing, while in the rest the evaluation percent value does not correspond to the quality capability<br />
rating (A or B) of the respective supplier (most probably a typo). Additional 8% of the quality<br />
capability records have no D-U-N-S R○ number associated to them. It is necessary to mention, however,<br />
that the supplier catalogue (used as basis for the statistics) includes quality capability records<br />
over a large time period and most of the identified incomplete records refer to suppliers, which<br />
are no longer active or have never been assigned a contract. Since such cases concern particularly<br />
old evaluations, which are not considered in suppliers’ rating, they also have little influence on<br />
the quality assurance process of Volkswagen. For the analysis presented in this paper it is only<br />
meaningful to concentrate on the records of active suppliers, therefore incomplete records of the<br />
former two types of suppliers are not of interest for this study and have been excluded.<br />
The number of evaluated quality performance records was significantly larger than the amount<br />
59
9%<br />
8%<br />
1,1% 1,3% 0,9%<br />
1,3%<br />
83%<br />
Records with incomplete QC score<br />
QC records without a DUNS-Nr.<br />
Complete QC records<br />
96%<br />
Missing Material Group Information<br />
Missing D&B DUNSâ<br />
Missing Delivery Quantity<br />
Operational Material<br />
Complete Records<br />
(a) Quality of the quality capability data.<br />
(b) Quality of the quality performance data.<br />
Figure 14: Quality of the available empirical data.<br />
Note that in the case of quality performance records the individual percentages sum up to more than 100% because<br />
part of the evaluated records miss more than one type of relevant information.<br />
of quality capability evaluations. The records were generated on a monthly basis and a single<br />
record includes information about components from a particular material group, which have been<br />
delivered from a single supplier production location to a single Volkswagen plant. Slightly above<br />
1% of all records contain incomplete material group information (see Figure 14b). 1.3% of all<br />
quality performance records do not contain D-U-N-S R○ information, and additional 0.8% lack information<br />
about the number of delivered components. 1.3% of the records contain information<br />
about operational material (from German Betriebsmaterial), which is not relevant for the analysis.<br />
The latter are supplies, which are not used in the Volkswagen Group production processes (do not<br />
end up in the final products), and are auxiliary materials such as protective wear for the employees,<br />
cleaning supplies, etc. All incomplete quality perfomance records were excluded from the<br />
analysis, presented in the subsequent chapters.<br />
To narrow down the sources of particular data non-integrities, the quality performance information<br />
was divided into several groups according to the location of the Volkswagen production<br />
plants, which reported the data. This step was motivated by the discussion on the trends of the<br />
automotive industry in the previous sections. Automotive production is spread over a number<br />
of regions which differ substantially in their manufacturing history and level of experience with<br />
60
20%<br />
9%<br />
57%<br />
7% 2,4%<br />
2,2%<br />
1,9%<br />
1,3%<br />
0,4%<br />
0,1%<br />
Germany<br />
Spain<br />
England<br />
Mexico<br />
India<br />
Czech Republic<br />
Brazil<br />
Belgium<br />
Argentina<br />
Russia<br />
Figure 15: Amount of quality capability records, reported by the individual Volkswagen production<br />
facilities sorted by region. The presented numbers are monthly averages over the period January,<br />
2008 – December, 2010.<br />
modern production management methods. It is therefore anticipated that historical and cultural<br />
differences influence quality awareness in these regions. As consequence this leads to potentially<br />
different perception of the importance of operational data and ultimately different data quality.<br />
Volkswagen Group has concentrated its production capacities into ten different regions – Germany<br />
(a total of 33 independent production units are located in this region), Argentina (2), Belgium (1),<br />
Brazil (4), Czech Republic (8), England (1), Mexico (1), Russia (1), India (1), and Spain (5). Other<br />
Volkswagen Group production facilities such as the ones in USA and China are not included in this<br />
analysis either due to the fact that they were not active throughout the evaluated time period or because<br />
the relevant performance information from the respective regions was not accessible through<br />
Volkswagen’s databases, which were available for this analysis.<br />
The amount of quality performance records generated by the individual regions is significantly<br />
different (see Figure 15), provided the uneven distribution of production capacities. As expected<br />
Germany contributed the largest number of records – comprising more than 56% of all records.<br />
Other regions, which generated particularly large amounts of quality performance information,<br />
are the Czech Republic (20% of all records) and Spain (9%). The available empirical data shows<br />
regional variations not only of the amount of incomplete records, but also of the proportion of<br />
factors, which influence data integrity (see Figure 16). However, the overall integrity of quality<br />
performance data, generated in the three regions 10 mentioned above, is exceptionally good and the<br />
10 These are also among the regions, in which Volkswagen has been present the longest and has established stable<br />
business processes. On the other hand, regions, which show certain deficits in data’s integrity, are mostly among the<br />
regions, in which Volkswagen has relatively small amount of experience.<br />
61
1,1% 1,4% 0,9% 1,3%<br />
0,4% 0,3% 0,2% 0,0%<br />
0,7% 0,8%<br />
0,4% 0,2%<br />
0,3% 1,1% 0,6% 0,0%<br />
0,5% 0,1% 0,9% 2,2%<br />
3,0%<br />
2,0%<br />
1,9%<br />
0,1%<br />
99%<br />
Region 2 Region 4 Region 7 Region 6 Region 8<br />
3,4% 4,0%1,0%<br />
0,1%<br />
98%<br />
0,0% 1,8% 6,3% 0,0%<br />
98%<br />
14%<br />
5,3%<br />
1,5%<br />
0,0%<br />
97%<br />
0,3%<br />
32%<br />
94%<br />
6,8%<br />
38%<br />
96%<br />
Worldwide<br />
0,2%<br />
92%<br />
92%<br />
82%<br />
68%<br />
0,0%<br />
56%<br />
0,2%<br />
0,2%<br />
Region 3 Region 9 Region 5 Region 10 Region 1<br />
Missing Material Group Info Missing Delivery Quantity Complete Records<br />
Missing DUNS Number<br />
Operational Material<br />
Figure 16: Quality of the supplier performance data reported by the individual Volkswagen production<br />
facilities. The data are sorted according to region. The numbers under each pie chart<br />
represent the average monthly amount of records reported by the according region over the period<br />
January, 2008 – December, 2010. Note that there are cases in which the individual percentages<br />
sum up to more than 100%. This is due to the fact that part of the evaluated records are missing<br />
more than one type of relevant information. The percentages for the amount of complete records<br />
in all cases are exact and can be used to deduce the overall amount of incomplete records.<br />
total amount of incomplete records remain below the overall average of 4.3%. This fact is very<br />
positive considering that recods coming from these regions comprise more than 85% of all monthly<br />
records. The observed differences between the quality of data records in the different regions show<br />
that splitting the records into regional datasets was indeed very meaningful.<br />
Note that not only the amount, but also the nature of data inconsistencies in the individual<br />
regions varies significantly. Thus for example, while almost 90% of all incomplete records in<br />
Region 1 are due to missing D-U-N-S R○ supplier identification, D-U-N-S R○ related data fragmentation<br />
is responsible for just below 30% of all problematic cases in Region 5. By contrast, almost<br />
80% of all incomplete records from this region are subject to incomplete material group information<br />
(some of the records are subject to both). In Region 6 60% of the incomplete data entries are<br />
subject to missing information about the amounts of delivered components. Significant variations<br />
in the extent to which certain factors influence data consistency, were identified not only on the regional,<br />
but also on the level of individual production units. On the one hand, such information can<br />
be quite useful to address the sources of differences and accordingly improve the coherence of the<br />
data collection process at different Volkswagen locations. This will ultimately improve the overall<br />
data consistency in the databases. On the other hand, the influence of the incomplete records on<br />
the subsequent steps of the analysis is limited only to reducing the available base of empirical data,<br />
since as already mentioned previously they are excluded from the calculations in the following<br />
62
sections. However, this is not particularly critical for the overall analysis since they represent less<br />
than 5% of the total available records (Figure 16).<br />
The numbers in Figure 16 are an average over the entire three-year time period covered by<br />
the analysis (from 2008 to 2010) and provide important information about the proportion of the<br />
major sources of data inconsistencies. However, it is equally important to also assess whether<br />
the influence of these factors on data consistency has diminished or increased over the according<br />
time period. Figures 17 through 21 present the longitudinal development of data quality in the<br />
individual regions. The graphs include the regional quality performance records on a monthly time<br />
scale and identify trends in the development of their quality.<br />
Worldwide<br />
Amount of Incomplete Records [%]<br />
10,00%<br />
9,00%<br />
8,00%<br />
7,00%<br />
6,00%<br />
5,00%<br />
4,00%<br />
3,00%<br />
2,00%<br />
1,00%<br />
Missing Material Group Information [%] Missing DUNS [%] Missing Delivery Quantity [%]<br />
0,00%<br />
2008 2009 2010<br />
Figure 17: Development over time of the quality of quality performance records, reported by all<br />
Volkswagen production facilities worldwide.<br />
63
Region 1<br />
60,00%<br />
Missing Material Group Information [%] Missing DUNS [%] Missing Delivery Quantity [%]<br />
Amount of Incomplete Records [%]<br />
40,00%<br />
20,00%<br />
0,00%<br />
2008 2009 2010<br />
Figure 18: Development over time of the quality of quality performance records, reported by<br />
Volkswagen’s production facilities in Region 1.<br />
Region 2<br />
Amount of Incomplete Records [%]<br />
10,00%<br />
9,00%<br />
8,00%<br />
7,00%<br />
6,00%<br />
5,00%<br />
4,00%<br />
3,00%<br />
2,00%<br />
1,00%<br />
Missing Material Group Information [%] Missing DUNS [%] Missing Delivery Quantity [%]<br />
0,00%<br />
2008 2009 2010<br />
Figure 19: Development over time of the quality of quality performance records, reported by<br />
Volkswagen’s production facilities in Region 2.<br />
64
Region 3<br />
Amount of Incomplete Records [%]<br />
10,00%<br />
9,00%<br />
8,00%<br />
7,00%<br />
6,00%<br />
5,00%<br />
4,00%<br />
3,00%<br />
2,00%<br />
1,00%<br />
Missing Material Group Information [%] Missing DUNS [%] Missing Delivery Quantity [%]<br />
Data point: 100%<br />
0,00%<br />
2008 2009 2010<br />
Figure 20: Development over time of the quality of quality performance records, reported by<br />
Volkswagen’s production facilities in Region 3.<br />
Region 4<br />
Amount of Incomplete Records [%]<br />
10,00%<br />
9,00%<br />
8,00%<br />
7,00%<br />
6,00%<br />
5,00%<br />
4,00%<br />
3,00%<br />
2,00%<br />
1,00%<br />
Missing Material Group Information [%] Missing DUNS [%] Missing Delivery Quantity [%]<br />
0,00%<br />
2008 2009 2010<br />
Figure 21: Development over time of the quality of quality performance records, reported by<br />
Volkswagen’s production facilities in the Region 4.<br />
65
Region 5<br />
30,00%<br />
Missing Material Group Information [%] Missing DUNS [%] Missing Delivery Quantity [%]<br />
Amount of Incomplete Records [%]<br />
20,00%<br />
10,00%<br />
0,00%<br />
2008 2009 2010<br />
Figure 22: Development over time of the quality of quality performance records, reported by<br />
Volkswagen’s production facilities in Region 5.<br />
Region 6<br />
Amount of Incomplete Records [%]<br />
10,00%<br />
9,00%<br />
8,00%<br />
7,00%<br />
6,00%<br />
5,00%<br />
4,00%<br />
3,00%<br />
2,00%<br />
1,00%<br />
Missing Material Group Information [%] Missing DUNS [%] Missing Delivery Quantity [%]<br />
0,00%<br />
2008 2009 2010<br />
Figure 23: Development over time of the quality of quality performance records, reported by<br />
Volkswagen’s production facilities in Region 6.<br />
66
Region 7<br />
10,00%<br />
Missing Material Group Information [%] Missing DUNS [%] Missing Delivery Quantity [%]<br />
Amount of Incomplete Records [%]<br />
9,00%<br />
8,00%<br />
7,00%<br />
6,00%<br />
5,00%<br />
4,00%<br />
3,00%<br />
2,00%<br />
1,00%<br />
0,00%<br />
2008 2009 2010<br />
Figure 24: Development over time of the quality of quality performance records, reported by<br />
Volkswagen’s production facilities in Region 7.<br />
67
Region 8<br />
10,00%<br />
Missing Material Group Information [%] Missing DUNS [%] Missing Delivery Quantity [%]<br />
Amount of Incomplete Records [%]<br />
9,00%<br />
8,00%<br />
7,00%<br />
6,00%<br />
5,00%<br />
4,00%<br />
3,00%<br />
2,00%<br />
1,00%<br />
0,00%<br />
2008 2009 2010<br />
Figure 25: Development over time of the quality of quality performance records, reported by<br />
Volkswagen’s production facilities in Region 8.<br />
The number of incomplete records with missing D-U-N-S R○ identification has increased slightly<br />
on a global scale during the first two years included in the evaluation – namely 2008 and 2009,<br />
while records with missing information about the quantity of delivered products or their material<br />
group classification have decreased over the same time period (Figure 17). The consistency of<br />
quality performance records marks an improvement during 2010. However, the overall result of<br />
the analysis shows that the cummulative quality performance dataset has a stable amount of affected<br />
records, which even slightly decreases over time. The total amount of complete records is<br />
particularly high – above 95%.<br />
On the regional level, the analysis shows that records from the plants in Region 6 and Region<br />
2 exhibit particularly stable data consistency with low share of incomplete records (see Figures<br />
23 and 19). Excellent examples for data quality improvement are the records from Volkswagen<br />
Group’s plants in Region 8 and Region 5. The plant in Region 8 managed to reduce the amount of<br />
records with missing D-U-N-S R○ information more than 10 times over a period of only three years<br />
(Figure 25). Similarly the production location in Region 5, managed to handle data inconsistency<br />
problems and drastically reduced the number of incomplete records – both for records with missing<br />
material group information as well as for records, which lack a D-U-N-S R○ number (Figure 22).<br />
Other regions also show positive development of data quality, even though not with the same pace.<br />
68
Thus for example, in Region 1 (Figure 18) the absolute number of records with missing D-U-N-S R○<br />
number show an overall improvement after a slight increase in 2008.<br />
From this several important conclusions can be derived both for quality management and the<br />
research questions of this study. On the one hand, benchmarking locations with positive development,<br />
such as Region 5 and Region 8, and applying their approaches for reducing the amount<br />
of incomplete records to other production locations, will improve the consistency of quality performance<br />
information in Volkswagen’s databases. As a result a number of Volkswagen internal<br />
processes, which need this information as input, will benefit from this improvement. Additionally,<br />
frequent monitoring of the quality of available data would also allow for the timely detection of<br />
potential negative trends. On the other hand, these observations show that the operational information,<br />
used to support Volkswagen’s business management process, with small exceptions shows<br />
good overall consistency especially given the size of the organization. Data management is a particularly<br />
critical issue and any organization has to take the due diligence to avoid data inconsistencies<br />
– most of all large international organizations.<br />
Generally speaking, integrity of the operational data is of extreme importance, not only for the<br />
purposes of the subsequent analysis, but also for the overall production process of Volkswagen<br />
Group (as well as for the quality management system of any other organization). Conclusions<br />
based on a reliable set of production records have a particularly important role in closing the PDCA<br />
(Plan-Do-Check-Act) feedback cycle, which is vital for the continuous improvement of business<br />
processes (ISO 9001, 2008, p. 9).<br />
69
7.2 Possible Sources of Data Bias<br />
Given the results on data quality, presented in the previous section, the next step is to analyze for<br />
potential sources of bias resulting from the found facts or other more general factors. This is the<br />
subject of Section 7.2. The following several sections discuss in detail a number of biasing factors<br />
and present possible approaches to account for and mitigate their influences.<br />
7.2.1 Revisions of Volkswagen’s Quality Auditing Process<br />
The first of the general influencing factors stems from the changes the auditing process has undergone<br />
over time. Quality auditing is not a new concept to Volkswagen (and other midsize and large<br />
companies) – the company conducted its first supplier quality evaluations already in the 1980s.<br />
Volkswagen’s quality audits were initially performed according to a number of process-specific<br />
lists of quality requirements, which later on were summarized into a single standardized quality<br />
auditing checklist 11 . The latter served as a basis for the first edition of Formel Q-Fähigkeit, which<br />
was released in 1991. Over the years the concept of quality auditing evolved and Volkswagen’s<br />
quality auditing policy was adjusted accordingly. There have been a total of seven editions of the<br />
document until now, as the most recent was issued in January, 2012 12 .<br />
All changes introduced by the subsequent editions of Formel Q-Fähigkeit aimed at optimizing<br />
Volkswagen’s auditing process and providing more reliable supplier quality capability evaluations.<br />
For example, one especially important change in the past is raising the requirements for a positive<br />
quality evaluation. The minimum quality capability fulfillment level for an A-rating was increased<br />
from 90% to 92% with the release of the fifth edition of Formel Q-Fähigkeit in January, 2005.<br />
The minimum requirement for a B-rating was also adjusted in the same edition and changed accordingly<br />
from 80% to 82%. Another example is the hurdle principle, which was introduced in<br />
the sixth edition of the document, released in August, 2009. Such modifications of the auditing<br />
requirements and respectively the auditing process can result into significant differences of the<br />
meaning of suppliers’ quality capability scores. However, comparisons between datasets which<br />
contain such differences would be undesirable. Therefore, an important part of the current analysis<br />
includes a number of comparisons between the distributions of quality capability scores performed<br />
according to the different editions of Formel Q-Fähigkeit. The purpose of these comparisons is<br />
11 Source: personal communication with Volkswagen AG employees.<br />
12 Empirical data included in this analysis does not include any supplier evaluations performed according to the<br />
requirements listed in Formel Q-Fähigkeit 7.0.<br />
70
to evaluate the extent to which changes introduced with the document revisions affect empirical<br />
data. Any significant variations in the distribution of supplier rankings suggest important differences<br />
and consequently in such cases it is meaningful to analyze quality capability data from the<br />
different periods separately.<br />
To determine what kind of tests are suitable for this part of the analysis – parametric or nonparametric<br />
statistical tests – it is first important to determine whether the analyzed data has a<br />
normal distribution or not. An important assumption of the parametric statistical tests is that the<br />
analyzed data is normally distributed. Thus, if the data are not normally distributed parametric tests<br />
will not be applicable and it will be necessary to use non-parametric distribution comparison tests.<br />
There are a number of methods to test for normality. Among those are the use of the one-sample<br />
Kolmogorov-Smirnov goodness of fit test performed with respect to the theoretical distribution of<br />
a normally distributed random variable (σ = 1 and µ = 0), the Liliefors test for normality, and the<br />
Shapiro-Wilk test for normality (Marques de Sá, 2007). This analysis employs the Shapiro-Wilk<br />
test since it is considered to be better for small sample sizes as compared to the two other tests<br />
mentioned here (Marques de Sá, 2007). The Shapiro-Wilk test ”is based on the observed distance<br />
between symmetrically positioned data values” (Marques de Sá, 2007, p. 187) and its statistic is<br />
defined as (Marques de Sá, 2007, p. 187):<br />
W =<br />
[ k∑<br />
i=1<br />
a i (x n−i+1 − x i )] 2<br />
/<br />
⎧<br />
n∑<br />
⎨<br />
(x i − x) 2 with k =<br />
⎩<br />
i=1<br />
n + 1<br />
2<br />
, if k is odd<br />
n<br />
2 , if k is even (4)<br />
The coefficients a i in (4) and the critical values of W can be obtained from table look-up<br />
(Marques de Sá, 2007). However, the normality test results presented here were obtained using<br />
the statistical software R 13 and no look-up was required. The R distribution comes with a ready<br />
implementation of the Shapiro-Wilk test (shapiro.test()). The R implementation of the test<br />
uses the null-hypothesis that sample data is normally distributed. In addition to the test statistic the<br />
software provides a p-value, which can be used to determine whether the null-hypothesis can be<br />
rejected at a particular significance level α, i.e. p < α.<br />
The available quality capability information was split into a number of subsets. Due to the<br />
great diversity of production processes the first criterion used to divide empirical data was sector<br />
of operation of the individual suppliers. Accordingly, three major sets of data resulted – quality<br />
13 All test results obtained in R and presented in this paper were generated using the software version 2.13.0 from<br />
13 th April 2011.<br />
71
evaluation scores of suppliers operating in the chemical, metal, and electrical industries accordingly.<br />
The resulting three sets of data were further divided based on which edition of the Formel<br />
Q-Fähigkeit was used as reference at the time the respective supplier evaluations were conducted.<br />
As a result, each of the three major evaluation score datasets was divided further into four subsets<br />
corresponding to evaluations performed according to Formel Q-Fähigkeit 3.0 (covering the time<br />
period between January 1997 and March 2000), Formel Q-Fähigkeit 4.0 (April 2000 – December<br />
2004), Formel Q-Fähigkeit 5.0 (January 2005 – July 2009), and Formel Q-Fähigkeit 6.0 (August<br />
2009 – December 2011).<br />
A Shapiro-Wilk normality test was performed on each of the resulting twelve groups of evaluation<br />
scores. Additionally, in order to visually ”double-check” the results of the normality tests for<br />
each dataset a Q-Q plot was generated. On the y-axis of each Q-Q plot are plotted the normalized<br />
quantiles of the given sample (µ = 0 and σ = 1), while on the x-axis are plotted the theoretical<br />
quantiles of a normal distribution with zero-mean and standard deviation of 1 (µ = 0 and σ = 1).<br />
If a particular sample is normally distributed, the according data points in the Q-Q plot would lie<br />
on the line y = x. Any systematic deviations from this line indicate that the data sample is not<br />
distributed normally (Marques de Sá, 2007). In each of the following Q-Q plots the line y = x<br />
was added (plotted in red). The results of the normality test on each of the twelve initial evaluation<br />
score samples are presented together with the respective Q-Q plots in Figures 26 through 28<br />
below. Each plot includes the Shapiro-Wilk test statistic W and the according p-value along with<br />
the sample size n.<br />
72
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ● ●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
● ● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
● ●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
−3 −2 −1 0 1 2 3<br />
−8 −6 −4 −2 0<br />
Normal Q−Q Plot for TOTAL_Chemie_FQF3<br />
Theoretical Quantiles<br />
Normalized Sample Quantiles ( σ = 1; µ = 0)<br />
Shapiro−Wilk Test<br />
W = 0.7665<br />
p = 4.6244e−32<br />
n = 793<br />
(a) Evaluations performed according to FQF 3.0.<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
−3 −2 −1 0 1 2 3<br />
−6 −4 −2 0 2<br />
Normal Q−Q Plot for TOTAL_Chemie_FQF4<br />
Theoretical Quantiles<br />
Normalized Sample Quantiles ( σ = 1; µ = 0)<br />
Shapiro−Wilk Test<br />
W = 0.8535<br />
p = 9.3982e−30<br />
n = 1022<br />
(b) Evaluations performed according to FQF 4.0.<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
−3 −2 −1 0 1 2 3<br />
−6 −4 −2 0 2<br />
Normal Q−Q Plot for TOTAL_Chemie_FQF5<br />
Theoretical Quantiles<br />
Normalized Sample Quantiles ( σ = 1; µ = 0)<br />
Shapiro−Wilk Test<br />
W = 0.9026<br />
p = 3.7678e−30<br />
n = 1531<br />
(c) Evaluations performed according to FQF 5.0.<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
−3 −2 −1 0 1 2 3<br />
−6 −4 −2 0 2<br />
Normal Q−Q Plot for TOTAL_Chemie_FQF6<br />
Theoretical Quantiles<br />
Normalized Sample Quantiles ( σ = 1; µ = 0)<br />
Shapiro−Wilk Test<br />
W = 0.8953<br />
p = 1.6998e−30<br />
n = 1469<br />
(d) Evaluations performed according to FQF 6.0.<br />
Figure 26: Shapiro-Wilk normality test on evaluation scores of chemical part suppliers. The samples<br />
include quality auditing data from all geographic regions.<br />
73
●<br />
Normal Q−Q Plot for TOTAL_Elektrik_FQF3<br />
Normal Q−Q Plot for TOTAL_Elektrik_FQF4<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●●●●<br />
●<br />
Normalized Sample Quantiles ( σ = 1; µ = 0)<br />
−5 −4 −3 −2 −1 0 1<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●●<br />
●●●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●●●<br />
●<br />
●●<br />
Shapiro−Wilk Test<br />
W = 0.8226<br />
p = 8.8423e−16<br />
n = 238<br />
Normalized Sample Quantiles ( σ = 1; µ = 0)<br />
−6 −4 −2 0<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
Shapiro−Wilk Test<br />
W = 0.8785<br />
p = 2.2182e−18<br />
n = 453<br />
−3 −2 −1 0 1 2 3<br />
Theoretical Quantiles<br />
(a) Evaluations performed according to FQF 3.0.<br />
−3 −2 −1 0 1 2 3<br />
Theoretical Quantiles<br />
(b) Evaluations performed according to FQF 4.0.<br />
Normal Q−Q Plot for TOTAL_Elektrik_FQF5<br />
Normal Q−Q Plot for TOTAL_Elektrik_FQF6<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●●<br />
●<br />
Normalized Sample Quantiles ( σ = 1; µ = 0)<br />
−4 −2 0<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●●●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
Shapiro−Wilk Test<br />
W = 0.9116<br />
p = 2.9814e−19<br />
n = 667<br />
Normalized Sample Quantiles ( σ = 1; µ = 0)<br />
−6 −4 −2 0<br />
●<br />
●<br />
●<br />
●●<br />
●<br />
●<br />
●●●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
Shapiro−Wilk Test<br />
W = 0.8523<br />
p = 2.274e−20<br />
n = 460<br />
−3 −2 −1 0 1 2 3<br />
Theoretical Quantiles<br />
(c) Evaluations performed according to FQF 5.0.<br />
−3 −2 −1 0 1 2 3<br />
Theoretical Quantiles<br />
(d) Evaluations performed according to FQF 6.0.<br />
Figure 27: Shapiro-Wilk normality test on evaluation scores of electrical part suppliers. The<br />
samples include quality auditing data from all geographic regions.<br />
74
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
−3 −2 −1 0 1 2 3<br />
−8 −6 −4 −2 0 2<br />
Normal Q−Q Plot for TOTAL_Metal_FQF3<br />
Theoretical Quantiles<br />
Normalized Sample Quantiles ( σ = 1; µ = 0)<br />
Shapiro−Wilk Test<br />
W = 0.8511<br />
p = 6.6782e−36<br />
n = 1551<br />
(a) Evaluations performed according to FQF 3.0.<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
−3 −2 −1 0 1 2 3<br />
−6 −4 −2 0 2<br />
Normal Q−Q Plot for TOTAL_Metal_FQF4<br />
Theoretical Quantiles<br />
Normalized Sample Quantiles ( σ = 1; µ = 0)<br />
Shapiro−Wilk Test<br />
W = 0.8829<br />
p = 2.2353e−34<br />
n = 1743<br />
(b) Evaluations performed according to FQF 4.0.<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ● ●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
● ●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
● ●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ● ● ● ●<br />
● ● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ● ●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ● ●<br />
●<br />
●<br />
●<br />
● ● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
−3 −2 −1 0 1 2 3<br />
−8 −6 −4 −2 0<br />
Normal Q−Q Plot for TOTAL_Metal_FQF5<br />
Theoretical Quantiles<br />
Normalized Sample Quantiles ( σ = 1; µ = 0)<br />
Shapiro−Wilk Test<br />
W = 0.8376<br />
p = 4.1559e−46<br />
n = 2682<br />
(c) Evaluations performed according to FQF 5.0.<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
−3 −2 −1 0 1 2 3<br />
−6 −4 −2 0 2<br />
Normal Q−Q Plot for TOTAL_Metal_FQF6<br />
Theoretical Quantiles<br />
Normalized Sample Quantiles ( σ = 1; µ = 0)<br />
Shapiro−Wilk Test<br />
W = 0.846<br />
p = 8.5925e−40<br />
n = 1932<br />
(d) Evaluations performed according to FQF 6.0.<br />
Figure 28: Shapiro-Wilk normality test on evaluation scores of metal part suppliers. The samples<br />
include quality auditing data from all geographic regions.<br />
75
The performed normality test gives strong statistical evidence that the respective distributions<br />
of the quality capability scores are not normally distributed. In all cases the p-value of each set<br />
of evaluation scores is substantially smaller than the 5%-significance level. The Shapiro-Wilk test<br />
results are also in accordance with the corresponding Q-Q plots, which show that the distribution<br />
of each group of points consistently deviate from the straight line representing y = x. Given<br />
these results, a parametric statistical test would not be suitable to compare the distributions of the<br />
individual sample groups. Alternatively, a two-sample Kolmogorov-Smirnov test was used for the<br />
comparisons.<br />
The two-sample Kolmogorov-Smirnov test ” is used to assess whether two independent samples<br />
were drawn from the same population or from populations with the same distribution” (Marques<br />
de Sá, 2007, p. 201). The test statistic measures the maximal deviation between the cumulative<br />
sample distributions of the two sets of data and is given by (Marques de Sá, 2007, p. 201):<br />
D m,n = max |S n (x) − S m (x)| (5)<br />
where S n (x) and S m (x) are the cumulative sample distributions of the two samples with sizes<br />
n and m respectively. Even for small samples the Kolmogorov-Smirnov test has a high powerefficiency<br />
14 of about 95% when compared to its parametric counterpart – the t-test (Marques de<br />
Sá, 2007, p. 201). The two-sample Kolmogorov-Smirnov test is implemented in R and can be<br />
executed using the function ks.test(). The null-hypothesis of the test is that the two samples<br />
have the same distributions. Each run of the Kolmogorov-Smirnov test in R provides the test<br />
statistic D along with the corresponding p-value, which is used to accept or alternatively reject<br />
the null-hypothesis at a particular significance level. In addition to the test statistic D, its p-value,<br />
and the sizes of the two compared samples n 1 and n 2 , each Kolmogorov-Smirnov test performed<br />
in this analysis is presented along with three plots. The first two plots present the distributions of<br />
each of the two samples, while the third plot superimposes their cumulative distribution functions.<br />
These graphs are particularly useful to understand the differences between the two samples and<br />
to identify regions in which they are similar. Figure 29 presents the results of the two-sample<br />
Kolmogorov-Smirnov test on evaluation score samples of chemical-part suppliers. The results for<br />
metal-part and electrical-part suppliers are presented in Figures 30 and 31 accordingly.<br />
14 The power efficiency of a non-parametric test η BA is the ratio between the sample size n A needed by the parametric<br />
test A and the sample size n B needed by its non-parametric counterpart B to achieve the same power at the<br />
same significance level, i.e. η BA = n A<br />
n B<br />
(Marques de Sá, 2007, p. 171).<br />
76
Frequency<br />
0 50 100 150<br />
TOTAL_Chemie_FQF3<br />
TOTAL_Chemie_FQF4<br />
TOTAL_Chemie_FQF5<br />
TOTAL_Chemie_FQF6<br />
TOTAL_Chemie_FQF3<br />
TOTAL_Chemie_FQF4<br />
TOTAL_Chemie_FQF5<br />
TOTAL_Chemie_FQF6<br />
Distribution of TOTAL_Chemie_FQF3<br />
30 40 50 60 70 80 90 100<br />
Frequency<br />
0 20 40 60 80 100<br />
Distribution of TOTAL_Chemie_FQF4<br />
40 50 60 70 80 90 100<br />
Frequency<br />
0 20 40 60 80 100<br />
Distribution of TOTAL_Chemie_FQF4<br />
30 40 50 60 70 80 90 100<br />
Frequency<br />
0 40 80 120<br />
Distribution of TOTAL_Chemie_FQF5<br />
40 50 60 70 80 90 100<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for TOTAL_Chemie_FQF3<br />
and TOTAL_Chemie_FQF4<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.15<br />
p = 3.72e−09<br />
n_1 = 793<br />
n_2 = 1022<br />
TOTAL_Chemie_FQF3<br />
TOTAL_Chemie_FQF4<br />
30 40 50 60 70 80 90 100<br />
Evaluation Scores [%]<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for TOTAL_Chemie_FQF4<br />
and TOTAL_Chemie_FQF5<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.0714<br />
p = 0.003863<br />
n_1 = 1022<br />
n_2 = 1531<br />
TOTAL_Chemie_FQF4<br />
TOTAL_Chemie_FQF5<br />
40 50 60 70 80 90 100<br />
Evaluation Scores [%]<br />
(a) Evaluations according to FQF 3.0 and FQF 4.0.<br />
(b) Evaluations according to FQF 4.0 and FQF 5.0.<br />
Frequency<br />
0 40 80 120<br />
Frequency<br />
0 50 100 150<br />
Distribution of TOTAL_Chemie_FQF5<br />
40 50 60 70 80 90 100<br />
Distribution of TOTAL_Chemie_FQF6<br />
40 50 60 70 80 90 100<br />
Shapiro-Wilk Normality test p-values<br />
p-values<br />
4,6E-32 9,4E-30 3,8E-30 1,7E-30<br />
Samples 793 1022 1531 1469<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for TOTAL_Chemie_FQF5<br />
and TOTAL_Chemie_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.03117<br />
p = 0.4602<br />
n_1 = 1531<br />
n_2 = 1469<br />
TOTAL_Chemie_FQF5<br />
TOTAL_Chemie_FQF6<br />
40 50 60 70 80 90 100<br />
Evaluation Scores [%]<br />
(c) Evaluations according to FQF 5.0 and FQF 6.0.<br />
TOTAL_Chemie_FQF3 1,000 3,7E-09 0 0<br />
TOTAL_Chemie_FQF4 3,7E-09 1,000 0,004 0,127<br />
TOTAL_Chemie_FQF5 0 0,004 1,000 0,460<br />
TOTAL_Chemie_FQF6 0 0,127 0,460 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(d) Distribution comparisons summary<br />
Figure 29: Distribution comparisons between the evaluation scores of chemical-part suppliers,<br />
performed according to different versions of Formel Q-Fähigkeit. The samples include quality<br />
auditing data from all geographic regions.<br />
77
In the case of chemical-part suppliers above (Figure 29) the obtained p-values are smaller than<br />
the significance level α = 5% in the cases in which results according to FQF 3.0 were compared<br />
to results according to FQF 4.0, as well as the comparison between results according to FQF 4.0<br />
and FQF 5.0. On the other hand, the p-value is greater than the 5%-significance level for the<br />
comparison between FQF 5.0 and FQF 6.0 evaluations. Thus, in the first two cases it is concluded<br />
that there is statistical significance and the null-hypothesis, that the two samples have the same<br />
distribution, is rejected, i.e. the distributions of the evaluations according to FQF 3.0, FQF 4.0 and<br />
FQF 5.0 are different from one another. The test results provide also statistical evidence that the<br />
evaluations performed according to FQF 5.0 and FQF 6.0 have similar distributions. In the latter<br />
case the null-hypothesis is not rejected.<br />
Similar conclusions can be made in the case of metal-part suppliers (Figure 30). Here again<br />
the p-values are smaller than the 5%-significance level for the comparisons between FQF 3.0 an<br />
FQF 4.0, as well as FQF 4.0 and FQF 5.0, meaning that the null-hypothesis, that the according<br />
samples have the same distributions, is rejected. Here again, when evaluations according to FQF<br />
5.0 are compared to evaluations according to FQF 6.0 the respective p-value is greater than the<br />
5%-significance level, meaning that the null-hypothesis is not rejected and that the two samples<br />
have the same distribution.<br />
78
Distribution of TOTAL_Metal_FQF3<br />
Distribution of TOTAL_Metal_FQF4<br />
Frequency<br />
0 50 150 250<br />
40 50 60 70 80 90 100<br />
Distribution of TOTAL_Metal_FQF4<br />
Frequency<br />
0 50 100 150<br />
Frequency<br />
0 50 100 150<br />
40 60 80 100<br />
Distribution of TOTAL_Metal_FQF5<br />
Frequency<br />
0 50 150 250<br />
TOTAL_Metal_FQF3<br />
TOTAL_Metal_FQF4<br />
TOTAL_Metal_FQF5<br />
TOTAL_Metal_FQF6<br />
TOTAL_Metal_FQF3<br />
TOTAL_Metal_FQF4<br />
TOTAL_Metal_FQF5<br />
TOTAL_Metal_FQF6<br />
40 50 60 70 80 90 100<br />
40 60 80 100<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for TOTAL_Metal_FQF3<br />
and TOTAL_Metal_FQF4<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.09903<br />
p = 2.044e−07<br />
n_1 = 1551<br />
n_2 = 1743<br />
TOTAL_Metal_FQF3<br />
TOTAL_Metal_FQF4<br />
40 50 60 70 80 90 100<br />
Evaluation Scores [%]<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for TOTAL_Metal_FQF4<br />
and TOTAL_Metal_FQF5<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.06194<br />
p = 0.0006026<br />
n_1 = 1743<br />
n_2 = 2682<br />
TOTAL_Metal_FQF4<br />
TOTAL_Metal_FQF5<br />
40 60 80 100<br />
Evaluation Scores [%]<br />
(a) Evaluations according to FQF 3.0 and FQF 4.0.<br />
(b) Evaluations according to FQF 4.0 and FQF 5.0.<br />
Shapiro-Wilk Normality test p-values<br />
Distribution of TOTAL_Metal_FQF5<br />
Frequency<br />
0 50 150 250<br />
40 60 80 100<br />
Distribution of TOTAL_Metal_FQF6<br />
p-values<br />
6,7E-36 2,2E-34 4,2E-46 8,6E-40<br />
Samples 1551 1743 2682 1932<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
Frequency<br />
0 50 100 200<br />
40 60 80 100<br />
TOTAL_Metal_FQF3 1,000 2,0E-07 0 0<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for TOTAL_Metal_FQF5<br />
and TOTAL_Metal_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.03381<br />
p = 0.1533<br />
n_1 = 2682<br />
n_2 = 1932<br />
TOTAL_Metal_FQF5<br />
TOTAL_Metal_FQF6<br />
40 60 80 100<br />
Evaluation Scores [%]<br />
TOTAL_Metal_FQF4 2,0E-07 1,000 0,001 0,001<br />
TOTAL_Metal_FQF5 0 0,001 1,000 0,153<br />
TOTAL_Metal_FQF6 0 0,001 0,153 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(c) Evaluations according to FQF 5.0 and FQF 6.0.<br />
(d) Distribution comparisons summary<br />
Figure 30: Distribution comparisons between the evaluation scores of metal-part suppliers, performed<br />
according to different versions of Formel Q-Fähigkeit. The samples include quality auditing<br />
data from all geographic regions.<br />
79
Distribution of TOTAL_Elektrik_FQF3<br />
Distribution of TOTAL_Elektrik_FQF4<br />
Frequency<br />
0 5 10 20 30<br />
60 70 80 90 100<br />
Distribution of TOTAL_Elektrik_FQF4<br />
Frequency<br />
0 10 20 30 40 50 60<br />
Frequency<br />
0 10 20 30 40 50 60<br />
60 70 80 90 100<br />
Distribution of TOTAL_Elektrik_FQF5<br />
Frequency<br />
0 20 40 60 80<br />
TOTAL_Elektrik_FQF3<br />
TOTAL_Elektrik_FQF4<br />
TOTAL_Elektrik_FQF5<br />
TOTAL_Elektrik_FQF6<br />
TOTAL_Elektrik_FQF3<br />
TOTAL_Elektrik_FQF4<br />
TOTAL_Elektrik_FQF5<br />
TOTAL_Elektrik_FQF6<br />
60 70 80 90 100<br />
60 70 80 90 100<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for TOTAL_Elektrik_FQF3<br />
and TOTAL_Elektrik_FQF4<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.05722<br />
p = 0.6866<br />
n_1 = 238<br />
n_2 = 453<br />
TOTAL_Elektrik_FQF3<br />
TOTAL_Elektrik_FQF4<br />
60 70 80 90 100<br />
Evaluation Scores [%]<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for TOTAL_Elektrik_FQF4<br />
and TOTAL_Elektrik_FQF5<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.07125<br />
p = 0.1292<br />
n_1 = 453<br />
n_2 = 667<br />
TOTAL_Elektrik_FQF4<br />
TOTAL_Elektrik_FQF5<br />
60 70 80 90 100<br />
Evaluation Scores [%]<br />
(a) Evaluations according to FQF 3.0 and FQF 4.0.<br />
(b) Evaluations according to FQF 4.0 and FQF 5.0.<br />
Shapiro-Wilk Normality test p-values<br />
Distribution of TOTAL_Elektrik_FQF5<br />
Frequency<br />
0 20 40 60 80<br />
50 60 70 80 90 100<br />
Distribution of TOTAL_Elektrik_FQF6<br />
p-values<br />
8,8E-16 2,2E-18 3,0E-19 2,3E-20<br />
Samples 238 453 667 460<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
Frequency<br />
0 20 40 60<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
50 60 70 80 90 100<br />
CDFs for TOTAL_Elektrik_FQF5<br />
and TOTAL_Elektrik_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.0518<br />
p = 0.4582<br />
n_1 = 667<br />
n_2 = 460<br />
TOTAL_Elektrik_FQF5<br />
TOTAL_Elektrik_FQF6<br />
50 60 70 80 90 100<br />
Evaluation Scores [%]<br />
(c) Evaluations according to FQF 5.0 and FQF 6.0.<br />
TOTAL_Elektrik_FQF3 1,000 0,687 0,091 0,094<br />
TOTAL_Elektrik_FQF4 0,687 1,000 0,129 0,157<br />
TOTAL_Elektrik_FQF5 0,091 0,129 1,000 0,458<br />
TOTAL_Elektrik_FQF6 0,094 0,157 0,458 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(d) Distribution comparisons summary<br />
Figure 31: Distribution comparisons between the evaluation scores of electrical-part suppliers,<br />
performed according to different versions of Formel Q-Fähigkeit. The samples include quality<br />
auditing data from all geographic regions.<br />
80
Things look a bit different in the case of electrical-part suppliers (Figure 31). Nevertheless,<br />
certain parallels to the previous two cases can be drawn. Here the Kolmogorov-Smirnov test<br />
does not show any statistically significant differences between the individual datasets on the 5%-<br />
significance level. However, p-values in the comparisons between the distributions of audit results<br />
according to the FQF 4.0 and the FQF 5.0 (p=0.129), the FQF 4.0 and FQF 6.0 (p=0.157), as well<br />
as in the cases in which audit result according to the FQF 3.0 are compared to audits according to<br />
FQF 5.0 (p=0.091) and FQF 6.0 (p=0.094) respectively, are significantly lower than the p-values<br />
for the comparisons between audits according to FQF 3.0 and FQF 4.0 (p=0.687) and between<br />
FQF 5.0 and FQF 6.0 (p=0.458). These results suggest that there are more parallels between the<br />
audits performed according to FQF 3.0 and FQF 4.0 and respectively FQF 5.0 and FQF 6.0 than<br />
between these two groups of results.<br />
The results of the Kolmogorov-Smirnov test show that indeed changes in the requirements defined<br />
in subsequent editions of Formel Q-Fähigkeit significantly influence the characteristics of<br />
empirical data. Evaluation score limits, which Volkswagen defines to differentiate between A-,<br />
B-, and C-suppliers, are potentially one of the major sources of difference between the individual<br />
groups of evaluations. This is supported by the fact that there is no statistically significant difference<br />
between the evaluation scores according to FQF 5.0 and FQF 6.0. The latter have identical<br />
requirements for an A- and a B-rating. On the other hand, the major differences in the empirical<br />
data occur between evaluations to FQF 3.0 and FQF 4.0 and the latter two compared to FQF 5.0<br />
and FQF 6.0 accordingly. The two editions – FQF 3.0 and FQF 4.0 – define also different limits<br />
for an A- and a B-rating respectively.<br />
The evaluations presented above analyze the similarities between twelve generalized sets of<br />
empirical data, which include audit results from a number of different regions. However, the<br />
results of the analysis presented in Section 7.1 identified considerable regional differences of the<br />
composition of empirical data. It is therefore meaningful to evaluate also the influence of the<br />
changes of Formel Q-Fähigkeit on the regional level. For the purpose, the initial twelve sets of<br />
empirical data were divided into a number of subsets based on the region in which the audits were<br />
performed. Using the same methodology as above, for each subset a Shapiro-Wilk normality test<br />
was performed. Since in almost all of the cases the data was found to be non-normally distributed a<br />
two-sample Kolmogorov-Smirnov test was used to identify any statistically significant differences<br />
between the distributions of the newly formed datasets. The results of the performed analysis are<br />
presented in Figures 58 through 78 in Section A in the Appendix.<br />
81
Chemical‐part Suppliers<br />
KS test p‐value<br />
Region<br />
Data Series<br />
FQF 4.0<br />
vs FQF 5.0<br />
FQF 5.0<br />
vs FQF 6.0<br />
Region A Team_A_Chemie_FQFx ‐ 0,046<br />
Region B Team_B_Chemie_FQFx 0,850 0,915<br />
Region C Team_C_Chemie_FQFx ‐ 1,000<br />
Region D Team_D_Chemie_FQFx ‐ 2,4E‐04<br />
Region E Team_E_Chemie_FQFx ‐ 0,011<br />
Region F Team_F_Chemie_FQFx 0,006 0,684<br />
Region G Team_G_Chemie_FQFx ‐ 0,006<br />
Region H Team_H_Chemie_FQFx ‐ 0,556<br />
Region I Team_I_Chemie_FQFx ‐ 0,661<br />
Metal‐part Suppliers<br />
KS test p‐value<br />
Region<br />
Data Series<br />
FQF 4.0<br />
vs FQF 5.0<br />
FQF 5.0<br />
vs FQF 6.0<br />
Region A Team_A_Metal_FQFx ‐ 0,104<br />
Region B Team_B_Metal_FQFx 0,041 0,531<br />
Region C Team_C_Metal_FQFx ‐ 3,3E‐04<br />
Region D Team_D_Metal_FQFx ‐ 1,5E‐05<br />
Region E Team_E_Metal_FQFx ‐ 0,472<br />
Region F Team_F_Metal_FQFx 5,4E‐06 0,448<br />
Region G Team_G_Metal_FQFx ‐ 0,446<br />
Region H Team_H_Metal_FQFx ‐ 1,5E‐07<br />
Region I Team_I_Metal_FQFx 0,177 0,993<br />
Region J Team_J_Metal_FQFx ‐ 0,374<br />
Electrical‐part Suppliers<br />
KS test p‐value<br />
Region<br />
Data Series<br />
FQF 4.0<br />
vs FQF 5.0<br />
FQF 5.0<br />
vs FQF 6.0<br />
Region C Team_C_Elektrik_FQFx ‐ 0,456<br />
Region E Team_E_Elektrik_FQFx ‐ 0,050<br />
Region F Team_F_Elektrik_FQFx 0,010 0,133<br />
Legend:<br />
significant on the 5% level<br />
x = 4, 5, or 6 ‐ FQF Version<br />
Figure 32: Summary of the two-sample Kolmogorov-Smirnov test p-values obtained by comparisons<br />
between the distributions of supplier evaluation scores according to the Formel Q-Fähigkeit<br />
versions 4.0, 5.0, and 6.0 respectively. The results are sorted by commodity type as well as region 1 .<br />
1 Note that the empirical information in some of the regional datasets refers to different supplier pools than the<br />
regional division used in Section 7.1. For this reason the region identification in this and the following sections does<br />
not correspond to the identification used in Section 7.1. However, this is irrelevant for the statistical analysis and its<br />
results.<br />
82
Figure 32 presents a summary of the performed Kolmogorov-Smirnov tests on the sets of regional<br />
empirical data. The test results show that also on the regional level evaluation scores obtained<br />
with respect to FQF 4.0 are significantly different from the evaluations performed according<br />
to FQF 5.0. The only two cases in which no statistical significance was observed are for the evaluations<br />
of metal-part suppliers in Region I, performed by the Volkswagen audit team in this region,<br />
and evaluations of chemical-part suppliers in Region B performed by the local audit team. In the<br />
first case the lack of statistical significance is most probably due to the relatively small sizes of<br />
the data samples – 19 (FQF 4.0) and 28 (FQF 5.0) accordingly. Moreover, the p-values of the<br />
Kolmogorov-Smirnov test for comparisons between audits to FQF 4.0 and FQF 5.0 (p=0.177) and<br />
between FQF 4.0 and FQF 6.0 (p=0.063) are much lower than the p-value of the Kolmogorov-<br />
Smirnov test comparing the evaluation results of metal-part suppliers in Region I with respect to<br />
FQF 5.0 and FQF 6.0 (p=0.993). These results suggest that there are definitely more parallels<br />
between the evaluations according to FQF 5.0 and FQF 6.0 than between these two and the audits<br />
performed according to FQF 4.0, which situation is in accordance with the observations in the rest<br />
of the regions. On the other hand, there is no obvious reason why the evaluations of chemical-part<br />
suppliers performed by the audit team in Region B do not differ for FQF 4.0 and FQF 5.0, especially<br />
provided that in the rest of the analyzed cases there is statistical evidence for such difference.<br />
Moreover, unlike the case with metal-part suppliers in Region I, the sample sizes in the second case<br />
are considerably larger – 69 evaluations for FQF 4.0 and 103 evaluations for FQF 5.0. The p-value<br />
of the test is also very high – p=0.850.<br />
The statistical test results show also that in the majority of cases there is no statistically significant<br />
difference between the evaluation scores performed according to FQF 5.0 and FQF 6.0<br />
accordingly. In the cases, in which statistical significance is observed, the differences in the empirical<br />
data are very probably not due to the changes in the Formel Q-Fähigkeit, and are rather due<br />
to regional trends in the supplier development. Region D for example is one of the markets with<br />
high strategic importance for Volkswagen Group but is subject to high personnel fluctuations and<br />
lack of quality sustainability 15 . This is the reason why Volkswagen has been running an intensive<br />
supplier qualification program in this region for some years now. The program aims at improving<br />
the quality capability of the local suppliers. It is therefore normal to expect that the overall quality<br />
capability of the supplier base in Region D will improve over time. This is indeed what is observed<br />
in Figures 60 and 72 in Section A in the Appendix. The quality audits performed to FQF 6.0 reveal<br />
higher evaluation scores compared to evaluations to FQF 5.0. This trend is consistent both for<br />
15 Source: personal communication with Volkswagen AG employees.<br />
83
chemical-part as well as metal-part suppliers.<br />
Other regions which show positive trends of development of the supplier audit evaluations<br />
are Region A (Figure 57) and Region G (Figure 63) for chemical-part suppliers, and Region C<br />
(Figure 71) and Region H (Figure 76) for metal-part suppliers. It is anticipated that in most of<br />
these cases the positive shift of the evaluation scores is due to improved quality performance of<br />
suppliers. It is not excluded, however, that other factors could also play a role and influence<br />
positively the supplier evaluation scores.<br />
7.2.2 Period of Relevance of Quality Audit Results<br />
The observations in the previous section lead to the next potential source of data bias - development<br />
of supplier’s quality capability over time. On the one hand, a particular supplier evaluation by<br />
Volkswagen auditors lasts on average several days, while the time between quality audits of the<br />
same supplier is of the order of months or even years (this point was already mentioned in the<br />
sampling discussion above). On the other hand, process variation is natural for all productions,<br />
as the intensity of variations is influenced by factors such as maturity of the production process,<br />
implementation of quality improvement programs, period of operation of the production facility,<br />
changes down the supply chain, and many others. This naturally poses the question: Even if at the<br />
time of the audit a quality evaluation result accurately reflects the quality capability of a particular<br />
supplier, how long is this audit result relevant?<br />
The answer to this question is quite important for the subsequent steps of the analysis. As<br />
proposed in the beginning of this paper, a possible way to assess the effectiveness of Volkswagen’s<br />
audit evaluations is to compare the quality capability scores of suppliers to their quality performance<br />
records, under the initial assumption that, if audit evaluations are performed effectively,<br />
suppliers with better audit scores would have better quality performance and vice versa. However,<br />
such comparison is only meaningful, if the quality capability evaluation of each supplier<br />
adequately reflects supplier’s quality capability at the time its quality performance was recorded.<br />
More precisely, the audit result provides only an assessment of the degree, to which a particular,<br />
temporary state of the constantly changing supplier production system meets Volkswagen Group<br />
customer requirements. By contrast, supplier’s quality performance is at any given point of time<br />
directly related to the actual state of the production process. It is influenced by the error sources<br />
down the supplier production chain, which can vary significantly over time. Thus, if the time separation<br />
between the last audit of a supplier and its quality performance record is large, it is highly<br />
84
probable that supplier’s quality capability, at the moment its quality performance was recorded,<br />
does not match its quality capability record in Volkswagen’s database.<br />
Let’s illustrate this with a concrete example – in this case a best-case scenario. Suppose the<br />
supplier A-Supplier was audited several years ago and received an A-rating. Let’s also suppose its<br />
production process fulfilled Volkswagen’s quality requirements to 93%. If we now assume that the<br />
quality audit has adequately evaluated supplier’s quality capability, this means that in a particular<br />
time window around the audit the supplier delivered to Volkswagen indeed with the performance<br />
of an A-93%-supplier. Suppose now that after its last audit the supplier implements a one-year<br />
improvement program, aiming at elimination of major sources of process variation. As a result,<br />
today A-Supplier employs a significantly better production process and does not fulfill any more<br />
just 93% of Volkswagen Group’s customer specific requirements – rather 98%. However, supplier’s<br />
official quality capability record of A 93% in Volkswagen’s database will not change, since no new<br />
audit at supplier’s site has been conducted. Thus, regardless of the fact that an effective supplier<br />
evaluation was performed, the quality audit score of A-Supplier does not match its actual quality<br />
capability any more. On the other hand, due to the implemented improvement measures the quality<br />
of supplier’s final products would increase and the amount of rejects at Volkswagen’s production<br />
lines would drop considerably. Consequently, even though its ”official” quality rating is A 93%, A-<br />
Supplier would supply with the quality of an A-98% rated supplier. It is very difficult to distinguish<br />
such a case from a case, in which the according supplier quality capability is indeed A 93% and<br />
matches its quality capability rating in Volkswagen’s database.<br />
Such mismatches of the measured quality capability (through an audit) and supplier’s actual<br />
quality performance introduce noise in the empirical data and could therefore distort analytical<br />
results. A possible workaround would be to introduce a suitable weighing function, so that quality<br />
performance data obtained around the time of the audit is given more importance than the rest of the<br />
empirical data. Finding the proper definition of the weighing function, however, is extremely difficult,<br />
as variations even for similar production processes are company-specific and could happen at<br />
substantially different time scales. Nevertheless, such a weighing function is especially important<br />
for the analysis of empirical data generated in regions characterized with unstable supplier quality<br />
capability and high process fluctuations. Suitable input for the definition of a weighing function<br />
are supplier-specific process variation indicators such as cpk-values, company-specific KPIs, etc.<br />
However, the empirical data analyzed here does not include any such information. This is the<br />
reason why, the comparisons between quality capability and quality performance presented later<br />
in this paper are restricted mainly to regions with relatively high stability of suppliers’ production<br />
85
processes.<br />
7.2.3 The Factor ”Human” and People Calibration<br />
When considering reliability of recorded data one must also take into account the influence of<br />
the ”human” element. The fact that data is recorded by human beings, each with their individual<br />
perceptions and judgments, can lead to significant variations. One of the important tasks in this<br />
project is to account for the latter when performing the analysis. The large number of people,<br />
responsible for data collection, naturally poses the problem of calibration. There are roughly about<br />
a hundred Volkswagen auditors worldwide. On the other hand, there are several thousand people<br />
involved in the evaluation of suppliers’ quality performance. The complexity of calibration is further<br />
increased also by the type of evaluations the two groups of quality inspectors should perform.<br />
In this function people can be regarded as measurement gauges, which have to provide the according<br />
reproducibility and repeatability of measurements. In order to do so they have to be properly<br />
calibrated and their ability to generate consistent measurements constantly monitored. In general<br />
it is meaningful to define calibration measures analogous to the Cg and Cgk indicators used for<br />
assessing a systems’ measuring capability as well as devise people calibration methods. Given<br />
the importance of people calibration, the next part of this analysis tests for differences in people’s<br />
judgments, i.e. differences in the measurement results of the ”human gauges”.<br />
7.2.3.1 Calibration of Supplier Quality Auditors<br />
Due to their relatively small number, calibration of quality auditors seems to be easier as<br />
compared to calibration of the large number of people, who evaluate suppliers’ quality performance.<br />
Nevertheless, deviations might occur and it is therefore necessary to evaluate their<br />
influence on the empirical data. Suppliers’ quality capability is evaluated by auditors, who have<br />
been particularly trained for the purpose. Each of Volkswagen’s auditors has to pass a number of<br />
quality management trainings before he is authorized to perform audits. Furthermore, the use of<br />
the standardized auditing questionnaire in Formel Q-Fähigkeit worldwide contributes significantly<br />
to a coherent supplier evaluation process. Additionally, Volkswagen requires from its auditors to<br />
perform supplier quality evaluations at least 125 days annually in order to keep their know-how<br />
up-to-date with the constant changes in the sector. Auditor calibration is aided also by a number<br />
of communication channels on the local and international level, which facilitate the exchange of<br />
know-how between the individual quality auditors. These include internal meetings on a weekly<br />
86
asis within the individual auditing units, as well as various regular trainings, workshops, and<br />
seminars (e.g.<br />
Internationale Tagung der Auditoren Leiter (ITAL), Jahres Auditoren Tagung),<br />
which cover important topics of the every-day auditing activities.<br />
Such know-how transfer<br />
suggests relatively good consistency of the auditing results and it is expected that, if two auditors<br />
evaluate the same supplier at the same point of time, their final evaluations will be coherent.<br />
However, in order to assess the effectiveness of auditor calibration in practice, a direct comparison<br />
between the audit results obtained from different auditors at the same production location<br />
is not informative, due to the fact that these are usually separated in time and in the meantime<br />
supplier’s quality capability is influenced by process variation as already discussed in detail in the<br />
previous section. This is the reason why here a different approach was used to evaluate how good<br />
the calibration of Volkswagen auditors is. Rather than concentrating on single suppliers, this part<br />
of the analysis considered entire groups of suppliers based on their location, under the assumption<br />
that the latter are subject to similar influencing factors common to the respective region – for example<br />
suppliers in India. Thus, if two groups of auditors 16 perform a large number of audits in the<br />
same supplier pool in the same time window, it is expected that their evaluations will have similar<br />
distributions, assuming that the two audit groups are equally calibrated. In the previous sections it<br />
was observed that in certain regions such as Region D or Region H, for example, there has been a<br />
positive development of the overall quality capability of suppliers in these regions from the time<br />
period of FQF 5.0 to the time period of FQF 6.0. To minimize the influence of such market-specific<br />
developments the analysis discussed in this section restricted the time period considered for each<br />
comparison to the period of relevance of a single edition of Formel Q-Fähigkeit.<br />
The analytic method used for the comparisons in this section is identical to the one introduced<br />
in Section 7.2.1. For the purpose the empirical quality capability evaluation data was divided into<br />
a number of datasets based on the audit team, which performed the evaluations, and the region<br />
in which the according suppliers are located. For each set of data a Shapiro-Wilk normality test<br />
was conducted. Subsequently the two-sample Kolmogorov-Smirnov test was used to identify any<br />
potential differences in the evaluation distributions. At this point it is necessary to mention that<br />
comparisons were carried out only for part of the regions introduced in the preceding sections. The<br />
only reason why audit data from the remaining regions, analyzed in the previous sections, were not<br />
included in this part of the analysis is because they do not contain large enough samples of quality<br />
evaluations performed by two different auditor teams. In such regions the only available datasets<br />
16 Note that evaluations focused on single auditors were not conducted due to an internal Volkswagen restriction,<br />
which forbids any kind of individual-related performance evaluations.<br />
87
large enough for statistical analysis are those generated by the respective local auditor teams. In<br />
total 10 different sets of empirical data over the three commodities were evaluated.<br />
The majority of the evaluated audit records showed very good consistency of the audit results,<br />
which implies that Volkswagen’s audit teams are well calibrated. In the few cases, in which statistically<br />
significant difference between the compared datasets was observed, either the number<br />
of observations was so low, that the small size of the datasets is the most probable reason for the<br />
statistical significance of the Kolmogorov-Smirnov test, or the empirical data were subject to specific<br />
regional influences and such differences had to be expected. Thus the observed deviations do<br />
not necessarily indicate different calibration of the audit teams. Moreover, these results show that<br />
the evaluations of the individual auditor teams can be subject to the influence of a small cultural<br />
human factor, even when the auditors are equally well calibrated. Based on the observations of<br />
this part of the analysis, to avoid any influence of potentially different supplier audit evaluations,<br />
the subsequent analysis evaluates audit results from different auditor teams separately.<br />
7.2.3.2 Calibration of Production Quality Assessment<br />
If quality capability data are to be compared to quality performance data, one has to account<br />
for any potential biases in suppliers’ quality performance as well. The evaluation process<br />
in the latter case is again highly influenced by the judgments of human beings and therefore also<br />
here it is important to make sure that the same evaluation standards are applied. However, the<br />
number of people involved in the evaluation process of supplier quality performance is much<br />
larger than the total number of Volkswagen auditors conducting quality audits worldwide. Apart<br />
from quality inspectors and supplier quality engineers, practically every single employee involved<br />
in the assembly process is also involved in the detection of eventual quality-related problems of<br />
the supplied components and consequently in the assessment of suppliers’ quality performance.<br />
Moreover, calibration of all participants in the performance evaluation process is particularly<br />
demanding due to the extremely heterogeneous character of the assessed quality. Calibrating the<br />
assessment of quality for products, whose quality can be measured using quantitative parameters<br />
such as length, weight, temperature, and other measurable physical characteristics, is much easier<br />
as compared to calibrating the evaluation of haptic quality for product characteristics such as<br />
color, look, feel, etc. Often, in the latter cases calibration is only possible with the help of shared<br />
master and limit samples. Their use, however, can still be subject to individual influences. For this<br />
reason, empirical data originating from different Volkswagen production plants might be subject<br />
88
to differences even for the same types of products.<br />
To assess the extent to which quality performance data from the different Volkswagen production<br />
facilities differ from one another, the supplier quality performance records from a number of<br />
different plants were compared. The available empirical data from each plant included in the analysis<br />
were divided into datasets based on the type of delivered products using the material group<br />
differentiation (used by the Volkswagen Procurement Department) already introduced in the previous<br />
sections. Similar to the analysis in Section 7.2.3.1 for each dataset a Shapiro-Wilk normality<br />
test was conducted and the individual sets of empirical data were compared pairwise with the help<br />
of a two-sample Kolmogorov-Smirnov test. The quality performance records of 98 suppliers were<br />
evaluated for two different material groups – 0058 (”Moulded parts < DIN A4 for body”) and<br />
0068 (”Moulded parts > DIN A4 for body”). The according empirical data was reported by 11<br />
different Volkswagen production facilities. Suppliers selected for the evaluation deliver the same<br />
type of products to two different Volkswagen plants. Each of the conducted pairwise comparisons<br />
includes the quality performance of all suppliers, which deliver components from a single material<br />
group simultaneously to two Volkswagen production facilities. Figure 33, for example, presents the<br />
test results for the comparison of two datasets – namely A_M_WSGR_0058_Plant_A_ppm and<br />
A_M_WSGR_0058_Plant_M_ppm. The two datasets include the quality performance records of<br />
only those suppliers (in the particular case 10), which deliver parts from the material group 0058 simultaneously<br />
to both Plant A and Plant M. The first dataset – A_M_WSGR_0058_Plant_A_ppm<br />
– contains the quality performance records of these suppliers for their deliveries to Plant A, while<br />
the second dataset – A_M_WSGR_0058_Plant_M_ppm – contains the quality performance<br />
records of the exactly same suppliers for their deliveries to Plant M. Using this methodology a<br />
total of seven different pairs of data were generated. The results from the comparisons of the remaining<br />
pairs of performance evaluations are presented in Figures 79 through 84 in Section B in<br />
the Appendix.<br />
This approach allows to directly compare the quality performance evaluations based on the<br />
same products coming from the same production processes conducted by two independent groups<br />
of evaluators in the respective Volkswagen production facilities. Any differences in the distributions<br />
of the compared pairs of data could suggest discrepancies in the evaluation methods.<br />
The conducted analysis shows, however, that the evaluations of the individual plants are rather<br />
consistent. In all cases the resulting p-values of the Kolmogorov-Smirnov test are greater than<br />
89
Distribution of A_M_WSGR_0058_Plant_A_ppm<br />
Frequency<br />
0.0 0.5 1.0 1.5 2.0<br />
Frequency<br />
0.0 1.0 2.0 3.0<br />
Plant: A<br />
WSGR: 0058<br />
Plant: M<br />
WSGR: 0058<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
Plant: A<br />
WSGR: 0058<br />
Plant: M<br />
WSGR: 0058<br />
1 2 3 4<br />
Distribution of A_M_WSGR_0058_Plant_M_ppm<br />
1 2 3 4<br />
CDFs for A_M_WSGR_0058_Plant_A_ppm<br />
and A_M_WSGR_0058_Plant_M_ppm<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.3077<br />
p = 0.5696<br />
n_1 = 13<br />
n_2 = 13<br />
A_M_WSGR_0058_Plant_A_ppm<br />
A_M_WSGR_0058_Plant_M_ppm<br />
1 2 3 4<br />
ppm Scores [log]<br />
Shapiro-Wilk Normality test p-values<br />
p-values 0,180 0,159<br />
Samples 13 13<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
Plant: A<br />
WSGR: 0058<br />
Plant: M<br />
WSGR: 0058<br />
1,000 0,570<br />
0,570 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(a) Performance records comparison based on ppm<br />
(b) Distribution comparison summary<br />
Figure 33: Comparison between the quality performance records of the same 13 suppliers reported<br />
by two different plants – Plant A and Plant M. The reported defective parts per million (ppm) refer<br />
to quality problems of components from the material group 0058 (”Moulded parts < DIN A4 for<br />
body”).<br />
0.05, which in statistical terms means that the null-hypothesis of the non-parametric test is not<br />
rejected on the 5%-significance level. This result is not surprising given that the evaluated parts<br />
are metal components used in the raw construction of the automobile, for which the most important<br />
characteristic is parts’ geometry, i.e. parts’ quality is defined by quantifiable metrics.<br />
Even though the analytical results above do not suggest any discrepancies between the evaluation<br />
methods of the individual Volkswagen production plants (at least not for the evaluated product<br />
groups), empirical data from the different Volkswagen locations may still contain certain differences.<br />
This is demonstrated by the analytical results presented in Figures 34 through 36. Here the<br />
quality performance records of suppliers which deliver a particular type of products reported by<br />
the different Volkswagen locations are compared to one another. The statistical evaluation contains<br />
performance records for three different material groups – 0058 (Figure 34), 0068 (Figure 35,<br />
and 0096 – ”Reservoirs, covers, pipes, wires” (Figure 36). Each figure presents a summary of the<br />
conducted Shapiro-Wilk normality test and the two-sample Kolmogorov-Smirnov test accordingly.<br />
90
Plant: A<br />
WSGR: 0058<br />
Plant: B<br />
WSGR: 0058<br />
Plant: E<br />
WSGR: 0058<br />
Plant: G<br />
WSGR: 0058<br />
Plant: H<br />
WSGR: 0058<br />
Plant: K<br />
WSGR: 0058<br />
Plant: L<br />
WSGR: 0058<br />
Plant: M<br />
WSGR: 0058<br />
Plant: N<br />
WSGR: 0058<br />
Plant: O<br />
WSGR: 0058<br />
Plant: P<br />
WSGR: 0058<br />
Plant: A<br />
WSGR: 0058<br />
Plant: B<br />
WSGR: 0058<br />
Plant: E<br />
WSGR: 0058<br />
Plant: G<br />
WSGR: 0058<br />
Plant: H<br />
WSGR: 0058<br />
Plant: K<br />
WSGR: 0058<br />
Plant: L<br />
WSGR: 0058<br />
Plant: M<br />
WSGR: 0058<br />
Plant: N<br />
WSGR: 0058<br />
Plant: O<br />
WSGR: 0058<br />
Plant: P<br />
WSGR: 0058<br />
Shapiro-Wilk Normality test p-values<br />
p-values 0,048 0,235 0,056 0,050 0,103 0,008 0,005 0,001 0,003 0,001 0,008<br />
Samples 25 17 20 23 32 27 27 31 24 22 29<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
Plant: A<br />
WSGR: 0058<br />
Plant: B<br />
WSGR: 0058<br />
Plant: E<br />
WSGR: 0058<br />
Plant: G<br />
WSGR: 0058<br />
Plant: H<br />
WSGR: 0058<br />
Plant: K<br />
WSGR: 0058<br />
Plant: L<br />
WSGR: 0058<br />
Plant: M<br />
WSGR: 0058<br />
Plant: N<br />
WSGR: 0058<br />
Plant: O<br />
WSGR: 0058<br />
Plant: P<br />
WSGR: 0058<br />
1,000 0,755 0,711 0,107 0,707 0,012 0,632 0,345 0,327 0,085 0,010<br />
0,755 1,000 0,804 0,208 0,632 0,026 0,807 0,984 0,386 0,552 0,215<br />
0,711 0,804 1,000 0,579 0,458 0,001 0,228 0,269 0,857 0,338 0,015<br />
0,107 0,208 0,579 1,000 0,093 4,9E-06 0,173 0,131 0,553 0,081 0,008<br />
0,707 0,632 0,458 0,093 1,000 0,004 0,193 0,222 0,228 0,043 0,005<br />
0,012 0,026 0,001 4,9E-06 0,004 1,000 0,004 0,003 2,7E-04 6,4E-05 1,1E-06<br />
0,632 0,807 0,228 0,173 0,193 0,004 1,000 0,786 0,139 0,474 0,121<br />
0,345 0,984 0,269 0,131 0,222 0,003 0,786 1,000 0,112 0,615 0,173<br />
0,327 0,386 0,857 0,553 0,228 2,7E-04 0,139 0,112 1,000 0,014 0,002<br />
0,085 0,552 0,338 0,081 0,043 6,4E-05 0,474 0,615 0,014 1,000 0,147<br />
0,010 0,215 0,015 0,008 0,005 1,1E-06 0,121 0,173 0,002 0,147 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
Figure 34: Comparison between the supplier quality performance records reported by eleven different<br />
plants based on ppm. The reported defective parts per million (ppm) refer to quality problems<br />
of components from the material group 0058 (”Moulded parts < DIN A4 for body”).<br />
91
Plant: B<br />
WSGR: 0068<br />
Plant: G<br />
WSGR: 0068<br />
Plant: H<br />
WSGR: 0068<br />
Plant: I<br />
WSGR: 0068<br />
Plant: K<br />
WSGR: 0068<br />
Plant: L<br />
WSGR: 0068<br />
Plant: N<br />
WSGR: 0068<br />
Plant: O<br />
WSGR: 0068<br />
Plant: P<br />
WSGR: 0068<br />
Plant: B<br />
WSGR: 0068<br />
Plant: G<br />
WSGR: 0068<br />
Plant: H<br />
WSGR: 0068<br />
Plant: I<br />
WSGR: 0068<br />
Plant: K<br />
WSGR: 0068<br />
Plant: L<br />
WSGR: 0068<br />
Plant: N<br />
WSGR: 0068<br />
Plant: O<br />
WSGR: 0068<br />
Plant: P<br />
WSGR: 0068<br />
Shapiro-Wilk Normality test p-values<br />
p-values 0,610 0,987 0,450 0,430 0,261 0,244 0,023 0,092 0,022<br />
Samples 25 19 28 24 35 37 26 23 21<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
Plant: B<br />
WSGR: 0068<br />
Plant: G<br />
WSGR: 0068<br />
Plant: H<br />
WSGR: 0068<br />
Plant: I<br />
WSGR: 0068<br />
Plant: K<br />
WSGR: 0068<br />
Plant: L<br />
WSGR: 0068<br />
Plant: N<br />
WSGR: 0068<br />
Plant: O<br />
WSGR: 0068<br />
Plant: P<br />
WSGR: 0068<br />
1,000 0,973 0,257 0,349 0,927 0,960 0,216 0,714 0,016<br />
0,973 1,000 0,145 0,124 0,290 0,961 0,316 0,445 0,010<br />
0,257 0,145 1,000 0,865 0,285 0,029 0,026 0,111 0,001<br />
0,349 0,124 0,865 1,000 0,952 0,273 0,080 0,089 0,012<br />
0,927 0,290 0,285 0,952 1,000 0,401 0,134 0,215 0,020<br />
0,960 0,961 0,029 0,273 0,401 1,000 0,053 0,414 0,005<br />
0,216 0,316 0,026 0,080 0,134 0,053 1,000 0,976 0,029<br />
0,714 0,445 0,111 0,089 0,215 0,414 0,976 1,000 0,018<br />
0,016 0,010 0,001 0,012 0,020 0,005 0,029 0,018 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
Figure 35: Comparison between the supplier quality performance records reported by nine different<br />
plants based on ppm. The reported defective parts per million (ppm) refer to quality problems<br />
of components from the material group 0068 (”Moulded parts > DIN A4 for body”).<br />
For material group 0058 supplier quality performance data from Plant K stands out as systematically<br />
different from the rest of the analyzed data. In all cases the Kolmogorov-Smirnov<br />
test’s low p-value for this particular dataset indicates statistically significant difference on the 5%-<br />
significance level. Other plants which show potential differences between the quality performance<br />
of their suppliers and the rest of the delivering companies are Plant O and Plant P. Plant P stands<br />
out as systematically different also in the case of material group 0068. Here the p-values of the<br />
92
Kolmogorov-Smirnov test are below the 5%-significance level in all 8 comparisons (Figure 35).<br />
Other plants for which the Kolmogorov-Smirnov test revealed statistically significant results are<br />
Plant H and Plant I as well as Plant N. Apart from the identified differences the rest of the evaluations<br />
were found to be statistically similar, meaning that suppliers which deliver the material<br />
groups 0058 and 0068 to the different Volkswagen plants have consistent quality.<br />
Shapiro‐Wilk Normality test p‐values<br />
Plant: C<br />
WSGR: 0096<br />
Plant: D<br />
WSGR: 0096<br />
Plant: F<br />
WSGR: 0096<br />
Plant: H<br />
WSGR: 0096<br />
Plant: J<br />
WSGR: 0096<br />
Plant: L<br />
WSGR: 0096<br />
p‐values 0,088 0,475 0,086 0,112 0,060 0,575<br />
Samples 17 15 24 23 15 15<br />
Kolmogorov‐Smirnov Distribution Comparison test<br />
p‐values matrix<br />
Plant: C<br />
WSGR: 0096<br />
Plant: D<br />
WSGR: 0096<br />
Plant: F<br />
WSGR: 0096<br />
Plant: H<br />
WSGR: 0096<br />
Plant: J<br />
WSGR: 0096<br />
Plant: L<br />
WSGR: 0096<br />
Plant: C<br />
WSGR: 0096<br />
Plant: D<br />
WSGR: 0096<br />
Plant: F<br />
WSGR: 0096<br />
Plant: H<br />
WSGR: 0096<br />
Plant: J<br />
WSGR: 0096<br />
Plant: L<br />
WSGR: 0096<br />
1,000 0,531 0,013 0,026 0,134 0,604<br />
0,531 1,000 7,8E‐05 1,1E‐04 0,003 0,026<br />
0,013 7,8E‐05 1,000 0,097 0,854 0,009<br />
0,026 1,1E‐04 0,097 1,000 0,105 0,409<br />
0,134 0,003 0,854 0,105 1,000 0,028<br />
0,604 0,026 0,009 0,409 0,028 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
Figure 36: Comparison between the supplier quality performance records reported by six different<br />
plants based on ppm. The reported defective parts per million (ppm) refer to quality problems of<br />
components from the material group 0096 (”Reservoirs, covers, pipes, wires”).<br />
93
Things are a bit different, however, for the case of material group 0096 (Figure 36). Here more<br />
than half of the performed comparisons revealed statistically significant differences between the<br />
individual datasets which made this particular case suitable for closer examination (a ”technical<br />
drill down”). The most probable reason for these differences is the high diversity of components<br />
which comprise this material group. A closer look at the available information revealed that material<br />
group 0096 has a total of 15 different subgroups in contrast to material group 0058 comprising<br />
of only 2 subgroups and material group 0068, which has only 4 subgroups 17 . The differences of<br />
the individual types of products, whose quality performance was evaluated, can also be derived<br />
from the fields of operation of the individual plants. Among the evaluated plants are plants, which<br />
build gearboxes, motors, after-sales replacement components, as well as plants involved in the final<br />
assembly of automobiles for different market segments.<br />
The analytical results included in this section underline important differences also in the character<br />
of quality performance data. Such discrepancies between the individual supplier records need<br />
to be handled with care, as in particular cases direct comparisons between the separate datasets are<br />
not desirable. To avoid potential negative influences of such different character of the empirical<br />
data on the final analytical results, quality performance data not only for different material groups<br />
but also data from the same material group reported by different plants of the Volkswagen Group<br />
are evaluated separately in the subsequent analysis.<br />
17 Note that the empirical data included in this analysis are not detailed enough and do not provide any information<br />
about the subgroups of the individual material groups.<br />
94
7.3 Matching Quality Capability and Quality Performance<br />
The steps of the project presented in Section 7.1 were necessary to eliminate any incomplete<br />
records and thus assure a reliable set of empirical data, while in Section 7.2 it was evaluated<br />
under which pre-conditions and circumstances it is reasonable to correlate quality capability and<br />
quality performance data. In this section the quality performance records of two groups of suppliers<br />
are compared to their quality audit scores in an attempt to give a possible answer to the<br />
research questions defined in the beginning of this paper. The two datasets were built considering<br />
the possible sources of data bias presented in the discussion in Section 7.2. However, to obtain the<br />
quality evaluations of the respective supplier production processes, which are relevant for the considered<br />
components, turned out to be a significant challenge and cost a significant amount of effort<br />
and time. This is because, while quality performance data are structured according to material<br />
groups defined by the Volkswagen Group Procurement department (Konzern Beschaffung), suppliers’<br />
quality capability is assessed with respect to the product groups defined by the Group Quality<br />
Assurance department (Konzern Qualitätssicherung). Even though the two types of categorization<br />
cover the same spectrum of components, there are little parallels between the two (as already discussed<br />
in one of the previous sections), and the only way to match them was to manually process<br />
the records (no automation of the process was possible). This was necessary due to the fact that<br />
at the time the analysis was performed there was no correspondence matrix, which described the<br />
relations between the two types of categorization. Defining such a matrix is a challenge in itself,<br />
due to the fact that the criteria used to define the two categorizations are of completely different<br />
character. While material groups use application as their main differentiating criterion, product<br />
groups are primarily based on the type of process, which is required to manufacture the according<br />
components.<br />
In the majority of cases each supplier production location has several quality capability evaluations<br />
according to different product groups. The number of evaluations depends on the diversity<br />
of produced components and accordingly the variety of production processes in the according facility.<br />
However, to tell which exactly of these audit evaluations assesses the quality capability of<br />
the production process, used to manufacture a particular component, which appears in supplier’s<br />
quality performance record, is not possible directly given the available data identification system<br />
currently in use at Volkswagen. Therefore, the relevant capability score had to be manually identified<br />
for every single supplier quality performance record included in the analysis. Given the high<br />
processing overhead, the analysis presented here was limited to the two presented cases.<br />
95
At this point it is necessary to mention that this particular challenge is relevant not only for<br />
the purposes of this analysis but also for Volkswagen’s internal processes such as the contract<br />
awarding by the Corporate Sourcing Committee (CSC), which needs inputs both from the quality<br />
assurance and procurement departments for the individual sourcing decisions. The problem is not<br />
new to Volkswagen, but at the time the evaluations were carried out there was still no functioning<br />
solution to it. In 2010 the Group Procurement, Volkswagen Research and Development, as well as<br />
the Group Quality Assurance departments discussed possible ways to unify the different product<br />
identification systems. One possible solution includes the definition of a relatively complex identification<br />
system (with several levels of identification), which will allow for the seamless interchange<br />
of information between the different databases of the according departments. The proposed solution<br />
has already been partially implemented. Once this concept is fully functional, it will allow for<br />
a more comprehensive, largely automated analysis of the type presented here.<br />
The first set of analyzed data contains the<br />
quality performance records of 31 suppliers,<br />
which deliver products from the material group<br />
0096 (”Reservoirs, covers, pipes, wires”) to Plant<br />
H. The quality performance records include the<br />
number of ppm each supplier accumulated over<br />
the period of three years – 2008, 2009, and 2010<br />
respectively. However, not all suppliers were active<br />
throughout the entire time period. What is<br />
also important to mention is that not all of the<br />
suppliers included in the analysis have had quality<br />
related problems in the respective period of<br />
time. On the contrary, the quality records of 18<br />
suppliers, representing almost 60% of the evaluated<br />
records, do not contain any reports of delivery<br />
of defective components. At this point, considering<br />
the hypothesis defined in the beginning<br />
of this paper, it is reasonable to expect, that suppliers<br />
which have no ppm have also better quality<br />
capability evaluations compared to suppliers,<br />
which experienced quality related problems. The<br />
Frequency<br />
0 1 2 3 4<br />
Frequency<br />
0 1 2 3 4<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
Distribution of Ep no ppms<br />
85 90 95 100<br />
Distribution of Ep with ppms<br />
85 90 95 100<br />
CDFs for Ep no ppms<br />
and Ep with ppms<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.1667<br />
p = 0.9848<br />
n_1 = 18<br />
n_2 = 13<br />
Ep no ppms<br />
Ep with ppms<br />
85 90 95 100<br />
Evaluation Scores [%]<br />
Figure 37: Distribution comparison between<br />
the evaluation scores of suppliers without reported<br />
quality related problems in the series<br />
production (E P no ppms) and suppliers with<br />
reported problems (E P with ppms). Both<br />
groups of suppliers delivered components from<br />
the material group 0096 (”Reservoirs, covers,<br />
pipes, wires”) to Plant H.<br />
96
overall quality capability evaluations of the two groups of suppliers – with and without ppm - are<br />
compared in Figure 37 using a two-sample Kolmogorov-Smirnov test. For both distributions a<br />
Shapiro-Wilk normality test revealed statistically significant evidence that they are non-normally<br />
distributed on the 5%-significance level with p-values of 0.021 in the case of suppliers without<br />
ppm, and 0.019 in the case of suppliers with ppm accordingly. The performed Kolmogorov-<br />
Smirnov test reveals no statistically significant result on the 5%-significance level, contrary to<br />
the expectations. Moreover, the according p-value of 0.985 of the non-parametric test is rather<br />
high, indicating that the evaluations of the two groups of suppliers are almost identical.<br />
Figure 38 presents the results of a linear regression<br />
between the quality performance of suppliers<br />
with ppm (the respective ppm records were<br />
evaluated on a logarithmic scale with base 10)<br />
and their quality capability evaluations. The estimated<br />
parameters of the least square linear fit are<br />
an intercept of β 0 = −6.439 (std. error= 3.977)<br />
and a slope of β 1 = 0.0939 (std. error= 0.044).<br />
Additionally an F-test was conducted to evaluated<br />
the significance of the estimated parameters<br />
of the least squares fit. The null-hypothesis of the<br />
performed test is that β 1 = 0, i.e. there is no correlation<br />
between the two compared datasets, with<br />
alternative hypothesis that β 1 ≠ 0. The p-value<br />
of the F-test of 0.056 on 1 and 11 degrees of freedom<br />
indicates a marginally significant result on<br />
the 5%-significance level, i.e. the null-hypothesis<br />
of the F-test can be rejected and the least squares<br />
fit shows indeed a positive correlation between<br />
PPM [log]<br />
20 50 100 200<br />
●<br />
Quality Capability vs PPM<br />
●<br />
●<br />
●<br />
●<br />
84 86 88 90 92<br />
Quality Capability Evaluation [%]<br />
Material Group: 0096 (Reservoirs, covers, pipes, wires)<br />
Plant: Plant H<br />
Figure 38: A linear regression fit between<br />
the quality performance and quality capability<br />
records of suppliers, which delivered products<br />
from the material group 0096 to Plant H and experienced<br />
quality related problems in the time<br />
period between the years 2008 and 2010.<br />
the quality performance records and quality capability scores of the respective suppliers (β 1 ≠ 0).<br />
The positive slope indicates that suppliers, which received higher quality capability evaluation during<br />
an audit, experience a larger number of quality related problems. These results are in contrast<br />
with the initial hypothesis presented in the beginning of this paper, which states that suppliers with<br />
higher capability scores are expected to have better quality performance and thus lower number of<br />
ppm respectively.<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
97
The second set of empirical data analyzed in this section comprises of the quality performance<br />
records of suppliers, which deliver parts from the material group 0337 (”Light metal cast gearbox<br />
parts”) to the Volkswagen production facility Plant C. The data contains the records of 44 different<br />
suppliers, as in this case again around 60% (26 suppliers) of all evaluated supplier records contain<br />
no information about any quality related problems in the respective time period between 2008 and<br />
2010. Note that also here not all suppliers were active during the entire period of evaluation.<br />
Distribution of Ep no ppms<br />
Quality Capability vs PPM<br />
●<br />
Frequency<br />
0 2 4 6 8<br />
Frequency<br />
0 1 2 3 4 5 6<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
85 90 95 100<br />
Distribution of Ep with ppms<br />
85 90 95 100<br />
CDFs for Ep no ppms<br />
and Ep with ppms<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.235<br />
p = 0.5994<br />
n_1 = 26<br />
n_2 = 18<br />
Ep no ppms<br />
Ep with ppms<br />
85 90 95 100<br />
Evaluation Scores [%]<br />
Figure 39: Distribution comparison between<br />
the evaluation scores of suppliers without reported<br />
quality related problems in the series<br />
production (E P no ppms) and suppliers with reported<br />
problems (E P with ppms). Both groups<br />
of suppliers delivered components from the<br />
material group 0337 (”Light metal cast gearbox<br />
parts”) to Plant C.<br />
PPM [log]<br />
1 10 100 1000 10000<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
86 88 90 92 94<br />
Quality Capability Evaluation [%]<br />
Material Group: 0337 (Light metal cast gearbox parts)<br />
Plant: Plant C<br />
Figure 40: A linear regression fit between<br />
the quality performance and quality capability<br />
records of suppliers, which delivered products<br />
from the material group 0337 to Plant C and experienced<br />
quality related problems in the time<br />
period between the years 2008 and 2010.<br />
Figure 39 presents the results of a two-sample Kolmogorov-Smirnov comparison between the<br />
quality capability scores of suppliers with and respectively without ppm, which deliver parts from<br />
the material group 0337 to Plant C. Additionally, for each set of evaluation scores a Shapiro-Wilk<br />
normality test was performed. The resulting Shapiro-Wilk test’s p-values are 0.008 for suppliers<br />
without ppm and 0.031 for suppliers with ppm, indicating statistical significance on the 5%-<br />
significance level in both cases. Based on these p-values it is concluded that both datasets are<br />
non-normally distributed. The Kolmogorov-Smirnov comparison between the two distributions<br />
reveals a particularly high p-value of 0.599, which is above the 5%-significance level, meaning<br />
●<br />
●<br />
●<br />
●<br />
●<br />
98
that the null-hypothesis of the test is not rejected and the two datasets are found to have statistically<br />
similar distributions. These results are in accordance with the accompanying plot of the two<br />
cumulative distribution functions included in Figure 39.<br />
Similar to the previous case, also here a linear regression was used to determine the relation<br />
between the quality performance records (on a decimal logarithmic scale) of suppliers, which were<br />
reported to have quality related problems, and their respective quality capability scores. The results<br />
of the linear regression are presented in Figure 40. The estimated parameters of the least square<br />
fit are accordingly an intercept β 0 = 8.212 (std. error= 9.61083) and a slope β 1 = −0.058 (std.<br />
error= 0.107). The slightly negative slope of the least squares fit suggests that suppliers with<br />
higher audit evaluation score would have less ppm and accordingly better quality performance.<br />
However, a F-test with the null-hypothesis of β 1 = 0 and alternative hypothesis β 1 ≠ 0 was used<br />
to check the significance of the obtained regression parameter estimates and revealed a p-value<br />
of 0.594. The high p-value of the F-test indicates that its null-hypothesis is not rejected on the<br />
5%-significance level, i.e. there is no statistically significant evidence supporting the claim that the<br />
linear fit has a negative slope indeed.<br />
For the second set of empirical data, apart from the comparison between the overall evaluation<br />
scores of suppliers with and respectively without ppm, the individual evaluation components (already<br />
presented in detail in Section 6.1), which comprise a particular audit evaluation, were also<br />
compared. The purpose of these comparisons is to possibly find potential reasons why these two<br />
groups of suppliers perform differently on the level of quality performance, while there is little<br />
difference between their quality capability evaluations. Note that the supplier catalogue, which as<br />
already described in Section 6.3 was the major source of audit evaluation data for the analysis,<br />
provides only the overall evaluation scores and does not contain any information about the individual<br />
evaluation elements of the audit results. For that reason here again the according scores<br />
had to be obtained manually from the audit report hard copies. For each of the resulting datasets a<br />
Shapiro-Wilk normality test was performed. The evaluation results were then compared pairwise<br />
with respect to the individual elements of a quality audit using a two-sample Kolmogorov-Smirnov<br />
test. The test results are presented in Figures 41 and 42 below.<br />
99
Distribution of Ez no ppms<br />
Distribution of Ek no ppms<br />
Frequency<br />
0 2 4 6 8<br />
80 85 90 95 100<br />
Distribution of Ez with ppms<br />
Frequency<br />
0 1 2 3 4<br />
Frequency<br />
0 1 2 3 4 5 6 7<br />
80 85 90 95 100<br />
Distribution of Ek with ppms<br />
Frequency<br />
0 1 2 3 4<br />
80 85 90 95 100<br />
80 85 90 95 100<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Ez no ppms<br />
and Ez with ppms<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.07692<br />
p = 1<br />
n_1 = 26<br />
n_2 = 18<br />
Ez no ppms<br />
Ez with ppms<br />
80 85 90 95 100<br />
Evaluation Scores [%]<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Ek no ppms<br />
and Ek with ppms<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.2009<br />
p = 0.7841<br />
n_1 = 26<br />
n_2 = 18<br />
Ek no ppms<br />
Ek with ppms<br />
80 85 90 95 100<br />
Evaluation Scores [%]<br />
(a) Comparison for the element Purchased Material.<br />
(b) Comparison for the element Customer Care / Customer<br />
Satisfaction.<br />
Distribution of EU1 no ppms<br />
Distribution of EU2 no ppms<br />
Frequency<br />
0 1 2 3 4<br />
85 90 95 100<br />
Frequency<br />
0.0 1.0 2.0 3.0<br />
75 80 85 90 95 100<br />
Distribution of EU1 with ppms<br />
Distribution of EU2 with ppms<br />
Frequency<br />
0 1 2 3 4<br />
85 90 95 100<br />
Frequency<br />
0.0 1.0 2.0 3.0<br />
75 80 85 90 95 100<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for EU1 no ppms<br />
and EU1 with ppms<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.1709<br />
p = 0.9151<br />
n_1 = 26<br />
n_2 = 18<br />
EU1 no ppms<br />
EU1 with ppms<br />
85 90 95 100<br />
Evaluation Scores [%]<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for EU2 no ppms<br />
and EU2 with ppms<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.188<br />
p = 0.8463<br />
n_1 = 26<br />
n_2 = 18<br />
EU2 no ppms<br />
EU2 with ppms<br />
75 80 85 90 95 100<br />
Evaluation Scores [%]<br />
(c) Comparison for the element Personnel / Personal<br />
Qualification.<br />
(d) Comparison for the element Machinery / Equipment.<br />
Figure 41: Distribution comparison between the individual components of the evaluation scores<br />
of suppliers without reported quality related problems in the series production (E P no ppms) and<br />
suppliers with reported problems (E P with ppms). Both groups of suppliers delivered components<br />
from the material group 0337 (”Light metal cast gearbox parts”) to Plant C.<br />
100
Distribution of EU3 no ppms<br />
Distribution of EU4 no ppms<br />
Frequency<br />
0 1 2 3 4 5 6<br />
Frequency<br />
0 2 4 6 8<br />
85 90 95 100<br />
80 85 90 95 100<br />
Distribution of EU3 with ppms<br />
Distribution of EU4 with ppms<br />
Frequency<br />
0 1 2 3 4 5<br />
85 90 95 100<br />
Frequency<br />
0.0 1.0 2.0 3.0<br />
80 85 90 95 100<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for EU3 no ppms<br />
and EU3 with ppms<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.1966<br />
p = 0.8056<br />
n_1 = 26<br />
n_2 = 18<br />
EU3 no ppms<br />
EU3 with ppms<br />
85 90 95 100<br />
Evaluation Scores [%]<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for EU4 no ppms<br />
and EU4 with ppms<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.1624<br />
p = 0.9418<br />
n_1 = 26<br />
n_2 = 18<br />
EU4 no ppms<br />
EU4 with ppms<br />
80 85 90 95 100<br />
Evaluation Scores [%]<br />
(a) Comparison for the element Transport / Parts Handling<br />
/ Storage / Packaging.<br />
(b) Comparison for the element Failure Analysis / Corrective<br />
Measures / Continuous Improvements.<br />
Figure 42: Distribution comparison between the individual components of the evaluation scores<br />
of suppliers without reported quality related problems in the series production (E P no ppms) and<br />
suppliers with reported problems (E P with ppms). Both groups of suppliers delivered components<br />
from the material group 0337 (”Light metal cast gearbox parts”) to Plant C.<br />
The conducted Kolmogorov-Smirnov tests reveal no statistically significant differences at the<br />
5%-significance level also for the individual components of the audit evaluations. In all cases<br />
the resulting p-values are considerably high ranging from 0.784 for the element ”Customer Care/-<br />
Customer Satisfaction” to 1.000 for the element ”Suppliers / Purchased Materials” 18 , for which<br />
element the audit scores of the two groups of suppliers are practically identical. A look on the<br />
cumulative distribution functions of each pair of datasets confirms the results of the Kolmogorov-<br />
Smirnov test. In almost all cases the compared pairs of distributions overlap completely. The only<br />
case in which the suppliers without ppm possibly have an advantage in terms of evaluation scores<br />
is for the element ”Customer Care/Customer Satisfaction”, where their distribution shows slightly<br />
better evaluation results up to the 91%-mark (Figure 41b). This is probably due to the fact that<br />
during a quality audit, the quality performance, which a supplier has shown up to the time of the<br />
audit, is also considered in the final evaluation.<br />
18 The names of the respective audit evaluation elements are used here as they appear in the Formel Q-Fähigkeit 6.0<br />
(2009). In the currently valid Formel Q-Fähigkeit 7.0 these may vary.<br />
101
7.4 Relation Between Number of Defective Parts and Delivery Amount<br />
The two case studies presented in Section 7.3 above do not provide any statistically significant<br />
evidence that suppliers, which have quality related problems expressed in terms of ppm, are rated<br />
differently in their quality capability evaluations than suppliers without any reported quality problems.<br />
The reader is reminded that the conducted evaluations are based on empirical data, which<br />
account for a number of influencing factors presented in Section 7.2. Therefore, these particular biasing<br />
factors could have little influence on the observed non-correlation between the two statistical<br />
quantities.<br />
Nevertheless, there is one particular indicator (other than ppm), based on which in both of the<br />
presented case studies suppliers with ppm records turn out to be very different from suppliers without<br />
quality problems. This is namely the number of components they delivered. Figure 43 presents<br />
the results of a two-sample Kolmogorov-Smirnov test, which compares the delivery amounts (on<br />
a decimal logarithmic scale) of the two groups of suppliers in each of the above cases. The test<br />
Distribution of Records w/o ppm (MG 0096/Plant H)<br />
Distribution of Records w/o ppm (MG 0337/Plant C)<br />
Frequency<br />
0 1 2 3 4 5<br />
3.5 4.0 4.5 5.0 5.5 6.0 6.5<br />
Frequency<br />
0 1 2 3 4 5<br />
2 3 4 5 6<br />
Distribution of Records with ppm (MG 0096/Plant H)<br />
Distribution of Records with ppm (MG 0337/Plant C)<br />
Frequency<br />
0 1 2 3 4 5<br />
3.5 4.0 4.5 5.0 5.5 6.0 6.5<br />
Frequency<br />
0 2 4 6 8<br />
2 3 4 5 6<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Records w/o ppm (MG 0096/Plant H)<br />
and Records with ppm (MG 0096/Plant H)<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.5598<br />
p = 0.009848<br />
n_1 = 18<br />
n_2 = 13<br />
Records w/o ppm (MG 0096/Plant H)<br />
Records with ppm (MG 0096/Plant H)<br />
3.5 4.0 4.5 5.0 5.5 6.0 6.5<br />
Delivery Amount [log]<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Records w/o ppm (MG 0337/Plant C)<br />
and Records with ppm (MG 0337/Plant C)<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.9444<br />
p = 1.149e−08<br />
n_1 = 26<br />
n_2 = 18<br />
Records w/o ppm (MG 0337/Plant C)<br />
Records with ppm (MG 0337/Plant C)<br />
2 3 4 5 6<br />
Delivery Amount [log]<br />
(a) Suppliers, which delivered parts from the material<br />
group 0096 to Plant H.<br />
(b) Suppliers, which delivered parts from the material<br />
group 0337 to Plant C.<br />
Figure 43: Comparison between the delivery amounts of suppliers with and without reported quality<br />
related problems.<br />
102
esults for material group 0096 are presented in Figure 43a and the test results for material group<br />
0337 – in Figure 43b respectively. The resulting p-values of the Kolmogorov-Smirnov test are<br />
0.010 in the case of material group 0096 and 1.15E-8 for material group 0337. Both p-values are<br />
below the 5%-significance level meaning that the null-hypothesis of the Kolmogorov-Smirnov test<br />
assuming similarity between the distributions of the compared data is rejected, as in the case of<br />
material group 0337 the especially low p-value indicates particularly large difference between the<br />
two distributions. The Kolmogorov-Smirnov test results are in accordance also with the accompanying<br />
cumulative distribution function plots in Figure 43. For suppliers which deliver components<br />
from the material group 0337 the ranges of the delivery amounts of the two groups barely overlap.<br />
The results presented in Figure 43 were not anticipated in the beginning of this project. However,<br />
given the fact that they identify a second important factor, which differentiates between suppliers<br />
with good quality performance and suppliers with quality related problems, this finding gives<br />
significant information, which can be used to improve the effectiveness of the quality auditing process<br />
employed by Volkswagen to manage the production quality down its supply chain. Therefore<br />
the topic was further investigated and the respective analytic results are presented in the current<br />
section.<br />
At this point it is important to answer the following questions: What is the relation between<br />
the number of components a supplier delivered and the amount of ppm it scored? And what information<br />
does this relation give about the quality capability of supplier’s production process?<br />
To provide an answer to these questions here first a mathematical approach is used to derive the<br />
relation between the delivery amount and ppm. Subsequently a number of empirical data sets are<br />
analyzed and the results are discussed with respect to the theoretical derivation.<br />
Let k ∈ N be the number of defective parts in a batch of N components. Then the amount of<br />
ppm in this batch is defined as follows:<br />
ppm = F (k, N) = k N × 106 (6)<br />
Note that ppm ∈ Q + . Let the event X refer to the absolute number of defective components k<br />
a supplier delivers in a batch of N components, and the event Y refer to the fraction of defective<br />
components in the same batch expressed in ppm. Further, let p = const 19 be the probability that<br />
19 It is important to keep in mind that the quality capability of a particular production process and accordingly the<br />
probability to produce functional parts can vary over time. In this calculation it is assumed that the supplier has a stable<br />
process over the time period in which the N components are produced and the according probability for no defects or<br />
defects accordingly remains constant.<br />
103
the supplier produces a functional part, and accordingly q = 1−p the probability that the produced<br />
part is defective. Note that according to this definition q is the failure rate of supplier’s production<br />
process. Under this assumption the event X would have a binomial distribution with N trials and<br />
success rate q, i.e. X ∼ B(N, q). Furthermore, the events X and Y are related as follows:<br />
Y = g(X) = βX, where β = 106<br />
N<br />
(7)<br />
Now let Ω Y = {kβ | k ∈ [0; N]} be the domain of all possible discrete values of ppm’s for a batch<br />
of N parts. The expected ppm-value for suppliers which have ppm (quality related problems) can<br />
be written as:<br />
E [Y | Y > 0] =<br />
∑<br />
y P (Y = y | Y > 0) (8)<br />
y ∈ Ω Y<br />
(<br />
P (Y = y | Y > 0) = P (βX = y | βX > 0) = P X = y )<br />
β ∣ X > 0 (9)<br />
(<br />
(<br />
P X = y ) P X > 0<br />
β ∣ X > 0 ∣ X = y ) (<br />
P X = y )<br />
β<br />
β<br />
=<br />
(10)<br />
P (X > 0)<br />
Using (10) one can write (8) as follows:<br />
E [Y | Y > 0] =<br />
∑<br />
y<br />
y ∈ Ω Y<br />
= 1, for y > 0<br />
and 0 otherwise<br />
{ ( }} {<br />
P X > 0<br />
∣ X = y β<br />
P (X > 0)<br />
)<br />
P<br />
(<br />
X = y )<br />
β<br />
(11)<br />
N∑<br />
E [Y | Y > 0] = kβ<br />
k = 1<br />
P (X = k)<br />
P (X > 0)<br />
(12)<br />
N∑<br />
E [Y | Y > 0] = kβ<br />
k = 1<br />
P (X = k)<br />
1 − P (X = 0)<br />
Now using the fact that X has a binomial distribution, the above equation is reduced to:<br />
N∑<br />
E [Y | Y > 0] = kβ<br />
k = 1<br />
( N<br />
k<br />
)<br />
q k p N−k<br />
1 − p N = β<br />
1 − p N N ∑<br />
k = 1<br />
( N<br />
k<br />
)<br />
kq k p N−k =<br />
(13)<br />
βNq<br />
1 − p N (14)<br />
104
E [Y | Y > 0] =<br />
q<br />
1 − p N × 106 (15)<br />
Analogically one can find the conditional expectation for the logarithm of the obtained ppm, only<br />
for suppliers with ppm, i.e. E [Z | Y > 0], where<br />
Z = log 10 (βX), with β = 106<br />
N<br />
(16)<br />
Since here only suppliers, which have ppm are of interest, the domain of the event Z is defined as<br />
Ω Z = {log 10 (kβ) | k ∈ [1; N]}.<br />
E [Z | Y > 0] =<br />
∑<br />
z P (Z = z | Y > 0) (17)<br />
z ∈ Ω Z<br />
P (Z = z | Y > 0) = P (log 10 (βX) = z | βX > 0) = P (βX = 10 z | X > 0) (18)<br />
(<br />
)<br />
)<br />
)<br />
P (Z = z | Y > 0) = P<br />
(X = 10z<br />
P X > 0<br />
β ∣ X > 0 ∣ X = 10z P<br />
(X = 10z<br />
β<br />
β<br />
=<br />
P (X > 0)<br />
(<br />
)<br />
) (19)<br />
E [Z | Y > 0] =<br />
∑ P X > 0<br />
∣ X = 10z P<br />
(X = 10z<br />
β<br />
β<br />
z<br />
(20)<br />
1 − P (X = 0)<br />
z ∈ Ω Z<br />
N∑<br />
E [Z | Y > 0] = log 10 (kβ)<br />
k = 1<br />
( ) N<br />
q k p N−k<br />
k<br />
N∑<br />
E [Z | Y > 0] = log 10 (kβ)<br />
k = 1<br />
= 1, for k > 0<br />
{ }} {<br />
P (X > 0 | X = k ) P (X = k)<br />
1 − P (X = 0)<br />
(21)<br />
N∑<br />
( ) ( ) k N q k<br />
= log<br />
1 − p N 10<br />
N × p N−k<br />
106 k 1 − p (22) N<br />
k = 1<br />
The resulting expressions in (15) and (22) were numerically evaluated for several different<br />
values of p with N ∈ [1; 10 6 ] (see Figure 44). As expected the two plots look very similar. Each of<br />
the curves is asymptotically limited by two straight lines (on a log-log plot). For a small number<br />
of delivered components the according curves are limited by the minimum number of ppm, while<br />
for a large number of delivered components the limiting line is y = log 10 (q10 6 ) – dependent only<br />
on the actual failure rate q of the respective production process.<br />
The plots have a straightforward practical interpretation. For a sufficiently small number of<br />
produced components in clear probabilistic terms the expected number of defective components<br />
105
E[ppm|ppm>0][log]<br />
1e+00 1e+02 1e+04 1e+06<br />
p = 70%<br />
y = log10(30% x 10^6)<br />
p = 90%<br />
y = log10(10% x 10^6)<br />
p = 99%<br />
y = log10(1% x 10^6)<br />
p = 99.9%<br />
y = log10(0.1% x 10^6)<br />
1e+00 1e+02 1e+04 1e+06<br />
Delivered Components N [log]<br />
E[log(ppm)|ppm]<br />
1 2 3 4 5 6<br />
●<br />
●<br />
●<br />
●<br />
p = 70%<br />
y = log10(30% x 10^6)<br />
● ●● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●<br />
●<br />
●<br />
●<br />
p = 90%<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
● ● p = 99%<br />
●<br />
●<br />
●<br />
● ● ●<br />
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ● ● ● ● ● ● ● ●<br />
●<br />
●<br />
y = log10(10% x 10^6)<br />
y = log10(1% x 10^6)<br />
p = 99.9%<br />
y = log10(0.1% x 10^6)<br />
1e+00 1e+02 1e+04 1e+06<br />
Delivered Components N [log]<br />
(a) Expectation of the event Y .<br />
(b) Expectation of the event Z.<br />
Figure 44: Numerical evaluation of the conditional expectations of the events Y and Z 1 , plotted<br />
for different values of p. The dashed line in both cases represents the minimal number of ppm for<br />
the according delivery amount N, which can be calculated using equation (6) for k = 1 2 .<br />
1 Note that due to the complex form of (22), the according numerical evaluation of the equation was limited by the<br />
floating point precision of the R software and therefore the respective simulation does not cover the entire range of the<br />
number of delivered components. Nevertheless, the key characteristics of the plotted curves are still discernible.<br />
2 Since both axises are on a logarithmic scale, the hyperbola described by (6) appears as a straight line on the plot<br />
G(N) = log 10 (F (k = 1, N)) = log 10<br />
( 1<br />
N 106 )<br />
= log 10 10 6 − log 10 N = 6 − log 10 N<br />
remains below 1. For that reason, for small N it is expected that, if any defects occur at all,<br />
they will be rather few and the resulting amount of ppm would be very close to the minimum<br />
possible number. In other words for small number of components a supplier is ”lucky” most of<br />
the time, i.e. has zero ppm. On the other hand, as the production volume increases and reaches a<br />
particular critical value (N crit = 1/q), even for very small failure rates q in the production process<br />
the expected number of defective components becomes larger than 1. Thus for a large number<br />
of components the supplier cannot escape his ppm value by luck any more, since the probability<br />
of defective components to occur is overwhelming due to the large number of produced parts.<br />
These results are a possible explanation why in the two cases discussed above suppliers, which<br />
lack quality related problems, tend to have smaller number of delivered components as compared<br />
to suppliers with reported ppm.<br />
106
Note that the critical amount of delivered components N crit for a given failure rate q can easily<br />
be determined graphically – it is the x-coordinate of the intersection point of the two asymptotes<br />
(see Figure 44). Note also that the larger the failure rate q of the process, the smaller the number of<br />
produced components N crit has to be to get the first defective parts. For N larger than the critical<br />
delivery amount N crit the number of ppm, which a particular supplier has, would be limited by the<br />
reliability of its production process.<br />
Automotive producers have particularly stringent quality requirements and expect their suppliers<br />
to manage the failure rates of their production processes at levels of less than several hundred<br />
ppm, meaning that in most of the cases the failure rate q is of the order of 0.01% and even less.<br />
At such small failure rates, the probability to produce defective components becomes practically<br />
significant for delivery amounts N crit = 1/q of the order of 10.000 or even more delivered components.<br />
In order to verify whether these observations hold true also for other areas of the supply<br />
chain, a two-sample Kolmogorov-Smirnov test was performed on a number of additional sets of<br />
empirical data. The results of the test are presented in Figures 45 through 47 below.<br />
Distribution of Records w/o ppm (MG 0068/Plant K)<br />
Distribution of Records w/o ppm (MG 0068/Plant L)<br />
Frequency<br />
0 1 2 3 4 5 6<br />
3 4 5 6<br />
Distribution of Records with ppm (MG 0068/Plant K)<br />
Frequency<br />
0 2 4 6 8 10 12<br />
Frequency<br />
0 5 10 15<br />
1 2 3 4 5 6 7<br />
Distribution of Records with ppm (MG 0068/Plant L)<br />
Frequency<br />
0 2 4 6 8 10<br />
3 4 5 6<br />
1 2 3 4 5 6 7<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Records w/o ppm (MG 0068/Plant K)<br />
and Records with ppm (MG 0068/Plant K)<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.4493<br />
p = 0.002056<br />
n_1 = 29<br />
n_2 = 35<br />
Records w/o ppm (MG 0068/Plant K)<br />
Records with ppm (MG 0068/Plant K)<br />
3 4 5 6<br />
Delivery Amount [log]<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Records w/o ppm (MG 0068/Plant L)<br />
and Records with ppm (MG 0068/Plant L)<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.7058<br />
p = 2.167e−10<br />
n_1 = 61<br />
n_2 = 37<br />
Records w/o ppm (MG 0068/Plant L)<br />
Records with ppm (MG 0068/Plant L)<br />
1 2 3 4 5 6 7<br />
Delivery Amount [log]<br />
(a) Suppliers, which delivered parts from the material<br />
group 0068 to Plant K.<br />
(b) Suppliers, which delivered parts from the material<br />
group 0068 to Plant L.<br />
Figure 45: Comparison between the delivery amounts of suppliers with and without reported quality<br />
related problems.<br />
107
Distribution of Records w/o ppm (MG 0096/Plant F)<br />
Distribution of Records w/o ppm (MG 0099/Plant L)<br />
Frequency<br />
0 1 2 3 4 5<br />
2 3 4 5 6<br />
Distribution of Records with ppm (MG 0096/Plant F)<br />
Frequency<br />
0 2 4 6 8<br />
Frequency<br />
0 1 2 3 4 5 6 7<br />
1 2 3 4 5 6<br />
Distribution of Records with ppm (MG 0099/Plant L)<br />
Frequency<br />
0 2 4 6 8 10<br />
2 3 4 5 6<br />
1 2 3 4 5 6<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Records w/o ppm (MG 0096/Plant F)<br />
and Records with ppm (MG 0096/Plant F)<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.7514<br />
p = 7.25e−07<br />
n_1 = 29<br />
n_2 = 24<br />
Records w/o ppm (MG 0096/Plant F)<br />
Records with ppm (MG 0096/Plant F)<br />
2 3 4 5 6<br />
Delivery Amount [log]<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Records w/o ppm (MG 0099/Plant L)<br />
and Records with ppm (MG 0099/Plant L)<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.4883<br />
p = 0.002592<br />
n_1 = 31<br />
n_2 = 22<br />
Records w/o ppm (MG 0099/Plant L)<br />
Records with ppm (MG 0099/Plant L)<br />
1 2 3 4 5 6<br />
Delivery Amount [log]<br />
(a) Suppliers, which delivered parts from the material<br />
group 0096 to Plant F.<br />
(b) Suppliers, which delivered parts from the material<br />
group 0099 to Plant L.<br />
Distribution of Records w/o ppm (MG 0105/Plant D)<br />
Distribution of Records w/o ppm (MG 0302/Plant F)<br />
Frequency<br />
0 5 10 20 30<br />
1 2 3 4 5 6<br />
Distribution of Records with ppm (MG 0105/Plant D)<br />
Frequency<br />
0 2 4 6 8 10<br />
Frequency<br />
0 2 4 6 8<br />
2 3 4 5 6<br />
Distribution of Records with ppm (MG 0302/Plant F)<br />
Frequency<br />
0 1 2 3 4 5 6 7<br />
1 2 3 4 5 6<br />
2 3 4 5 6<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Records w/o ppm (MG 0105/Plant D)<br />
and Records with ppm (MG 0105/Plant D)<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.6792<br />
p = 1.287e−10<br />
n_1 = 167<br />
n_2 = 30<br />
Records w/o ppm (MG 0105/Plant D)<br />
Records with ppm (MG 0105/Plant D)<br />
1 2 3 4 5 6<br />
Delivery Amount [log]<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Records w/o ppm (MG 0302/Plant F)<br />
and Records with ppm (MG 0302/Plant F)<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.6387<br />
p = 2.496e−05<br />
n_1 = 25<br />
n_2 = 31<br />
Records w/o ppm (MG 0302/Plant F)<br />
Records with ppm (MG 0302/Plant F)<br />
2 3 4 5 6<br />
Delivery Amount [log]<br />
(c) Suppliers, which delivered parts from the material<br />
group 0105 to Plant D.<br />
(d) Suppliers, which delivered parts from the material<br />
group 0302 to Plant F.<br />
Figure 46: Comparison between the delivery amounts of suppliers with and without reported quality<br />
related problems.<br />
108
Distribution of Records w/o ppm (MG 0480/Plant F)<br />
Distribution of Records w/o ppm (MG 0485/Plant C)<br />
Frequency<br />
0 1 2 3 4 5 6<br />
1 2 3 4 5 6<br />
Distribution of Records with ppm (MG 0480/Plant F)<br />
Frequency<br />
0 2 4 6 8 10<br />
Frequency<br />
0 1 2 3 4 5 6<br />
3 4 5 6 7<br />
Distribution of Records with ppm (MG 0485/Plant C)<br />
Frequency<br />
0 2 4 6 8<br />
1 2 3 4 5 6<br />
3 4 5 6 7<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Records w/o ppm (MG 0480/Plant F)<br />
and Records with ppm (MG 0480/Plant F)<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.6877<br />
p = 2.603e−06<br />
n_1 = 25<br />
n_2 = 26<br />
Records w/o ppm (MG 0480/Plant F)<br />
Records with ppm (MG 0480/Plant F)<br />
1 2 3 4 5 6<br />
Delivery Amount [log]<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Records w/o ppm (MG 0485/Plant C)<br />
and Records with ppm (MG 0485/Plant C)<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.7797<br />
p = 2.597e−08<br />
n_1 = 30<br />
n_2 = 23<br />
Records w/o ppm (MG 0485/Plant C)<br />
Records with ppm (MG 0485/Plant C)<br />
3 4 5 6 7<br />
Delivery Amount [log]<br />
(a) Suppliers, which delivered parts from the material<br />
group 0480 to Plant F.<br />
(b) Suppliers, which delivered parts from the material<br />
group 0485 to Plant C.<br />
Figure 47: Comparison between the delivery amounts of suppliers with and without reported quality<br />
related problems.<br />
In each of the above cases the p-values of the respective Kolmogorov-Smirnov test provide<br />
strong evidence for statistical significance on the 5%-significance level. Based on the small p-<br />
values the null-hypothesis of the test is rejected and in every single case it is concluded that the<br />
distribution of the number of delivered components of suppliers with ppm is different from the distribution<br />
of the number of delivered components of suppliers without ppm. Furthermore, the plots<br />
with the cumulative distribution functions of each pair of empirical data systematically show that<br />
suppliers which experience quality related problems deliver more components than suppliers without<br />
any reported problems. These results are identical with the situation observed in the other two<br />
cases discussed previously in this section as well as in accordance with the theoretical derivation.<br />
The second part of the analysis presented in this section investigates the relation between the<br />
amount of components a certain supplier delivered and the reported number of defective parts per<br />
million. The theoretical derivation above revealed that the relation between the two quantities is<br />
asymptotically limited by two straight lines on a log-log scale for a given failure rate. This is why<br />
in each of the following cases a linear regression between the logarithm of the reported ppm and<br />
109
the logarithm of the respective delivery quantities was carried out. The corresponding results are<br />
presented in Figures 48 through 50 below.<br />
Cummulative Delivery Quantity vs Cummulative PPM<br />
Cummulative Delivery Quantity vs Cummulative PPM<br />
Cummulative PPM [log]<br />
5 10 50 100 500 5000<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
Least Squares fit<br />
Minimal number of ppms<br />
MG 0068 | Plant K<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
Cummulative PPM [log]<br />
5 10 20 50 100 200 500 1000 2000 5000<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
Least Squares fit<br />
Minimal number of ppms<br />
MG 0068 | Plant L<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
2e+04 5e+04 2e+05 5e+05 2e+06 5e+06<br />
Cummulative Delivery Quantity [log]<br />
(a) Suppliers, which delivered parts from the material<br />
group 0068 to Plant K.<br />
5e+03 2e+04 1e+05 5e+05 2e+06<br />
Cummulative Delivery Quantity [log]<br />
(b) Suppliers, which delivered parts from the material<br />
group 0068 to Plant L.<br />
Cummulative PPM [log]<br />
1 10 100 1000 10000<br />
Cummulative Delivery Quantity vs Cummulative PPM<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
Least Squares fit<br />
Minimal number of ppms<br />
MG 0096 | Plant F<br />
●<br />
●<br />
●<br />
●<br />
Cummulative PPM [log]<br />
20 50 100 200<br />
Cummulative Delivery Quantity vs Cummulative PPM<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
Least Squares fit<br />
Minimal number of ppms<br />
MG 0096 | Plant H<br />
●<br />
●<br />
5e+03 2e+04 1e+05 5e+05 2e+06<br />
Cummulative Delivery Quantity [log]<br />
2e+04 5e+04 2e+05 5e+05 2e+06<br />
Cummulative Delivery Quantity [log]<br />
(c) Suppliers, which delivered parts from the material<br />
group 0096 to Plant F.<br />
(d) Suppliers, which delivered parts from the material<br />
group 0096 to Plant H.<br />
Figure 48: Linear regression between the delivery amount of suppliers with reported quality problems<br />
and their ppm record. Please note that both axises are on a logarithmic scale.<br />
110
●<br />
Cummulative Delivery Quantity vs Cummulative PPM<br />
Cummulative Delivery Quantity vs Cummulative PPM<br />
Cummulative PPM [log]<br />
1 10 100 1000 10000<br />
●<br />
●<br />
●<br />
●<br />
●<br />
Least Squares fit<br />
Minimal number of ppms<br />
MG 0099 | Plant L<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
Cummulative PPM [log]<br />
100 200 500 1000 2000 5000 10000 20000<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
Least Squares fit<br />
Minimal number of ppms<br />
MG 0105 | Plant D<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
2e+03 1e+04 5e+04 2e+05 1e+06<br />
Cummulative Delivery Quantity [log]<br />
(a) Suppliers, which delivered parts from the material<br />
group 0099 to Plant L.<br />
1e+04 5e+04 2e+05 5e+05 2e+06 5e+06<br />
Cummulative Delivery Quantity [log]<br />
(b) Suppliers, which delivered parts from the material<br />
group 0105 to Plant D.<br />
Cummulative Delivery Quantity vs Cummulative PPM<br />
Cummulative Delivery Quantity vs Cummulative PPM<br />
●<br />
●<br />
Cummulative PPM [log]<br />
5 10 50 100 500 5000<br />
●<br />
●<br />
●<br />
●<br />
●<br />
Least Squares fit<br />
Minimal number of ppms<br />
MG 0302 | Plant F<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
Cummulative PPM [log]<br />
1 10 100 1000 10000<br />
●<br />
●<br />
●<br />
Least Squares fit<br />
Minimal number of ppms<br />
MG 0480 | Plant F<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ● ●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
1e+02 1e+03 1e+04 1e+05 1e+06<br />
Cummulative Delivery Quantity [log]<br />
(c) Suppliers, which delivered parts from the material<br />
group 0302 to Plant F.<br />
1e+02 1e+03 1e+04 1e+05 1e+06<br />
Cummulative Delivery Quantity [log]<br />
(d) Suppliers, which delivered parts from the material<br />
group 0480 to Plant F.<br />
Figure 49: Linear regression between the delivery amount of suppliers with reported quality problems<br />
and their ppm record. Please note that both axises are on a logarithmic scale.<br />
111
●<br />
Cummulative Delivery Quantity vs Cummulative PPM<br />
Cummulative Delivery Quantity vs Cummulative PPM<br />
●<br />
●<br />
●<br />
●<br />
Cummulative PPM [log]<br />
1 100 10000<br />
●<br />
●<br />
●<br />
●<br />
●<br />
Least Squares fit<br />
Minimal number of ppms<br />
MG 0485 | Plant C<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
Cummulative PPM [log]<br />
1 10 100 1000 10000<br />
●<br />
Least Squares fit<br />
Minimal number of ppms<br />
MG 0337 | Plant C<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
1e+05 2e+05 5e+05 1e+06 2e+06 5e+06<br />
Cummulative Delivery Quantity [log]<br />
(a) Suppliers, which delivered parts from the material<br />
group 0485 to Plant C.<br />
5e+03 2e+04 5e+04 2e+05 5e+05 2e+06<br />
Cummulative Delivery Quantity [log]<br />
(b) Suppliers, which delivered parts from the material<br />
group 0337 to Plant C.<br />
Figure 50: Linear regression between the delivery amount of suppliers with reported quality problems<br />
and their ppm record. Please note that both axises are on a logarithmic scale.<br />
112
Table 4: Summary of the performed linear regression analysis between suppliers’ quality performance<br />
and delivery quantity (both on a log scale).<br />
Material<br />
Group<br />
Plant ID<br />
Sample<br />
Size N<br />
Intersect Estimate β 0<br />
(Standard Error)<br />
Slope Estimate β 1<br />
(Standard Error)<br />
F-test p-value<br />
0068 K 35 6.963 (1.625) −0.833 (0.287) 0.007<br />
0068 L 37 2.830 (1.221) −0.089 (0.219) 0.686<br />
0096 F 24 5.991 (1.129) −0.820 (0.200) 4.7 × 10 −4<br />
0096 H 13 2.745 (0.940) −0.130 (0.170) 0.463<br />
0099 L 22 7.198 (1.625) −0.911 (0.309) 0.008<br />
0105 D 30 5.314 (0.716) −0.404 (0.138) 0.007<br />
0302 F 31 4.613 (0.696) −0.537 (0.130) 2.7 × 10 −4<br />
0337 C 18 2.339 (3.442) 0.114 (0.604) 0.852<br />
0480 F 26 5.100 (0.604) −0.623 (0.114) 1.3 × 10 −5<br />
0485 C 23 2.347 (3.996) 0.072 (0.656) 0.914<br />
Theoretical values<br />
vertical asymptote (N < N crit ) 6.000 −1.000<br />
horizontal asymptote (N > N crit ) log 10 (q10 6 ) 0.000<br />
Table 4 presents a summary of the estimated parameters of the linear fit in each of the above<br />
ten cases. For each set of suppliers which deliver parts from a particular material group to a certain<br />
Volkswagen production facility the table provides the sample size N as well as the estimated<br />
intercept β 0 and accordingly slope β 1 of the least squares linear fit. The numbers included in<br />
the brackets represent the standard error of each of the estimated parameters. These values were<br />
obtained using the R function summary(). In addition, summary() provides the results of<br />
an F-test, which evaluates the reliability of the slope estimate. The performed F-test in each of<br />
the cases uses the null-hypothesis that the linear fit has a zero slope (β 1<br />
= 0) and alternative<br />
hypothesis of β 1 ≠ 0. The p-values of the performed F-test are listed in the last column in the table<br />
above. In six of the cases the resulting p-values of the F-test are below the 5%-significance level<br />
providing support to reject the null-hypothesis and accept that the slope of the linear fit is indeed<br />
non-zero. In all of these cases the estimated slopes of the least squared fit are particularly high. In<br />
the remaining four cases the p-values of the F-test are above the 5%-significance level, meaning<br />
that the null-hypothesis of the F-test cannot be rejected and the according slope estimates, even<br />
though in most of the cases negative, are not significant. In the latter cases the slope values are also<br />
substantially smaller, as in the case of suppliers which deliver components from material groups<br />
0337 and 0485 to Plant C the estimated slopes are even positive. To aid the comparison between<br />
the parameter estimates in Table 4 they are presented also graphically in Figure 51.<br />
113
Slope (β_1)<br />
-1,5 -1,0 -0,5 0,0 0,5 1,0<br />
10<br />
8<br />
0099; Plant L<br />
0068; Plant K<br />
Zero Slope<br />
(N > N_crit)<br />
Intercept (β_0)<br />
6<br />
4<br />
2<br />
Theoretical<br />
value (min ppm)<br />
0096; Plant F<br />
0105; Plant D<br />
0068; Plant L<br />
0480; Plant F 0302; Plant F<br />
0096; Plant H<br />
0337; Plant C<br />
0485; Plant C<br />
0<br />
-2<br />
-4<br />
Figure 51: Graphical representation of the estimated linear regression parameters presented in<br />
Table 4. The error bars denote the standard error of the estimates. The label of each data point<br />
consists of the respective Material group; Plant ID. A list of all material groups, which appear in<br />
this paper, can be found in the Appendix.<br />
For each of the performed linear regressions above the influence of the individual data points<br />
on the estimated parameters was evaluated using Cook’s distance, which measures the change<br />
of the estimated linear model parameters induced by the removal of the respective point from the<br />
dataset (Cook, 1977, 1979). For every set of data a plot was generated, which displays the residuals<br />
versus their leverage. Each of these plots includes also two dashed pairs of lines, representing a<br />
Cook’s distance of 0.5 and 1.0 respectively. These plots were used to identify points, which have a<br />
particularly strong influence on the estimated parameters, and thus could lead to serious offsets of<br />
the linear fit. For three of the presented cases above these evaluations revealed indeed that there are<br />
points, which have particularly strong influence on the parameter estimates, with Cook’s distance<br />
of more than 0.5. The plots for the respective three cases are presented in Figure 52. Due to their<br />
strong influence on the data parameters, the respective data points were consequently excluded<br />
from the datasets and a new linear regression based on the remaining data points was carried out.<br />
The new linear fits are presented in Figures 53 and 54. The resulting new plots of residuals versus<br />
points’ leverage are also included alongside the linear regression plots.<br />
114
Residuals vs Leverage<br />
Residuals vs Leverage<br />
Standardized residuals<br />
−2 −1 0 1 2<br />
● ●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
● ●<br />
●<br />
●<br />
●<br />
●<br />
3<br />
Cook's distance<br />
●<br />
13<br />
0.0 0.1 0.2 0.3 0.4 0.5<br />
Leverage<br />
lm(log10(ppm.1) ~ log10(liefermenge.1))<br />
9<br />
●<br />
1<br />
0.5<br />
0.5<br />
1<br />
Standardized residuals<br />
−2 −1 0 1 2<br />
●<br />
14<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
4<br />
Cook's distance 1<br />
0.00 0.05 0.10 0.15 0.20 0.25 0.30<br />
Leverage<br />
lm(log10(ppm.1) ~ log10(liefermenge.1))<br />
●<br />
8<br />
●<br />
1<br />
0.5<br />
0.5<br />
(a) Suppliers, which delivered parts from the material<br />
group 0096 to Plant F.<br />
(b) Suppliers, which delivered parts from the material<br />
group 0099 to Plant L.<br />
Residuals vs Leverage<br />
●<br />
●<br />
Standardized residuals<br />
−2 −1 0 1<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
18<br />
13<br />
●<br />
1<br />
0.5<br />
0.5<br />
1<br />
●<br />
15<br />
Cook's distance<br />
0.0 0.2 0.4 0.6 0.8<br />
Leverage<br />
lm(log10(ppm.1) ~ log10(liefermenge.1))<br />
(c) Suppliers, which delivered parts from the material<br />
group 0337 to Plant C.<br />
Figure 52: Datasets, which include data points with particularly large leverage. The leverage each<br />
data point has on the estimated parameters of the linear regression was determined by computing<br />
its Cook’s distance.<br />
115
●<br />
Cummulative Delivery Quantity vs Cummulative PPM<br />
Residuals vs Leverage<br />
Cummulative PPM [log]<br />
2 5 10 20 50 100 200 500<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
Least Squares fit<br />
Minimal number of ppms<br />
●<br />
MG 0096 | Plant F / Excluded Points: 9<br />
1e+05 2e+05 5e+05 1e+06 2e+06 5e+06<br />
Cummulative Delivery Quantity [log]<br />
Standardized residuals<br />
−2 −1 0 1 2<br />
●<br />
●<br />
●<br />
Cook's distance<br />
●<br />
13<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
0.00 0.05 0.10 0.15 0.20<br />
Leverage<br />
lm(log10(ppm.1) ~ log10(liefermenge.1))<br />
3<br />
●<br />
18<br />
●<br />
●<br />
0.5<br />
0.5<br />
(a) Suppliers, which delivered parts from the material<br />
group 0096 to Plant F.<br />
Cummulative Delivery Quantity vs Cummulative PPM<br />
(b) Leverage of the individual data points for suppliers,<br />
which delivered parts from the material group 0096 to<br />
Plant F.<br />
Residuals vs Leverage<br />
Cummulative PPM [log]<br />
1 10 100 1000 10000<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
Least Squares fit<br />
Minimal number of ppms<br />
MG 0099 | Plant L / Excluded Points: 8<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
5e+03 2e+04 1e+05 5e+05 2e+06<br />
Cummulative Delivery Quantity [log]<br />
Standardized residuals<br />
−2 −1 0 1 2<br />
●<br />
13<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
Cook's distance<br />
4<br />
●<br />
●<br />
●<br />
17<br />
●<br />
0.0 0.1 0.2 0.3 0.4<br />
Leverage<br />
lm(log10(ppm.1) ~ log10(liefermenge.1))<br />
1<br />
0.5<br />
0.5<br />
1<br />
(c) Suppliers, which delivered parts from the material<br />
group 0099 to Plant L.<br />
(d) Leverage of the individual data points for suppliers,<br />
which delivered parts from the material group 0099 to<br />
Plant L.<br />
Figure 53: Recomputed linear regressions, excluding the influencing data points.<br />
116
●<br />
Cummulative Delivery Quantity vs Cummulative PPM<br />
Residuals vs Leverage<br />
0.5<br />
●<br />
●<br />
●<br />
Cummulative PPM [log]<br />
1 10 100 1000 10000<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
Least Squares fit<br />
Minimal number of ppms<br />
●<br />
MG 0337 | Plant C / Excluded Points: 13<br />
500000 1000000 1500000<br />
Cummulative Delivery Quantity [log]<br />
●<br />
●<br />
●<br />
Standardized residuals<br />
−2 −1 0 1<br />
Cook's distance<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
14<br />
10<br />
3●<br />
0.00 0.05 0.10 0.15 0.20 0.25<br />
●<br />
Leverage<br />
lm(log10(ppm.1) ~ log10(liefermenge.1))<br />
●<br />
0.5<br />
1<br />
(a) Suppliers, which delivered parts from the material<br />
group 0337 to Plant C.<br />
(b) Leverage of the individual data points for suppliers,<br />
which delivered parts from the material group 0337 to<br />
Plant C.<br />
Figure 54: Recomputed linear regressions, excluding the influencing data points.<br />
117
Table 5: Summary of the performed linear regression analysis between suppliers’ quality performance<br />
and delivery quantity (both on a log scale) for the corrected datasets.<br />
Material<br />
Group<br />
Plant ID<br />
Sample<br />
Size N<br />
Intersect Estimate β 0<br />
(Standard Error)<br />
Slope Estimate β 1<br />
(Standard Error)<br />
F-test p-value<br />
0096 F 24 (old) 5.991 (1.129) −0.820 (0.200) 4.7 × 10 −4<br />
0096 F 23 4.124 (1.516) −0.498 (0.265) 0.074<br />
0099 L 22 (old) 7.198 (1.625) −0.911 (0.309) 0.008<br />
0099 L 21 8.974 (1.894) −1.233 (0.355) 0.003<br />
0337 C 18 (old) 2.339 (3.442) 0.114 (0.604) 0.852<br />
0337 C 17 7.986 (7.681) −0.856 (1.325) 0.528<br />
Theoretical values<br />
vertical asymptote (N < N crit ) 6.000 −1.000<br />
horizontal asymptote (N > N crit ) 1/q 0.000<br />
The new plots show that there are no more data points with Cook’s distance greater than 0.5.<br />
The resulting new estimates for the intercept and slope of the respective linear regressions are<br />
presented in Table 5 and graphically in Figure 55. Removing the respective influential data points<br />
from the three datasets brought significant changes of the estimates of the intercept and slope of<br />
the least squares fit. The largest change is observed in the case of suppliers of products from the<br />
material group 0337 to Plant C. Here the initially estimated positive slope with small absolute value<br />
(β 1 = 0.114) changed to a strongly negative slope (β 1 = −0.856). Furthermore, the according<br />
intercept shifted as well from β 0 = 2.339 to β 0 = 7.986 (Figure 55). The new parameters of the<br />
linear fit are very close to the theoretical line representing a minimal number of ppm with intercept<br />
β 0 = 6 and slope β 1 = −1. However, the results of the performed F-test indicate that the estimates<br />
are not reliable. The very high p-value of the F-test (p=0.528) indicates that the null-hypothesis<br />
of the test that β 1 = 0 cannot be rejected on the 5%-significance level. In the other two cases the<br />
p-values of the F-test show that the estimated slopes of the respective linear fits are statistically<br />
significant even after the removal of the influential points. In the case of suppliers, which deliver<br />
components from the material group 0096 to Plant F, the removal of the influential data point lead<br />
to a smaller slope β 1 = −0.498 (in contrast to β 1 = −0.820 previously). The intercept of the<br />
fitting line β 0 = 5.991 dropped down to β 0 = 4.124. In the case of material group 0099 for Plant<br />
L, the removal of the influential point has an opposite effect. The slope of the least squares fit<br />
increased in absolute terms from β 1 = −0.911 to β 1 = −1.233. The according intercept shifted<br />
from β 0 = 7.198 to β 0 = 8.974 (see Figure 55).<br />
118
Slope (β_1)<br />
-2,5 -2,0 -1,5 -1,0 -0,5 0,0 0,5 1,0<br />
16<br />
14<br />
12<br />
Zero Slope<br />
(N > N_crit)<br />
Intercept (β_0)<br />
10<br />
8<br />
6<br />
0099; Plant L<br />
0337; Plant C<br />
4<br />
Theoretical<br />
value (min ppm)<br />
0096; Plant F<br />
2<br />
0<br />
-2<br />
Figure 55: Graphical representation of the shift of the estimated linear regression parameters presented<br />
in Table 5, after the influential points have been removed from the data sets. The error bars<br />
denote the standard error of the estimates. The label of each data point consists of the respective<br />
Material group; Plant. A list of all material groups and plant codes, which appear in this paper,<br />
can be found in the Appendix.<br />
In two of the cases above – namely for suppliers, which deliver products from the material<br />
group 0068 to Plant K, and suppliers, which deliver components from material group 0099 to Plant<br />
L respectively (in the latter case even after the removal of the influential data points) – the estimated<br />
linear fits are particularly close to the theoretical line limiting the number of ppm, which has an<br />
intersect β 0 = 6 and a slope β 1 = −1 (as already shown previously). These results suggest that the<br />
respective groups of suppliers did not yet reach their critical delivery amount of N crit = 1/q, and<br />
that even though they have quality related problems on their records, their production processes<br />
seem to be particularly stable.<br />
On the other hand, the datasets which reveal slopes of the linear fit close to zero, such as in<br />
the case of suppliers delivering material group 0068 to Plant L – β 1 = −0.089, or suppliers of<br />
parts from the material group 0485 to Plant C – β 1 = 0.072, for example, suggest that part of the<br />
suppliers in the respective group have reached their critical delivery amount and their performance<br />
record is now limited by the second asymptote derived in the theoretical part above. Such groups<br />
of suppliers require closer attention, since it is very probable that their production processes are<br />
not stable enough.<br />
119
These findings are particularly important for the supplier management process, since the analysis<br />
presented in this section provides a good approach to identify the areas, in which the probability<br />
of arising quality related problems is rather high. The results show that suppliers who supply a<br />
large number of components with a confirmed zero slope need to be handled with high priority.<br />
On the other hand, suppliers with a slope near the ”detectability” asymptote (see Figure 44) are<br />
only second priority since the probability of them causing problems is low, since their processes<br />
are either really stable in absolute terms, or are at least stable enough for the lower number of<br />
delivered components to have a low probability for causing problems. Such type of information is<br />
a particularly important input for the supplier management process and can be used to efficiently<br />
distribute auditing resources down the supply chain.<br />
120
8 Conclusion<br />
Second-party quality auditing is an essential part of the quality assurance process especially in<br />
businesses with high levels of outsourcing. The goal of the presented project was to assess the<br />
effectiveness of the practical implementation of this quality management tool in the automotive<br />
industry and using a case study approach to answer the following two research questions:<br />
<br />
Can quality performance of automotive suppliers in the mass production be anticipated based<br />
on quality evaluation of the capability of their production processes?<br />
<br />
What is the quality of the quality auditing process at Volkswagen Group and what is its<br />
improvement potential?<br />
Figure 56 shows once again the flow of the discussion and summarizes the contents of the individual<br />
sections of this paper. Section 1 introduced the concept of quality and its importance for the<br />
business environment. It also briefly addressed the quality-related challenges faced by companies,<br />
which manufacture complex products in cooperation with external business partners. Section 2<br />
presented the research questions and the motivation for this paper. One important aspect in this<br />
regard was that automotive companies need a quality management tool, which allows them to effectively<br />
assess the quality capability of their potential business partners before the contracts are<br />
in place and thus strategically select the most suitable suppliers, which can guarantee the stability<br />
of OEMs’ production processes. At the same time it is important that the selected suppliers<br />
maintain their quality capability even after the start of production and any quality related issues<br />
are addressed in a timely manner. At this point the generally accepted quality standards such as<br />
ISO 9001 and ISO/TS16949 and the Total Quality Management principle upon which they build<br />
were introduced as a possible answer to the needs of automotive OEMs. The discussion focused<br />
on the ability of such standards to guarantee the quality of certified companies. The section reviewed<br />
scientific literature, which describes potential deficiencies of the certification according to<br />
these quality standards and explains why second-party quality auditing is its preferred counterpart<br />
in the quality management practices in a lot of industries including the automotive. Pursuing<br />
the presented research questions was additionally motivated by the fact that there is small scientific<br />
base, which provides information whether this quality management tool is indeed effectively<br />
implemented in practice in the automotive industry.<br />
121
1.<br />
Introduction<br />
2.<br />
Research Question and Research Motivation<br />
Research Questions:<br />
• Can quality performance of automotive suppliers in the mass production be anticipated based<br />
on quality evaluation of the capability of their production processes?<br />
• What is the quality of the quality auditing process at Volkswagen Group and what is its<br />
improvement potential?<br />
3.<br />
Research Approach<br />
Framework for the analysis (case study):<br />
• Similarities between quality auditing and sampling<br />
• Two aspects, which are important for effective quality auditing:<br />
− Frequency of audits in the context of the general audit planning (sampling frequency)<br />
− Effectiveness of individual audits (accuracy of the samples)<br />
General Auditing Strategy<br />
(Sampling Frequency)<br />
4. Specifics of the Automotive Industry<br />
Fundamental Principles of the<br />
5.<br />
Quality Audit<br />
• Specifics of the industry:<br />
− Internationalization of the<br />
automotive operations<br />
− Rising development costs<br />
− Increased complexity of the<br />
automotive supply chain<br />
• Summary of the quality challenges and<br />
Implications for the quality auditing<br />
strategy in the automotive sector<br />
• Evaluation of Volkswagen’s quality<br />
auditing process with respect to the<br />
insights from the discussion<br />
8.<br />
Conclusion<br />
• Summary of the analytical results<br />
• Comments of the results with<br />
respect to the research questions<br />
of the paper<br />
• Outlook<br />
6.<br />
Effectiveness of Individual<br />
Audits (Accuracy of Samples)<br />
Empirical Data<br />
• Overview of the analyzed empirical data<br />
and the evaluation methods relevant for<br />
data collection:<br />
− Quality performance records<br />
− Quality capability records<br />
• Technical aspects of the analysis<br />
• Important recommendations regarding<br />
Volkswagen’s data management<br />
7. Results and Discussion<br />
• Comprehensive evaluation of the available<br />
empirical data w/r/t incomplete records<br />
• Account for potential biasing factors<br />
− Revisions of Volkswagen’s quality<br />
auditing process<br />
− Relevance of audit results with time<br />
− Influences resulting from the<br />
calibration of auditors and quality<br />
inspectors<br />
• Correlation of quality performance and<br />
quality capability for two data sets<br />
• Theoretical and empirical evaluation of the<br />
delivery amount as a second predictor for<br />
quality performance<br />
Figure 56: Flow of the discussion presented in this paper.<br />
122
Section 3 presented the research approach. The analysis in this paper was based on a case study<br />
carried out in cooperation with one of the largest automotive producers in the world – Volkswagen<br />
Group. An important contribution of this section is the analytical framework it introduced, which<br />
draws similarities between the quality auditing process and one of the fundamental techniques in<br />
the field of Electrical Engineering – sampling. An important benefit of this particular analytic<br />
approach is that it is independent of the specifics of the automotive industry and can be used to<br />
assess the effectiveness of quality auditing also in other industry sectors. The two aspects of the assessment<br />
framework are the frequency of auditing (analog to the sampling frequency in Electrical<br />
Engineering) and the ability of the individual audits to evaluate adequately the quality capability<br />
of suppliers’ production processes (analog to the accuracy of the individual measurements). As a<br />
consequence of the defined evaluation framework the rest of the paper was divided into two conceptually<br />
different but mutually complementing parts. The aspect of auditing frequency was dealt<br />
with in Section 4, while the remaining sections (Sections 5 through 7) focused on the effectiveness<br />
of the individual audits. This division is also reflected in the current section.<br />
8.1 General Auditing Strategy (Sampling Frequency)<br />
Section 4 examined the specifics of the automotive market and the associated quality management<br />
challenges down the automotive supply chain. On the one hand, strategic moves such as the relocation<br />
of production to cost-efficient regions, increasing the share of outsourced production, and<br />
at the same time the introduction of new production approaches such as the use of modular assembly<br />
components, offer substantial competitive advantages to the individual market players and are<br />
essential for the competitiveness of automotive manufacturers. On the other hand, factors such as<br />
personnel fluctuation in developing regions, large production volume increases and the resulting<br />
lack of sustainability of quality of suppliers’ production processes, rapidly growing complexity of<br />
the supplier pool, product portfolio integration and quality risk concentration in production modules,<br />
which are extensively used in smaller varieties and increasing numbers and complexity, not<br />
only pose serious challenges to quality management in the automotive sector, but also potentially<br />
endanger OEMs’ overall business operations.<br />
Based on the insights of the discussion a possible approach was suggested (see Figure 8 in<br />
Section 4.4), which can be used to prioritize quality audits in the general quality auditing plan. To<br />
respond to the challenges of the sector automotive players need to adjust accordingly their audit<br />
planning and the frequency of audits in areas, which are subject to particularly strong influence of<br />
123
the factors mentioned above. Special attention should be paid to sub-supplier management as well<br />
as securing the stability of the production processes of key partners on the newly opened markets.<br />
Production concepts similar to Volkswagen’s modular strategy MQB must also be handled with<br />
priority in the general audit planning, especially considering the significantly larger volumes of the<br />
sourced components.<br />
Already in this section followed the first important conclusion regarding the research questions.<br />
The average age of audit evaluations of Volkswagen suppliers from a number of different<br />
regions were compared to the recommended audit prioritizing (Figure 8). The results showed that<br />
Volkswagen’s audit planning accounts for the specifics of the automotive industry. Thus suppliers<br />
in developing regions such as China, India, or Russia were audited much more frequently than<br />
suppliers in developed regions such as Germany or North America. Furthermore, the number of<br />
the sub-supplier audits conducted by Volkswagen is constantly rising. One aspect, which needs<br />
yet to be considered in Volkswagen’s general quality audit planning, is the start of production of<br />
Volkswagen models based on the MQB platform planned for 2012.<br />
However, this conclusion is based on a qualitative evaluation of the empirical data only. It is<br />
difficult to determine whether the auditing frequency is optimal for the according regions, e.g. is<br />
the average age of quality audits in the individual regions small enough to prevent quality problems<br />
or should it be reduced even further? The answer to such a question is limited by the a vast number<br />
of parameters. Nevertheless, this particular point represents an interesting topic for subsequent<br />
research.<br />
8.2 Effectiveness of Individual Audits (Accuracy of Samples)<br />
Once the aspects regarding the general audit planning (the sampling frequency) were treated, the<br />
remaining part of the paper concentrated on the ability of individual audits to adequately evaluate<br />
the quality capability of suppliers’ production processes (accuracy of the samples). This part included<br />
the bulk of statistical evaluations on the available empirical data. In the beginning of this<br />
paper (Section 3) it was hypothesized, that if individual quality audits are conducted effectively, it<br />
should be expected that suppliers, which receive a higher quality capability rating during an audit,<br />
should also perform better in terms of quality performance indicators, such as ppm. This is the<br />
reason why the ultimate purpose of the analysis in this part of the paper was to correlate quality<br />
capability to quality performance information in order to search for hints, which can help to find an<br />
answer to the research questions. Before the according correlations could be performed, however,<br />
124
several additional considerations had to be taken into account such as the theoretical aspects of<br />
the quality audit, specifics and quality of the empirical data, as well as a number of factors, which<br />
could potentially bias the analytical results and obscure important trends in the evaluated data.<br />
Section 5 introduced the theoretical aspects of the quality audit. It focused on the specifics of<br />
its implementation and presented the individual stages of a single quality audit. The purpose of this<br />
chapter was to introduce the different auditing approaches and thus the possible diversity of implementation<br />
of this quality management tool in terms of auditing techniques, team size, scoping,<br />
technical content, etc. Alongside the presented literature review the quality audit implementation<br />
specific to Volkswagen was roughly outlined. One particular insight of this section is the fact that<br />
the effectiveness of the audit can be strongly influenced by auditors’ knowledge of the state-ofthe-art<br />
of the audited production processes. In this regard Volkswagen pursues a good strategy<br />
of maintaining a pool of auditing experts, each with profound process knowledge in a particular<br />
strategic area.<br />
Section 6 presented in detail the two major types of empirical data, which were analyzed<br />
throughout the paper – quality capability and quality performance information. Along with the<br />
processes, which generate the empirical data, the section discussed also particular technical aspects<br />
of the analysis regarding the automated data processing. This discussion outlined one especially<br />
important finding, which is relevant for the more subjective research question defined in the<br />
beginning of the paper regarding improvment potential of Volkswagen’s internal processes. The<br />
challenges faced during the data processing underscored the importance of data management in<br />
large companies and showed once again how complex this task can be.<br />
The challenge at Volkswagen in particular originates, on the one hand, in the different types of<br />
categorizations used by the two Volkswagen departments, which provided the empirical data – material<br />
groups (Volkswagen Purchasing) and product groups (Volkswagen Quality Assurance). This<br />
problem is on the conceptual level and in the meantime has already been addressed by the according<br />
departments in order to improve the information exchange in Volkswagen’s internal processes<br />
such as the corporate sourcing process. Once sufficient amount of empirical data is generated<br />
using the new information management approach, this improvement will also benefit any future<br />
analyses similar to the one presented in this work.<br />
On the other hand, one important aspect which still has to be addressed is the fact that valuable<br />
detailed quality capability information collected during supplier evaluations is stored in a form,<br />
which makes it practically unusable for analysis on a large scale (see Section 6.3). These are<br />
125
data regarding the complexity of the evaluated production processes (number of processing steps),<br />
partial evaluation results for the individual blocks of the auditing questionnaire, etc. Fortunately,<br />
the solution to this particular problem is relatively easy, since the data are generated in an electronic<br />
form. Addressing these problems will allow Volkswagen to gain valuable additional knowledge<br />
about its internal processes and accordingly use it to improve its operations even further.<br />
Section 7 presented the results of the conducted evaluations. Section 7.1 evaluated the quality<br />
of the available empirical data as the integrity of the operational records was of particular interest.<br />
The results of the analysis revealed differences in the completeness of quality performance information<br />
on the regional level. However, considering the specifics of the automotive sector such<br />
differences were to be expected. Nevertheless, the quality of the generated quality performance<br />
records improved over the evaluated time period as regions such as Region 5 (Figure 22) and Region<br />
8 (Figure 25in Section 7.1) demonstrated particularly positive development. All incomplete<br />
records were excluded from the analysis. Due to their small overall amount (less than 5%) it is<br />
assumed that the remaining data set is still representative. Even though the incomplete records<br />
were eliminated, the identified differences between the quality of the regional datasets were the<br />
reason to differentiate between regional data also in the subsequent analysis.<br />
Section 7.2 dealt with potential sources of data bias. Among the evaluated factors were the<br />
influence of revisions of Volkswagen’s quality auditing process (Section 7.2.1), the period of relevance<br />
of quality audit results (Section 7.2.2), as well as calibration of the people involved in<br />
supplier quality evaluation (Section 7.2.3.1) and production quality assessment (Section 7.2.3.2).<br />
Section 7.2.1 compared the quality evaluation results of suppliers carried out according to four<br />
different versions of the Formel Q-Fähigkeit. The initial analysis divided the data based only on<br />
industry type – chemical, electrical, and metal accordingly. The results of the evaluations showed<br />
that quality evaluations according to the fifth and sixth edition of the Formel Q-Fähigkeit (FQF 5<br />
and FQF 6) are not statistically different in all cases, while the conducted Kolmogorov-Smirnov<br />
test showed statistically significant differences between the evaluations according to FQF 3 and<br />
FQF 4 and between these datasets and the former two. Similar results were observed also in most<br />
of the cases on the regional level. These observations showed that the changes of the quality auditing<br />
evaluation criteria do have an influence on the empirical data and if they are not accounted<br />
for could bias the analytical results.<br />
However, there were also several datasets, which were found to be statistically different even<br />
with respect to FQF 5 and FQF 6. Among these are the quality capability evaluations of chemical-<br />
126
part suppliers in Region A (Figure 57 in Section A in the Appendix) and Region G (Figure 63), and<br />
metal-part suppliers in Region C (Figure 71) and Region H (Figure 76). In Region D statistically<br />
significant difference between the datasets was observed for both types of suppliers (Figures 60<br />
and 72).<br />
These observations suggested the presence of another factor, which affects the quality capability<br />
records. This is namely the period of relevance of the auditing results (Section 7.2.2). In<br />
this case the influencing factor is time. Due to the fact that a quality audit is an assessment of<br />
a momentary state of the constantly changing production process, audit’s validity over time also<br />
changes. Thus even effectively conducted audits do not correspond to the actual quality capability<br />
of a supplier after a sufficiently long period of time. To account for this issue it is necessary to<br />
define a suitable weighing function, which will give quality performance records collected around<br />
the time of the audit more importance than the rest of supplier’s quality performance records (see<br />
Section 7.2.2). The definition of such a weighing function is particularly important for regions with<br />
very high fluctuations of the process quality capability such as developing regions. Suitable input<br />
for the definition of a weighing function are supplier-specific process variation indicators such as<br />
process cpk-values, company-specific KPIs, etc. However, the empirical data analyzed here does<br />
not include any such information. This is the reason why, the comparisons between quality capability<br />
and quality performance were restricted mainly to regions with relatively high stability of<br />
suppliers’ production processes such as developed regions.<br />
Section 7.2.3.1 assessed whether there are differences between the evaluations of audit teams in<br />
the individual regions. The majority of the evaluated audit records showed very good consistency,<br />
which implies that Volkswagen’s audit teams are well calibrated. In the few cases, in which statistically<br />
significant difference between the compared datasets was observed, either the number of<br />
observations was so low, that the small size of the datasets is the most probable reason for the statistical<br />
significance of the Kolmogorov-Smirnov test, or the empirical data were subject to specific<br />
regional influences and such differences had to be expected. Thus the observed differences do not<br />
necessarily indicate different calibration of the audit teams. These results show that differences between<br />
the evaluations of the auditors are possible, even when they are equally well calibrated. It is<br />
therefore necessary in any subsequent analysis to also test for differences between the evaluations<br />
of the individual audit teams.<br />
The last section, which addressed a possible biasing factor was Section 7.2.3.2. It evaluated<br />
how similar or how different the quality performance of suppliers is assessed by the individual<br />
127
Volkswagen production facilities. Most of the calculations revealed consistent evaluations of the<br />
individual production plants. However, in several cases the evaluations of one or more production<br />
plants showed systematic deviations from the evaluations of other plants, which received the same<br />
components from the same material group (see Figures 34 through 36 in Section 7.2.3.2). In<br />
these cases the differences are due rather to the heterogeneity of the components comprising a<br />
particular product group (Section 7.2.3.2), than to lack of calibration between the assessment<br />
of individual production facilities. Thus, these results reveal another factor, which could have a<br />
negative influence on the correlation study of quality capability and quality performance of the<br />
suppliers.<br />
After the possible sources of bias in the empirical data were examined, for two datasets quality<br />
performance information was tested for correlation against the quality capability records of the respective<br />
suppliers (Section 7.3). The purpose of these calculations was to test for the initial hypothesis<br />
of this study introduced in Section 3. The results of the linear regressions between the number<br />
of ppm and the respective supplier evaluation scores in the two case studies presented in Figures 38<br />
and 40 do not show any statistically significant evidence that supports the null-hypothesis. Furthermore,<br />
the results of the two case studies did not show any statistically significant evidence also<br />
for differences between the quality capability evaluation results of suppliers, which are reported<br />
to have quality related problems (expressed in terms of ppm), and suppliers, which delivered with<br />
excellent performance record (0 ppm) throughout the entire time period covered by the analysis.<br />
The last section of the presented analysis describes one particularly important finding, which<br />
was not anticipated in the beginning of this project. It turned out that suppliers with quality related<br />
problems differ from suppliers without any problems, not only based on their ppm record<br />
but also on the amount of delivered components. Section 7.4 presents a mathematical derivation<br />
of the relation between the performance record expressed in ppm of suppliers with quality related<br />
problems and the amount of components they delivered. These two quantities have a relatively<br />
simple relationship. Expressed on a double logarithmic scale the mathematical expectation of the<br />
ppm-value as a function of the delivery amount N is asymptotically bound by two straight lines<br />
(see Figure 44 in Section 7.4). Using this information one can deduct valuable information about<br />
the quality capability of a supplier’s production process.<br />
This issue was further investigated for several sets of empirical data, as the relation between<br />
the ppm and the delivered amount of components was tested with a linear regression. Cases, in<br />
which the estimated linear fits are particularly close to the theoretical line limiting the number<br />
128
of ppm, which has an intersect β 0 = 6 and a slope β 1 = −1 (see Section 7.4), suggest that the<br />
respective groups of suppliers did not yet reach their critical delivery amount of N crit , and that<br />
even though they have quality related problems on their records, their production processes seem<br />
to be particularly stable. By contrast, datasets, which reveal slopes of the linear fit close to zero,<br />
require closer attention, since it is very probable that their production processes are not stable<br />
enough. Such information might be a useful contribution for the supplier management process and<br />
can provide assistance in distributing the auditing resources.<br />
8.3 Answers of the Research Questions<br />
Can quality performance of automotive suppliers in the mass production be anticipated based<br />
on quality evaluation of the capability of their production processes?<br />
A definitive answer to this question is not possible at this stage due to several important reasons.<br />
First of all, due to the technical difficulties experienced during the analytical part of this<br />
project (such as difficult access to empirical data) the analysis was seriously restricted in its scope.<br />
The only two cases with relatively small datasets, which were evaluated in Section 7.3, are not representative<br />
for the entire set of empirical data. Therefore, even though the results of the empirical<br />
analysis of the two datasets does not provide sufficient evidence, that the quality performance of<br />
suppliers in these two particular cases can be anticipated based on the evaluation on the capability<br />
of their production processes, this does not definitely mean that such relation does not exist. On the<br />
contrary, further analysis surely has the potential to provide evidence that support this hypothesis.<br />
Furthermore, as already discussed above there are a number of factors, which could potentially<br />
introduce bias. Through statistical evaluation it was shown that the following factors can surely<br />
influence the character of empirical information:<br />
<br />
Revisions of the Formel Q-Fähigkeit;<br />
<br />
Period of relevance of the audit results;<br />
<br />
Coherence of the evaluations of different auditors.<br />
The conducted evaluations surely account for the potential biases mentioned above. On the<br />
other hand, the evaluations are based on quality performance data, which are averaged over the<br />
entire time period considered in the analysis. The data averages were computed without the use<br />
of a weighing function. This approach might not be optimal, given that quality problems with<br />
significantly large time offset from a particular supplier evaluation are given the same weight as<br />
129
problems arising in a point of time close to the actual quality audit. The analysis was also technically<br />
limited given the fact that some of the available operational data was difficult to access.<br />
For example, the analyzed supplier quality capability records do not provide information about<br />
the structure of the evaluated processes. This could eventually result in handling relatively simple<br />
processes the same as processes, which are considerably more complex. Further, the quality<br />
performance information included in the analysis is clustered based on the material group classification,<br />
which in many cases includes a large number of diverse components in the same material<br />
group. These particular factors were not regarded during the analytical stages, and any influences<br />
they might have on the conducted evaluations is not excluded. Such potential bias should definitely<br />
be investigated in any subsequent studies.<br />
Even though the current analysis could not provide a definitive answer to this research question,<br />
it employs important analytical methods, which can serve for guidance of any future works on the<br />
topic. The comprehensive evaluation of the empirical data derived valuable relations and identified<br />
important biasing factors, which have to be definitely avoided in any subsequent analysis. One of<br />
the major contributions of the current analysis was the identification of a second quality predictor<br />
– the amount of delivered components. This paper presented a formal mathematical derivation of<br />
the identified relationship as well as a set of empirical evaluations, which demonstrate its practical<br />
significance. This finding definitely has a lot of potential for further research and can reveal further<br />
valuable aspects of the quality management process.<br />
What is the quality of the quality auditing process at Volkswagen Group and what is its<br />
improvement potential?<br />
The analysis presented in Section 4 showed that the general quality auditing strategy employed<br />
by Volkswagen adequately responds to the quality challenges of the automotive sector. In this<br />
regard the more frequent auditing in developing regions accounts for process instability resulting<br />
from factors such as personnel fluctuations and rapid growth. Furthermore, the increased focus<br />
on sub-supplier audits allows Volkswagen to counteract to another trend in the automotive sector<br />
– the growing complexity of the automotive supply chain. However, as already mentioned previously,<br />
these conclusions are based on qualitative evaluation of the empirical data, and therefore a<br />
quantitative analysis of this aspect is a particularly interesting topic for further research.<br />
Volkswagen’s quality auditing process is defined by good practices also on the level of individual<br />
audits. As already mentioned in Section 5, the comprehensive know-how about the state-of-theart<br />
of the audited processes gives auditors an advantage during the quality audit. Such knowledge<br />
130
allows them to identify potential process weaknesses more quickly and therefore to provide a more<br />
adequate process evaluation in the limited amount of time. The strategy pursued by Volkswagen<br />
to maintain strategic know-how in expert circles is particularly useful to this end. Volkswagen<br />
auditors perform audits mainly in one particular area of specialization, which allows them to accumulate<br />
a very comprehensive knowledge base in that particular area. This approach is much more<br />
efficient than auditing in all possible areas.<br />
Along with the positive aspects of the quality auditing process, based on the experiences gained<br />
throughout this project several particularly important improvement potentials of Volkswagen’s information<br />
management were identified. These can be summarized within the following:<br />
<br />
Categorization of quality capability and quality performance information needs improvement<br />
(mismatch between the material / product group classification, see Section 6.2); A<br />
solution to this particular point has already been implemented by the responsible departments.<br />
However, at the time this project was carried out, the generated empirical data using<br />
the new information management approach was not sufficient for a comprehensive analysis.<br />
The new database, however, is of particular interest for future studies.<br />
<br />
Valuable data from audit reports is very difficult to access (stored as document attachment,<br />
not suitable for analysis, see Section 6.3); however, as already mentioned in the previous<br />
sections a solution is possible with little overhead.<br />
<br />
Incomplete quality performance records are present (Section 7.1); nevertheless, the overall<br />
amount of complete records amounts to more than 95% and the quality of empirical data<br />
shows considerable improvement over time.<br />
Addressing the identified improvement areas will contribute to the quality of the operational<br />
data and will ultimately have positive effect on the learning cycles in the organization.<br />
Even though the discussion presented in this paper focuses on topics from the automotive production,<br />
the analytical approaches and methods used here are applicable in any other business<br />
context. Several important conclusions stem from the presented analysis:<br />
<br />
In order to properly assess the effectiveness of the implementation of second-party quality<br />
auditing specific to a particular company or sector it is necessary to evaluate the auditing process<br />
from two perspectives – strategic employment of the audit expressed in the general audit<br />
plan and its technical implementation expressed in terms of the structuring and execution of<br />
individual quality audits.<br />
131
Frequency of auditing is affected by factors specific to the industry of operation, as well as<br />
the characteristics of the individual markets. To appropriately schedule the quality audits<br />
it is important to understand the specific conditions of a particular business field, which<br />
determine the direction of its development. Example of such factors are regional differences<br />
as well as the economic challenges, which influence companies’ decisions and production<br />
strategies.<br />
<br />
It is important to consider factors, which could potentially influence the individual evaluations<br />
such as calibration of the respective quality auditors.<br />
<br />
One particularly important conclusion, which results from the presented case study is the fact<br />
that a proper management of the quality auditing process is strongly influenced by company’s<br />
approach for management of the operational data obtained during the supplier visits. For this<br />
aspect it is very important to develop a suitable concept and provide the required technical resources,<br />
which allow for seamless regular evaluation of data generated in different phases of<br />
the overall business process. Special attention is required in matching the respective records<br />
especially for data resulting from different departments of the organization. Data analysis is<br />
a necessary step, which boils down the collected knowledge into specific recommendations<br />
for management and continuous improvement of the quality auditing process.<br />
132
Team_A_Chemie_FQF5<br />
Team_A_Chemie_FQF6<br />
Team_A_Chemie_FQF5<br />
Team_A_Chemie_FQF6<br />
Appendices<br />
A Results from the analysis presented in Section 7.2.1<br />
Distribution of Team_A_Chemie_FQF5<br />
Frequency<br />
0 5 10 15 20 25<br />
60 70 80 90 100<br />
Shapiro-Wilk Normality test p-values<br />
Distribution of Team_A_Chemie_FQF6<br />
Frequency<br />
0 5 10 15<br />
60 70 80 90 100<br />
p-values 0,002 2,9E-06<br />
Samples 174 154<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_A_Chemie_FQF5<br />
and Team_A_Chemie_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.152<br />
p = 0.04596<br />
n_1 = 174<br />
n_2 = 154<br />
Team_A_Chemie_FQF5<br />
Team_A_Chemie_FQF6<br />
60 70 80 90 100<br />
Evaluation Scores [%]<br />
(a) Evaluations according to FQF 5.0 and FQF 6.0.<br />
Team_A_Chemie_FQF5 1,000 0,046<br />
Team_A_Chemie_FQF6 0,046 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(b) Distribution comparisons summary<br />
Figure 57: Distribution comparisons between the evaluation scores of chemical-part suppliers<br />
located in Region A, performed according to different versions of Formel Q-Fähigkeit by the local<br />
audit team.<br />
133
Distribution of Team_B_Chemie_FQF4<br />
Distribution of Team_B_Chemie_FQF5<br />
Frequency<br />
0 2 4 6 8<br />
50 60 70 80 90 100<br />
Distribution of Team_B_Chemie_FQF5<br />
Frequency<br />
0 2 4 6 8 10 12<br />
Frequency<br />
0 2 4 6 8 10 12<br />
60 70 80 90 100<br />
Distribution of Team_B_Chemie_FQF6<br />
Frequency<br />
0 1 2 3 4 5<br />
Team_B_Chemie_FQF4<br />
Team_B_Chemie_FQF5<br />
Team_B_Chemie_FQF6<br />
Team_B_Chemie_FQF4<br />
Team_B_Chemie_FQF5<br />
Team_B_Chemie_FQF6<br />
50 60 70 80 90 100<br />
60 70 80 90 100<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_B_Chemie_FQF4<br />
and Team_B_Chemie_FQF5<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.09498<br />
p = 0.8501<br />
n_1 = 69<br />
n_2 = 103<br />
Team_B_Chemie_FQF4<br />
Team_B_Chemie_FQF5<br />
50 60 70 80 90 100<br />
Evaluation Scores [%]<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_B_Chemie_FQF5<br />
and Team_B_Chemie_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.1068<br />
p = 0.9154<br />
n_1 = 103<br />
n_2 = 37<br />
Team_B_Chemie_FQF5<br />
Team_B_Chemie_FQF6<br />
60 70 80 90 100<br />
Evaluation Scores [%]<br />
(a) Evaluations according to FQF 4.0 and FQF 5.0.<br />
(b) Evaluations according to FQF 5.0 and FQF 6.0.<br />
Shapiro-Wilk Normality test p-values<br />
p-values 4,3E-09 2,3E-09 0,003<br />
Samples 69 103 37<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
Team_B_Chemie_FQF4 1,000 0,850 0,780<br />
Team_B_Chemie_FQF5 0,850 1,000 0,915<br />
Team_B_Chemie_FQF6 0,780 0,915 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(c) Distribution comparisons summary<br />
Figure 58: Distribution comparisons between the evaluation scores of chemical-part suppliers<br />
located in Region B, performed according to different versions of Formel Q-Fähigkeit by the local<br />
audit team.<br />
134
Team_C_Chemie_FQF5<br />
Team_C_Chemie_FQF6<br />
Team_C_Chemie_FQF5<br />
Team_C_Chemie_FQF6<br />
Distribution of Team_C_Chemie_FQF5<br />
Frequency<br />
0 5 10 15 20 25<br />
50 60 70 80 90 100<br />
Shapiro-Wilk Normality test p-values<br />
Distribution of Team_C_Chemie_FQF6<br />
Frequency<br />
0 5 10 15 20<br />
50 60 70 80 90 100<br />
p-values 1,3E-12 0,003<br />
Samples 202 120<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_C_Chemie_FQF5<br />
and Team_C_Chemie_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.03408<br />
p = 1<br />
n_1 = 202<br />
n_2 = 120<br />
Team_C_Chemie_FQF5<br />
Team_C_Chemie_FQF6<br />
50 60 70 80 90 100<br />
Evaluation Scores [%]<br />
(a) Evaluations according to FQF 5.0 and FQF 6.0.<br />
Team_C_Chemie_FQF5 1,000 1,000<br />
Team_C_Chemie_FQF6 1,000 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(b) Distribution comparisons summary<br />
Figure 59: Distribution comparisons between the evaluation scores of chemical-part suppliers<br />
located in Region C, performed according to different versions of Formel Q-Fähigkeit by the local<br />
audit team.<br />
135
Team_D_Chemie_FQF5<br />
Team_D_Chemie_FQF6<br />
Team_D_Chemie_FQF5<br />
Team_D_Chemie_FQF6<br />
Distribution of Team_D_Chemie_FQF5<br />
Frequency<br />
0 5 10 15 20<br />
60 70 80 90 100<br />
Shapiro-Wilk Normality test p-values<br />
Distribution of Team_D_Chemie_FQF6<br />
Frequency<br />
0 5 10 20 30<br />
60 70 80 90 100<br />
p-values<br />
5,7E-06 7,4E-06<br />
Samples 79 155<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_D_Chemie_FQF5<br />
and Team_D_Chemie_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.294<br />
p = 0.0002357<br />
n_1 = 79<br />
n_2 = 155<br />
Team_D_Chemie_FQF5<br />
Team_D_Chemie_FQF6<br />
60 70 80 90 100<br />
Evaluation Scores [%]<br />
(a) Evaluations according to FQF 5.0 and FQF 6.0.<br />
Team_D_Chemie_FQF5 1,000 2,4E-04<br />
Team_D_Chemie_FQF6 2,4E-04 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(b) Distribution comparisons summary<br />
Figure 60: Distribution comparisons between the evaluation scores of chemical-part suppliers<br />
located in Region D, performed according to different versions of Formel Q-Fähigkeit by the local<br />
audit team.<br />
136
Team_E_Chemie_FQF5<br />
Team_E_Chemie_FQF6<br />
Team_E_Chemie_FQF5<br />
Team_E_Chemie_FQF6<br />
Distribution of Team_E_Chemie_FQF5<br />
Frequency<br />
0 5 10 15<br />
60 70 80 90 100<br />
Shapiro-Wilk Normality test p-values<br />
Distribution of Team_E_Chemie_FQF6<br />
Frequency<br />
0 2 4 6 8 10<br />
60 70 80 90 100<br />
p-values<br />
3,0E-08 2,3E-07<br />
Samples 154 119<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_E_Chemie_FQF5<br />
and Team_E_Chemie_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.1971<br />
p = 0.01086<br />
n_1 = 154<br />
n_2 = 119<br />
Team_E_Chemie_FQF5<br />
Team_E_Chemie_FQF6<br />
60 70 80 90 100<br />
Evaluation Scores [%]<br />
(a) Evaluations according to FQF 5.0 and FQF 6.0.<br />
Team_E_Chemie_FQF5 1,000 0,011<br />
Team_E_Chemie_FQF6 0,011 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(b) Distribution comparisons summary<br />
Figure 61: Distribution comparisons between the evaluation scores of chemical-part suppliers<br />
located in Region E, performed according to different versions of Formel Q-Fähigkeit by the local<br />
audit team.<br />
137
Distribution of Team_F_Chemie_FQF4<br />
Distribution of Team_F_Chemie_FQF5<br />
Frequency<br />
0 10 20 30 40<br />
65 70 75 80 85 90 95 100<br />
Distribution of Team_F_Chemie_FQF5<br />
Frequency<br />
0 5 10 15 20 25 30<br />
Frequency<br />
0 5 10 15 20 25 30<br />
70 75 80 85 90 95 100<br />
Distribution of Team_F_Chemie_FQF6<br />
Frequency<br />
0 5 10 15 20<br />
Team_F_Chemie_FQF4<br />
Team_F_Chemie_FQF5<br />
Team_F_Chemie_FQF6<br />
Team_F_Chemie_FQF4<br />
Team_F_Chemie_FQF5<br />
Team_F_Chemie_FQF6<br />
65 70 75 80 85 90 95 100<br />
70 75 80 85 90 95 100<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_F_Chemie_FQF4<br />
and Team_F_Chemie_FQF5<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.1747<br />
p = 0.005529<br />
n_1 = 177<br />
n_2 = 212<br />
Team_F_Chemie_FQF4<br />
Team_F_Chemie_FQF5<br />
65 70 75 80 85 90 95 100<br />
Evaluation Scores [%]<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_F_Chemie_FQF5<br />
and Team_F_Chemie_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.08201<br />
p = 0.6845<br />
n_1 = 212<br />
n_2 = 119<br />
Team_F_Chemie_FQF5<br />
Team_F_Chemie_FQF6<br />
70 75 80 85 90 95 100<br />
Evaluation Scores [%]<br />
(a) Evaluations according to FQF 4.0 and FQF 5.0.<br />
(b) Evaluations according to FQF 5.0 and FQF 6.0.<br />
Shapiro-Wilk Normality test p-values<br />
p-values<br />
3,3E-11 5,5E-10 4,5E-08<br />
Samples 177 212 119<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
Team_F_Chemie_FQF4 1,000 0,006 1,7E-04<br />
Team_F_Chemie_FQF5 0,006 1,000 0,684<br />
Team_F_Chemie_FQF6 1,7E-04 0,684 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(c) Distribution comparisons summary<br />
Figure 62: Distribution comparisons between the evaluation scores of chemical-part suppliers<br />
located in Region F, performed according to different versions of Formel Q-Fähigkeit by the local<br />
audit team.<br />
138
Team_G_Chemie_FQF5<br />
Team_G_Chemie_FQF6<br />
Team_G_Chemie_FQF5<br />
Team_G_Chemie_FQF6<br />
Distribution of Team_G_Chemie_FQF5<br />
Frequency<br />
0 5 10 15<br />
75 80 85 90 95 100<br />
Shapiro-Wilk Normality test p-values<br />
Distribution of Team_G_Chemie_FQF6<br />
Frequency<br />
0 10 20 30 40<br />
75 80 85 90 95 100<br />
p-values<br />
9,2E-06 9,6E-09<br />
Samples 102 125<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_G_Chemie_FQF5<br />
and Team_G_Chemie_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.2264<br />
p = 0.006305<br />
n_1 = 102<br />
n_2 = 125<br />
Team_G_Chemie_FQF5<br />
Team_G_Chemie_FQF6<br />
75 80 85 90 95 100<br />
Evaluation Scores [%]<br />
(a) Evaluations according to FQF 5.0 and FQF 6.0.<br />
Team_G_Chemie_FQF5 1,000 0,006<br />
Team_G_Chemie_FQF6 0,006 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(b) Distribution comparisons summary<br />
Figure 63: Distribution comparisons between the evaluation scores of chemical-part suppliers<br />
located in Region G, performed according to different versions of Formel Q-Fähigkeit by the local<br />
audit team.<br />
139
Team_H_Chemie_FQF5<br />
Team_H_Chemie_FQF6<br />
Team_H_Chemie_FQF5<br />
Team_H_Chemie_FQF6<br />
Distribution of Team_H_Chemie_FQF5<br />
Frequency<br />
0 5 10 15<br />
40 50 60 70 80 90 100<br />
Shapiro-Wilk Normality test p-values<br />
Distribution of Team_H_Chemie_FQF6<br />
Frequency<br />
0 5 10 15 20 25<br />
40 50 60 70 80 90 100<br />
p-values<br />
2,1E-04 1,5E-11<br />
Samples 139 147<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_H_Chemie_FQF5<br />
and Team_H_Chemie_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.09382<br />
p = 0.5556<br />
n_1 = 139<br />
n_2 = 147<br />
Team_H_Chemie_FQF5<br />
Team_H_Chemie_FQF6<br />
40 50 60 70 80 90 100<br />
Evaluation Scores [%]<br />
(a) Evaluations according to FQF 5.0 and FQF 6.0.<br />
Team_H_Chemie_FQF5 1,000 0,556<br />
Team_H_Chemie_FQF6 0,556 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(b) Distribution comparisons summary<br />
Figure 64: Distribution comparisons between the evaluation scores of chemical-part suppliers<br />
located in Region H, performed according to different versions of Formel Q-Fähigkeit by the local<br />
audit team.<br />
140
Team_I_Chemie_FQF5<br />
Team_I_Chemie_FQF6<br />
Team_I_Chemie_FQF5<br />
Team_I_Chemie_FQF6<br />
Distribution of Team_I_Chemie_FQF5<br />
Frequency<br />
0 2 4 6 8 10<br />
80 85 90 95 100<br />
Shapiro-Wilk Normality test p-values<br />
Frequency<br />
0 1 2 3 4 5 6 7<br />
Distribution of Team_I_Chemie_FQF6<br />
80 85 90 95 100<br />
p-values 0,020 0,087<br />
Samples 41 42<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_I_Chemie_FQF5<br />
and Team_I_Chemie_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.1603<br />
p = 0.6608<br />
n_1 = 41<br />
n_2 = 42<br />
Team_I_Chemie_FQF5<br />
Team_I_Chemie_FQF6<br />
80 85 90 95 100<br />
Evaluation Scores [%]<br />
Team_I_Chemie_FQF5 1,000 0,661<br />
Team_I_Chemie_FQF6 0,661 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(a) Evaluations according to FQF 5.0 and FQF 6.0.<br />
(b) Distribution comparisons summary<br />
Figure 65: Distribution comparisons between the evaluation scores of chemical-part suppliers<br />
located in Region I, performed according to different versions of Formel Q-Fähigkeit by the local<br />
audit team.<br />
141
Team_C_Elektrik_FQF5<br />
Team_C_Elektrik_FQF6<br />
Team_C_Elektrik_FQF5<br />
Team_C_Elektrik_FQF6<br />
Distribution of Team_C_Elektrik_FQF5<br />
Frequency<br />
0 1 2 3 4 5 6<br />
65 70 75 80 85 90 95 100<br />
Shapiro-Wilk Normality test p-values<br />
Distribution of Team_C_Elektrik_FQF6<br />
Frequency<br />
0 1 2 3 4 5 6 7<br />
65 70 75 80 85 90 95 100<br />
p-values 0,017 0,001<br />
Samples 50 55<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_C_Elektrik_FQF5<br />
and Team_C_Elektrik_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.1673<br />
p = 0.4562<br />
n_1 = 50<br />
n_2 = 55<br />
Team_C_Elektrik_FQF5<br />
Team_C_Elektrik_FQF6<br />
65 70 75 80 85 90 95 100<br />
Evaluation Scores [%]<br />
(a) Evaluations according to FQF 5.0 and FQF 6.0.<br />
Team_C_Elektrik_FQF5 1,000 0,456<br />
Team_C_Elektrik_FQF6 0,456 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(b) Distribution comparisons summary<br />
Figure 66: Distribution comparisons between the evaluation scores of electrical-part suppliers<br />
located in Region C, performed according to different versions of Formel Q-Fähigkeit by the local<br />
audit team.<br />
142
Team_E_Elektrik_FQF5<br />
Team_E_Elektrik_FQF6<br />
Team_E_Elektrik_FQF5<br />
Team_E_Elektrik_FQF6<br />
Distribution of Team_E_Elektrik_FQF5<br />
Frequency<br />
0 2 4 6 8 10<br />
75 80 85 90 95 100<br />
Shapiro-Wilk Normality test p-values<br />
Distribution of Team_E_Elektrik_FQF6<br />
Frequency<br />
0 1 2 3 4 5 6 7<br />
75 80 85 90 95 100<br />
p-values 0,005 0,017<br />
Samples 52 23<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_E_Elektrik_FQF5<br />
and Team_E_Elektrik_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.3403<br />
p = 0.04977<br />
n_1 = 52<br />
n_2 = 23<br />
Team_E_Elektrik_FQF5<br />
Team_E_Elektrik_FQF6<br />
75 80 85 90 95 100<br />
Evaluation Scores [%]<br />
(a) Evaluations according to FQF 5.0 and FQF 6.0.<br />
Team_E_Elektrik_FQF5 1,000 0,050<br />
Team_E_Elektrik_FQF6 0,050 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(b) Distribution comparisons summary<br />
Figure 67: Distribution comparisons between the evaluation scores of electrical-part suppliers<br />
located in Region E, performed according to different versions of Formel Q-Fähigkeit by the local<br />
audit team.<br />
143
Distribution of Team_F_Elektrik_FQF4<br />
Distribution of Team_F_Elektrik_FQF5<br />
Frequency<br />
0 5 10 15 20<br />
70 80 90 100<br />
Distribution of Team_F_Elektrik_FQF5<br />
Frequency<br />
0 5 10 15<br />
Frequency<br />
0 5 10 15<br />
70 80 90 100<br />
Distribution of Team_F_Elektrik_FQF6<br />
Frequency<br />
0 2 4 6 8 10<br />
Team_F_Elektrik_FQF4<br />
Team_F_Elektrik_FQF5<br />
Team_F_Elektrik_FQF6<br />
Team_F_Elektrik_FQF4<br />
Team_F_Elektrik_FQF5<br />
Team_F_Elektrik_FQF6<br />
70 80 90 100<br />
70 80 90 100<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_F_Elektrik_FQF4<br />
and Team_F_Elektrik_FQF5<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.2556<br />
p = 0.009894<br />
n_1 = 126<br />
n_2 = 60<br />
Team_F_Elektrik_FQF4<br />
Team_F_Elektrik_FQF5<br />
70 80 90 100<br />
Evaluation Scores [%]<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_F_Elektrik_FQF5<br />
and Team_F_Elektrik_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.2358<br />
p = 0.1333<br />
n_1 = 60<br />
n_2 = 41<br />
Team_F_Elektrik_FQF5<br />
Team_F_Elektrik_FQF6<br />
70 80 90 100<br />
Evaluation Scores [%]<br />
(a) Evaluations according to FQF 4.0 and FQF 5.0.<br />
(b) Evaluations according to FQF 5.0 and FQF 6.0.<br />
Shapiro-Wilk Normality test p-values<br />
p-values 1,7E-08 1,3E-08 0,001<br />
Samples 126 60 41<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
Team_F_Elektrik_FQF4 1,000 0,010 0,024<br />
Team_F_Elektrik_FQF5 0,010 1,000 0,133<br />
Team_F_Elektrik_FQF6 0,024 0,133 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(c) Distribution comparisons summary<br />
Figure 68: Distribution comparisons between the evaluation scores of electrical-part suppliers<br />
located in Region F, performed according to different versions of Formel Q-Fähigkeit by the local<br />
audit team.<br />
144
Team_A_Metal_FQF5<br />
Team_A_Metal_FQF6<br />
Team_A_Metal_FQF5<br />
Team_A_Metal_FQF6<br />
Distribution of Team_A_Metal_FQF5<br />
Frequency<br />
0 5 10 20 30<br />
50 60 70 80 90 100<br />
Shapiro-Wilk Normality test p-values<br />
Distribution of Team_A_Metal_FQF6<br />
Frequency<br />
0 5 10 20 30<br />
50 60 70 80 90 100<br />
p-values<br />
1,5E-12 1,1E-09<br />
Samples 209 303<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_A_Metal_FQF5<br />
and Team_A_Metal_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.1094<br />
p = 0.1037<br />
n_1 = 209<br />
n_2 = 303<br />
Team_A_Metal_FQF5<br />
Team_A_Metal_FQF6<br />
50 60 70 80 90 100<br />
Evaluation Scores [%]<br />
(a) Evaluations according to FQF 5.0 and FQF 6.0.<br />
Team_A_Metal_FQF5 1,000 0,104<br />
Team_A_Metal_FQF6 0,104 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(b) Distribution comparisons summary<br />
Figure 69: Distribution comparisons between the evaluation scores of metal-part suppliers located<br />
in Region A, performed according to different versions of Formel Q-Fähigkeit by the local audit<br />
team.<br />
145
Distribution of Team_B_Metal_FQF4<br />
Distribution of Team_B_Metal_FQF5<br />
Frequency<br />
0 5 10 15<br />
60 70 80 90 100<br />
Distribution of Team_B_Metal_FQF5<br />
Frequency<br />
0 2 4 6 8 10 14<br />
Frequency<br />
0 2 4 6 8 10 14<br />
80 85 90 95 100<br />
Distribution of Team_B_Metal_FQF6<br />
Frequency<br />
0 2 4 6 8 10<br />
Team_B_Metal_FQF4<br />
Team_B_Metal_FQF5<br />
Team_B_Metal_FQF6<br />
Team_B_Metal_FQF4<br />
Team_B_Metal_FQF5<br />
Team_B_Metal_FQF6<br />
60 70 80 90 100<br />
80 85 90 95 100<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_B_Metal_FQF4<br />
and Team_B_Metal_FQF5<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.2226<br />
p = 0.04086<br />
n_1 = 126<br />
n_2 = 57<br />
Team_B_Metal_FQF4<br />
Team_B_Metal_FQF5<br />
60 70 80 90 100<br />
Evaluation Scores [%]<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_B_Metal_FQF5<br />
and Team_B_Metal_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.1602<br />
p = 0.5309<br />
n_1 = 57<br />
n_2 = 46<br />
Team_B_Metal_FQF5<br />
Team_B_Metal_FQF6<br />
80 85 90 95 100<br />
Evaluation Scores [%]<br />
(a) Evaluations according to FQF 4.0 and FQF 5.0.<br />
(b) Evaluations according to FQF 5.0 and FQF 6.0.<br />
Shapiro-Wilk Normality test p-values<br />
p-values 7,0E-10 0,002 0,002<br />
Samples 126 57 46<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
Team_B_Metal_FQF4 1,000 0,041 0,479<br />
Team_B_Metal_FQF5 0,041 1,000 0,531<br />
Team_B_Metal_FQF6 0,479 0,531 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(c) Distribution comparisons summary<br />
Figure 70: Distribution comparisons between the evaluation scores of metal-part suppliers located<br />
in Region B, performed according to different versions of Formel Q-Fähigkeit by the local audit<br />
team.<br />
146
Team_C_Metal_FQF5<br />
Team_C_Metal_FQF6<br />
Team_C_Metal_FQF5<br />
Team_C_Metal_FQF6<br />
Distribution of Team_C_Metal_FQF5<br />
Frequency<br />
0 5 10 20 30<br />
70 75 80 85 90 95 100<br />
Shapiro-Wilk Normality test p-values<br />
Distribution of Team_C_Metal_FQF6<br />
Frequency<br />
0 5 10 15 20 25<br />
70 75 80 85 90 95 100<br />
p-values<br />
4,6E-06 6,4E-07<br />
Samples 171 157<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_C_Metal_FQF5<br />
and Team_C_Metal_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.2306<br />
p = 0.0003314<br />
n_1 = 171<br />
n_2 = 157<br />
Team_C_Metal_FQF5<br />
Team_C_Metal_FQF6<br />
70 75 80 85 90 95 100<br />
Evaluation Scores [%]<br />
(a) Evaluations according to FQF 5.0 and FQF 6.0.<br />
Team_C_Metal_FQF5 1,000 3,3E-04<br />
Team_C_Metal_FQF6 3,3E-04 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(b) Distribution comparisons summary<br />
Figure 71: Distribution comparisons between the evaluation scores of metal-part suppliers located<br />
in Region C, performed according to different versions of Formel Q-Fähigkeit by the local audit<br />
team.<br />
147
Team_D_Metal_FQF5<br />
Team_D_Metal_FQF6<br />
Team_D_Metal_FQF5<br />
Team_D_Metal_FQF6<br />
Distribution of Team_D_Metal_FQF5<br />
Frequency<br />
0 2 4 6 8 10<br />
50 60 70 80 90 100<br />
Shapiro-Wilk Normality test p-values<br />
Distribution of Team_D_Metal_FQF6<br />
Frequency<br />
0 5 10 20 30<br />
50 60 70 80 90 100<br />
p-values<br />
4,8E-07 5,8E-06<br />
Samples 88 116<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_D_Metal_FQF5<br />
and Team_D_Metal_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.3433<br />
p = 1.513e−05<br />
n_1 = 88<br />
n_2 = 116<br />
Team_D_Metal_FQF5<br />
Team_D_Metal_FQF6<br />
50 60 70 80 90 100<br />
Evaluation Scores [%]<br />
(a) Evaluations according to FQF 5.0 and FQF 6.0.<br />
Team_D_Metal_FQF5 1,000 1,5E-05<br />
Team_D_Metal_FQF6 1,5E-05 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(b) Distribution comparisons summary<br />
Figure 72: Distribution comparisons between the evaluation scores of metal-part suppliers located<br />
in Region D, performed according to different versions of Formel Q-Fähigkeit by the local audit<br />
team.<br />
148
Team_E_Metal_FQF5<br />
Team_E_Metal_FQF6<br />
Team_E_Metal_FQF5<br />
Team_E_Metal_FQF6<br />
Distribution of Team_E_Metal_FQF5<br />
Frequency<br />
0 2 4 6 8 10 14<br />
70 80 90 100<br />
Shapiro-Wilk Normality test p-values<br />
Distribution of Team_E_Metal_FQF6<br />
Frequency<br />
0 5 10 15<br />
70 80 90 100<br />
p-values 1,5E-04 0,006<br />
Samples 111 125<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_E_Metal_FQF5<br />
and Team_E_Metal_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.1103<br />
p = 0.4722<br />
n_1 = 111<br />
n_2 = 125<br />
Team_E_Metal_FQF5<br />
Team_E_Metal_FQF6<br />
70 80 90 100<br />
Evaluation Scores [%]<br />
(a) Evaluations according to FQF 5.0 and FQF 6.0.<br />
Team_E_Metal_FQF5 1,000 0,472<br />
Team_E_Metal_FQF6 0,472 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(b) Distribution comparisons summary<br />
Figure 73: Distribution comparisons between the evaluation scores of metal-part suppliers located<br />
in Region E, performed according to different versions of Formel Q-Fähigkeit by the local audit<br />
team.<br />
149
Distribution of Team_F_Metal_FQF4<br />
Distribution of Team_F_Metal_FQF5<br />
Frequency<br />
0 10 20 30 40 50 60<br />
70 80 90 100<br />
Distribution of Team_F_Metal_FQF5<br />
Frequency<br />
0 10 20 30 40<br />
Frequency<br />
0 10 20 30 40<br />
70 80 90 100<br />
Distribution of Team_F_Metal_FQF6<br />
Frequency<br />
0 5 10 20 30<br />
Team_F_Metal_FQF4<br />
Team_F_Metal_FQF5<br />
Team_F_Metal_FQF6<br />
Team_F_Metal_FQF4<br />
Team_F_Metal_FQF5<br />
Team_F_Metal_FQF6<br />
70 80 90 100<br />
70 80 90 100<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_F_Metal_FQF4<br />
and Team_F_Metal_FQF5<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.2195<br />
p = 5.419e−06<br />
n_1 = 310<br />
n_2 = 233<br />
Team_F_Metal_FQF4<br />
Team_F_Metal_FQF5<br />
70 80 90 100<br />
Evaluation Scores [%]<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_F_Metal_FQF5<br />
and Team_F_Metal_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.08616<br />
p = 0.4483<br />
n_1 = 233<br />
n_2 = 175<br />
Team_F_Metal_FQF5<br />
Team_F_Metal_FQF6<br />
70 80 90 100<br />
Evaluation Scores [%]<br />
(a) Evaluations according to FQF 4.0 and FQF 5.0.<br />
(b) Evaluations according to FQF 5.0 and FQF 6.0.<br />
Shapiro-Wilk Normality test p-values<br />
p-values<br />
1,9E-14 4,8E-09 1,2E-13<br />
Samples 310 233 175<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
Team_F_Metal_FQF4 1,000 5,4E-06 3,7E-04<br />
Team_F_Metal_FQF5 5,4E-06 1,000 0,448<br />
Team_F_Metal_FQF6 3,7E-04 0,448 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(c) Distribution comparisons summary<br />
Figure 74: Distribution comparisons between the evaluation scores of metal-part suppliers located<br />
in Region F, performed according to different versions of Formel Q-Fähigkeit by the local audit<br />
team.<br />
150
Team_G_Metal_FQF5<br />
Team_G_Metal_FQF6<br />
Team_G_Metal_FQF5<br />
Team_G_Metal_FQF6<br />
Distribution of Team_G_Metal_FQF5<br />
Frequency<br />
0 5 10 15 20 25<br />
75 80 85 90 95 100<br />
Shapiro-Wilk Normality test p-values<br />
Distribution of Team_G_Metal_FQF6<br />
Frequency<br />
0 10 20 30<br />
75 80 85 90 95 100<br />
p-values<br />
8,4E-07 3,0E-10<br />
Samples 85 89<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_G_Metal_FQF5<br />
and Team_G_Metal_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.1309<br />
p = 0.4459<br />
n_1 = 85<br />
n_2 = 89<br />
Team_G_Metal_FQF5<br />
Team_G_Metal_FQF6<br />
75 80 85 90 95 100<br />
Evaluation Scores [%]<br />
(a) Evaluations according to FQF 5.0 and FQF 6.0.<br />
Team_G_Metal_FQF5 1,000 0,446<br />
Team_G_Metal_FQF6 0,446 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(b) Distribution comparisons summary<br />
Figure 75: Distribution comparisons between the evaluation scores of metal-part suppliers located<br />
in Region G, performed according to different versions of Formel Q-Fähigkeit by the local audit<br />
team.<br />
151
Team_H_Metal_FQF5<br />
Team_H_Metal_FQF6<br />
Team_H_Metal_FQF5<br />
Team_H_Metal_FQF6<br />
Distribution of Team_H_Metal_FQF5<br />
Frequency<br />
0 5 10 20 30<br />
50 60 70 80 90 100<br />
Shapiro-Wilk Normality test p-values<br />
Distribution of Team_H_Metal_FQF6<br />
Frequency<br />
0 5 15 25 35<br />
50 60 70 80 90 100<br />
p-values<br />
4,2E-14 2,1E-13<br />
Samples 160 185<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_H_Metal_FQF5<br />
and Team_H_Metal_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.3093<br />
p = 1.487e−07<br />
n_1 = 160<br />
n_2 = 185<br />
Team_H_Metal_FQF5<br />
Team_H_Metal_FQF6<br />
50 60 70 80 90 100<br />
Evaluation Scores [%]<br />
(a) Evaluations according to FQF 5.0 and FQF 6.0.<br />
Team_H_Metal_FQF5 1,000 1,5E-07<br />
Team_H_Metal_FQF6 1,5E-07 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(b) Distribution comparisons summary<br />
Figure 76: Distribution comparisons between the evaluation scores of metal-part suppliers located<br />
in Region H, performed according to different versions of Formel Q-Fähigkeit by the local audit<br />
team.<br />
152
Distribution of Team_I_Metal_FQF4<br />
Distribution of Team_I_Metal_FQF5<br />
Frequency<br />
0.0 1.0 2.0 3.0<br />
80 85 90 95 100<br />
Frequency<br />
0 1 2 3 4 5<br />
65 70 75 80 85 90 95 100<br />
Distribution of Team_I_Metal_FQF5<br />
Distribution of Team_I_Metal_FQF6<br />
Frequency<br />
0 1 2 3 4 5<br />
Frequency<br />
0 2 4 6 8 10<br />
Team_I_Metal_FQF4<br />
Team_I_Metal_FQF5<br />
Team_I_Metal_FQF6<br />
Team_I_Metal_FQF4<br />
Team_I_Metal_FQF5<br />
Team_I_Metal_FQF6<br />
80 85 90 95 100<br />
65 70 75 80 85 90 95 100<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_I_Metal_FQF4<br />
and Team_I_Metal_FQF5<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.3271<br />
p = 0.1774<br />
n_1 = 19<br />
n_2 = 28<br />
Team_I_Metal_FQF4<br />
Team_I_Metal_FQF5<br />
80 85 90 95 100<br />
Evaluation Scores [%]<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_I_Metal_FQF5<br />
and Team_I_Metal_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.09921<br />
p = 0.9934<br />
n_1 = 28<br />
n_2 = 54<br />
Team_I_Metal_FQF5<br />
Team_I_Metal_FQF6<br />
65 70 75 80 85 90 95 100<br />
Evaluation Scores [%]<br />
(a) Evaluations according to FQF 4.0 and FQF 5.0.<br />
(b) Evaluations according to FQF 5.0 and FQF 6.0.<br />
Shapiro-Wilk Normality test p-values<br />
p-values 0,694 0,647 7,7E-08<br />
Samples 19 28 54<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
Team_I_Metal_FQF4 1,000 0,177 0,063<br />
Team_I_Metal_FQF5 0,177 1,000 0,993<br />
Team_I_Metal_FQF6 0,063 0,993 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(c) Distribution comparisons summary<br />
Figure 77: Distribution comparisons between the evaluation scores of metal-part suppliers located<br />
in Region I, performed according to different versions of Formel Q-Fähigkeit by the local audit<br />
team.<br />
153
Team_J_Metal_FQF5<br />
Team_J_Metal_FQF6<br />
Team_J_Metal_FQF5<br />
Team_J_Metal_FQF6<br />
Distribution of Team_J_Metal_FQF5<br />
Frequency<br />
0 1 2 3 4 5<br />
80 85 90 95 100<br />
Shapiro-Wilk Normality test p-values<br />
Distribution of Team_J_Metal_FQF6<br />
Frequency<br />
0 2 4 6 8 10<br />
80 85 90 95 100<br />
p-values 0,036 0,146<br />
Samples 31 72<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
CDFs for Team_J_Metal_FQF5<br />
and Team_J_Metal_FQF6<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.1962<br />
p = 0.3744<br />
n_1 = 31<br />
n_2 = 72<br />
Team_J_Metal_FQF5<br />
Team_J_Metal_FQF6<br />
80 85 90 95 100<br />
Evaluation Scores [%]<br />
(a) Evaluations according to FQF 5.0 and FQF 6.0.<br />
Team_J_Metal_FQF5 1,000 0,374<br />
Team_J_Metal_FQF6 0,374 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(b) Distribution comparisons summary<br />
Figure 78: Distribution comparisons between the evaluation scores of metal-part suppliers located<br />
in Region J, performed according to different versions of Formel Q-Fähigkeit by the local audit<br />
team.<br />
154
B Results from the analysis presented in Section 7.2.3.2<br />
Distribution of E_G_WSGR_0058_Plant_E_ppm<br />
Frequency<br />
0 1 2 3 4 5 6<br />
Frequency<br />
0 1 2 3 4 5 6 7<br />
Plant: E<br />
WSGR: 0058<br />
Plant: G<br />
WSGR: 0058<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
Plant: E<br />
WSGR: 0058<br />
Plant: G<br />
WSGR: 0058<br />
1 2 3 4<br />
Distribution of E_G_WSGR_0058_Plant_G_ppm<br />
1 2 3 4<br />
CDFs for E_G_WSGR_0058_Plant_E_ppm<br />
and E_G_WSGR_0058_Plant_G_ppm<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.2667<br />
p = 0.6781<br />
n_1 = 15<br />
n_2 = 15<br />
E_G_WSGR_0058_Plant_E_ppm<br />
E_G_WSGR_0058_Plant_G_ppm<br />
1 2 3 4<br />
ppm Scores [log]<br />
Shapiro-Wilk Normality test p-values<br />
p-values 0,015 0,028<br />
Samples 15 15<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
Plant: E<br />
WSGR: 0058<br />
Plant: G<br />
WSGR: 0058<br />
1,000 0,678<br />
0,678 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(a) Performance records comparison based on ppm<br />
(b) Distribution comparison summary<br />
Figure 79: Comparison between the quality performance records of the same 15 suppliers reported<br />
by two different plants – Plant E and Plant G. The reported defective parts per million (ppm) refer<br />
to quality problems of components from the material group 0058 (”Moulded parts < DIN A4 for<br />
body”).<br />
155
Distribution of H_I_WSGR_0058_Plant_H_ppm<br />
Frequency<br />
0.0 1.0 2.0 3.0<br />
Frequency<br />
0 1 2 3 4<br />
Plant: H<br />
WSGR: 0058<br />
Plant: I<br />
WSGR: 0058<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
Plant: H<br />
WSGR: 0058<br />
Plant: I<br />
WSGR: 0058<br />
0 1 2 3<br />
Distribution of H_I_WSGR_0058_Plant_I_ppm<br />
0 1 2 3<br />
CDFs for H_I_WSGR_0058_Plant_H_ppm<br />
and H_I_WSGR_0058_Plant_I_ppm<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.3<br />
p = 0.7869<br />
n_1 = 10<br />
n_2 = 10<br />
H_I_WSGR_0058_Plant_H_ppm<br />
H_I_WSGR_0058_Plant_I_ppm<br />
0 1 2 3<br />
ppm Scores [log]<br />
Shapiro-Wilk Normality test p-values<br />
p-values 0,601 0,044<br />
Samples 10 10<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
Plant: H<br />
WSGR: 0058<br />
Plant: I<br />
WSGR: 0058<br />
1,000 0,787<br />
0,787 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(a) Performance records comparison based on ppm<br />
(b) Distribution comparison summary<br />
Figure 80: Comparison between the quality performance records of the same 10 suppliers reported<br />
by two different plants – Plant H and Plant I. The reported defective parts per million (ppm) refer<br />
to quality problems of components from the material group 0058 (”Moulded parts < DIN A4 for<br />
body”).<br />
156
Distribution of O_P_WSGR_0058_Plant_O_ppm<br />
Frequency<br />
0 1 2 3 4 5 6 7<br />
Frequency<br />
0 1 2 3 4 5 6<br />
Plant: O<br />
WSGR: 0058<br />
Plant: P<br />
WSGR: 0058<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
Plant: O<br />
WSGR: 0058<br />
Plant: P<br />
WSGR: 0058<br />
0 1 2 3 4<br />
Distribution of O_P_WSGR_0058_Plant_P_ppm<br />
0 1 2 3 4<br />
CDFs for O_P_WSGR_0058_Plant_O_ppm<br />
and O_P_WSGR_0058_Plant_P_ppm<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.4667<br />
p = 0.07626<br />
n_1 = 15<br />
n_2 = 15<br />
O_P_WSGR_0058_Plant_O_ppm<br />
O_P_WSGR_0058_Plant_P_ppm<br />
0 1 2 3 4<br />
ppm Scores [log]<br />
Shapiro-Wilk Normality test p-values<br />
p-values 0,002 0,009<br />
Samples 15 15<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
Plant: O<br />
WSGR: 0058<br />
Plant: P<br />
WSGR: 0058<br />
1,000 0,076<br />
0,076 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(a) Performance records comparison based on ppm<br />
(b) Distribution comparison summary<br />
Figure 81: Comparison between the quality performance records of the same 15 suppliers reported<br />
by two different plants – Plant O and Plant P. The reported defective parts per million (ppm) refer<br />
to quality problems of components from the material group 0058 (”Moulded parts < DIN A4 for<br />
body”).<br />
157
Distribution of B_K_WSGR_0068_Plant_B_ppm<br />
Frequency<br />
0.0 1.0 2.0 3.0<br />
Frequency<br />
0 1 2 3 4 5 6<br />
Plant: B<br />
WSGR: 0068<br />
Plant: K<br />
WSGR: 0068<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
Plant: B<br />
WSGR: 0068<br />
Plant: K<br />
WSGR: 0068<br />
0 1 2 3 4<br />
Distribution of B_K_WSGR_0068_Plant_K_ppm<br />
0 1 2 3 4<br />
CDFs for B_K_WSGR_0068_Plant_B_ppm<br />
and B_K_WSGR_0068_Plant_K_ppm<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.2667<br />
p = 0.6604<br />
n_1 = 15<br />
n_2 = 15<br />
B_K_WSGR_0068_Plant_B_ppm<br />
B_K_WSGR_0068_Plant_K_ppm<br />
0 1 2 3 4<br />
ppm Scores [log]<br />
Shapiro-Wilk Normality test p-values<br />
p-values 0,720 0,085<br />
Samples 15 15<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
Plant: B<br />
WSGR: 0068<br />
Plant: K<br />
WSGR: 0068<br />
1,000 0,660<br />
0,660 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(a) Performance records comparison based on ppm<br />
(b) Distribution comparison summary<br />
Figure 82: Comparison between the quality performance records of the same 15 suppliers reported<br />
by two different plants – Plant B and Plant K. The reported defective parts per million (ppm) refer<br />
to quality problems of components from the material group 0068 (”Moulded parts > DIN A4 for<br />
body”).<br />
158
Distribution of H_I_WSGR_0068_Plant_H_ppm<br />
Frequency<br />
0 1 2 3 4<br />
Frequency<br />
0.0 1.0 2.0 3.0<br />
Plant: H<br />
WSGR: 0068<br />
Plant: I<br />
WSGR: 0068<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
Plant: H<br />
WSGR: 0068<br />
Plant: I<br />
WSGR: 0068<br />
0 1 2 3<br />
Distribution of H_I_WSGR_0068_Plant_I_ppm<br />
0 1 2 3<br />
CDFs for H_I_WSGR_0068_Plant_H_ppm<br />
and H_I_WSGR_0068_Plant_I_ppm<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.3571<br />
p = 0.3338<br />
n_1 = 14<br />
n_2 = 14<br />
H_I_WSGR_0068_Plant_H_ppm<br />
H_I_WSGR_0068_Plant_I_ppm<br />
0 1 2 3<br />
ppm Scores [log]<br />
Shapiro-Wilk Normality test p-values<br />
p-values 0,219 0,851<br />
Samples 14 14<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
Plant: H<br />
WSGR: 0068<br />
Plant: I<br />
WSGR: 0068<br />
1,000 0,334<br />
0,334 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(a) Performance records comparison based on ppm<br />
(b) Distribution comparison summary<br />
Figure 83: Comparison between the quality performance records of the same 14 suppliers reported<br />
by two different plants – Plant H and Plant I. The reported defective parts per million (ppm) refer<br />
to quality problems of components from the material group 0068 (”Moulded parts > DIN A4 for<br />
body”).<br />
159
Distribution of N_O_WSGR_0068_Plant_N_ppm<br />
Frequency<br />
0 1 2 3 4 5<br />
Frequency<br />
0 1 2 3 4 5<br />
Plant: N<br />
WSGR: 0068<br />
Plant: O<br />
WSGR: 0068<br />
f(x)<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
Plant: N<br />
WSGR: 0068<br />
Plant: O<br />
WSGR: 0068<br />
0 1 2 3 4<br />
Distribution of N_O_WSGR_0068_Plant_O_ppm<br />
0 1 2 3 4<br />
CDFs for N_O_WSGR_0068_Plant_N_ppm<br />
and N_O_WSGR_0068_Plant_O_ppm<br />
Two−sample Kolmogorov−Smirnov test<br />
D = 0.1875<br />
p = 0.9412<br />
n_1 = 16<br />
n_2 = 16<br />
N_O_WSGR_0068_Plant_N_ppm<br />
N_O_WSGR_0068_Plant_O_ppm<br />
0 1 2 3 4<br />
ppm Scores [log]<br />
Shapiro-Wilk Normality test p-values<br />
p-values 0,269 0,124<br />
Samples 16 16<br />
Kolmogorov-Smirnov Distribution Comparison test<br />
p-values matrix<br />
Plant: N<br />
WSGR: 0068<br />
Plant: O<br />
WSGR: 0068<br />
1,000 0,941<br />
0,941 1,000<br />
significant on the 10% significance level<br />
significant on the 5% significance level<br />
(a) Performance records comparison based on ppm<br />
(b) Distribution comparison summary<br />
Figure 84: Comparison between the quality performance records of the same 16 suppliers reported<br />
by two different plants – Plant N and Plant O. The reported defective parts per million (ppm) refer<br />
to quality problems of components from the material group 0068 (”Moulded parts > DIN A4 for<br />
body”).<br />
160
B.1 List of Material Groups<br />
Table 6 includes the codes of all material groups which appear in the analysis presented in this<br />
paper and their names. Note that the complete definition of the Volkswagen material groups comprises<br />
of three levels of detail. However, here only the second level is listed, since this is the only<br />
material group level available in suppliers’ quality performance records analyzed here.<br />
Table 6: List of product groups and their names. (Source: Volkswagen AG, 2010)<br />
Material Group Number Description Number of Subgroups<br />
0058 Moulded parts < DIN A4 for body 2<br />
0068 Moulded parts > DIN A4 for body 4<br />
0096 Reservoirs, covers, pipes, wires 15<br />
0099 Column trim, sill plates 4<br />
0105 Bumper, Spoiler 2<br />
0302 Light metal cast engine parts 1<br />
0337 Light metal cast gearbox parts 4<br />
0480 Pipes, pipe parts per drawing 13<br />
0485 Swivel plates 1<br />
161
References<br />
Abele, E., Meyer, T., Näher, U., Strube, G., & Sykes, R. (2008). Global production: A handbook<br />
for strategy and implementation. Springer-Verlag Berlin Heidelberg.<br />
Agência AutoData Weekly Edition (multiple editions). (2009–2012). AutoData Editora Ltda. São<br />
Paulo, Brazil.<br />
Arter, D. R. (1989). Quality audits for improved performance. ASQC Quality Press.<br />
Benton, Jr., W. C. (2007). Purchasing and supply management. McGraw-Hill.<br />
Bunkley, N., & Maynard, M. (2010, 29 January). With recall expanding, Toyota gives an apology.<br />
The New York Times. (Accessed on http://www.nytimes.com/2010/01/<br />
30/business/30toyota.html? r=1 in August, 2011)<br />
Business (definition). (2012). Cambridge Dictionaries Online. (Accessed on http://<br />
dictionary.cambridge.org/ in May, 2012)<br />
Car sales slump, but exports pick up. (2011, 13 October). Economist Inteligence Unit (Access<br />
China newsletter).<br />
Choy, K., Lee, W., & Lo, V. (2002). An intelligent supplier management tool for benchmarking<br />
suppliers in outsource manufacturing. Expert Systems with Applications, 22, 213–224.<br />
Cook, R. D. (1977, February). Detection of influential observation in linear regression. Technometrics,<br />
19(1), 15–18.<br />
Cook, R. D. (1979, March). Influential observations in linear regression. Journal of the American<br />
Statistical Association, 74(365), 169–174.<br />
D’Antona, G., & Ferrero, A. (2006). Digital signal processing for measurements systems: Theory<br />
and applications. Springer Science + Business Media, Inc.<br />
Davis, T. (1993, Summer). Effective supply chain management. Sloan Management Review,<br />
35–46.<br />
DIN ISO 19011: Leitfaden zur Auditierung von Managementsystemen (ISO 19011:2011). (2011,<br />
December). DIN Deutsches Insistut für Normung e.V.<br />
DIN ISO 9000: Qualitätsmanagementsysteme – Grundlagen und Begriffe (ISO 9000:2005); Dreisprachige<br />
Fassung EN ISO 9000:2005. (2005, December). DIN Deutsches Insistut für Normung<br />
e.V.<br />
DIN ISO 9001: Qualitätsmanagementsysteme – Anforderungen (ISO 9001:2008); Dreisprachige<br />
Fassung EN ISO 9001:2008. (2008, December). DIN Deutsches Insistut für Normung e.V.<br />
DIN ISO/TS 16949: Qualitätsmanagementsysteme – Besondere Anforderungnen bei Anwendung<br />
von ISO 9001:2008 für die Serien- und Ersatzteil-Produktion in der Automobilindustrie<br />
(ISO/TS 16949:2009). (2009, November). DIN Deutsches Insistut für Normung e.V.<br />
Dun & Bradstreet (D&B). (2012). (Accessed on http://www.dnb.com/ in 2012)<br />
The EFQM Excellence Model. (2012). European Foundation for Quality Management. (Accessed<br />
on http://www.efqm.org in May, 2012)<br />
Faraway, J. J. (2002, July). Practical regression and anova using R.<br />
Flat file copies of NHTSA/ODI databases: Recalls. (2011). National Highway Traffic<br />
Safety Administration. Washington DC, USA. (Accessed in September, 2011 via<br />
www.safecar.gov)<br />
Formel Q-Capability: Quality Capability Suppliers Assessment Guidelines. (2009, August).<br />
Volkswagen AG. Wolfsburg, Germany. (sixth edition (English))<br />
Formel Q-Fähigkeit: Qualitätsfähigkeit Lieferanten Beurteilungsrichtlinie. (2009, August).<br />
Volkswagen AG. Wolfsburg, Germany. (sixth edition)<br />
Formel Q-konkret: Qualitätsmanagementvereinbarung zwischen den Gesellschaften des<br />
VOLKSWAGEN-KONZERNS und seinen Lieferanten. (2008, September). Volkswagen AG.<br />
Wolfsburg, Germany. (fourth edition)<br />
162
Formula Q-concrete: Quality management agreement between the companies of the<br />
VOLKSWAGEN GROUP and its suppliers. (2008, September). Volkswagen AG. Wolfsburg,<br />
Germany. (fourth edition (English))<br />
Gomes, C. (2011, March 29). Global auto industry faces component-shortage risk – cutbacks<br />
are spreading beyond Japan (Tech. Rep.). Scotiabank Group. (Global Economic Research:<br />
Global Auto Report)<br />
Green, D. (1997). ISO 9000 quality systems auditing. Gower.<br />
Gropp, M. (2009). Konzept zur erhöhung der effektivität von zertifizierungsaudits im<br />
qualitätsmanagement. Unpublished doctoral dissertation, Fakultät V – Verkehrs- und<br />
Maschinensysteme der Technischen Universität Berlin.<br />
Hadzhiev, B. (2009, August). Quality auditing in the automotive industry: In-depth methodology<br />
analysis. Bremen, Germany.<br />
Hendricks, K. B., & Singhal, V. R. (2000, March). The impact of total quality management (TQM)<br />
on financial performance: Evidence from quality award winners. <strong>University</strong> of Western<br />
Ontario / Georgia Institute of Technology.<br />
Hoyle, D. (2005). Automotive quality systems handbook: Incorporating ISO/TS 16949:2002<br />
(second ed.). Elsevier Butterworth-Heinemann.<br />
Karrenberg, U. (2002). An interactive multimedia introduction to signal processing (second ed.).<br />
Springer-Verlag Berlin Heidelberg.<br />
Lan, Y., & Unhelkar, B. (2006). Global integrated supply chain systems. London, United Kingdom:<br />
Idea Group Inc.<br />
Lee, E.-K., Ha, S., & Kim, S.-K. (2001, August). Supplier selection and management system<br />
considering relationships in supply chain management. IEEE Transactions on Engineering<br />
Management, 48(3), 307–318.<br />
Liker, J. K., & Choi, T. (2004, December). Building deep supplier relationships. Harvard Business<br />
Review, 104–113.<br />
Liker, J. K., Kamath, R. R., Wasti, S. N., & Nagamachi, M. (1996). Supplier involvement in<br />
automotive component design: Are there really large us japan differences. Research Policy,<br />
25, 59–89.<br />
Lorenz, F. O. (1987, April). Teaching about influence in simple regression. Teaching Sociology,<br />
15(2), 173–177. (Teaching Research Methods and Statistics)<br />
Manderscheid, L. V. (1965, December). Significance levels. 0.05, 0.001, or ? Journal of Farm<br />
Economics, 47(5), 1381–1385.<br />
Marques de Sá, J. P. (2007). Appliead statistics using SPSS, STATISTICA, MATLAB, and R.<br />
Springer-Verlag Berlin Heidelberg.<br />
McDonald, B. (2002). A teaching note on cook’s distance - a guideline. Institute of Information<br />
and Mathematical Science, Massey <strong>University</strong> at Albany, Auckland, New Zealand.<br />
Meyer-Baese, U. (2007). Digital signal processing with field programmable gate arrays (third<br />
ed.). Springer-Verlag Berlin Heidelberg.<br />
Meyr, H. (2004). Supply chain planning in the German automotive industry. OR Spectrum, 26,<br />
447–470.<br />
Napier, N. K., & Vu, V. T. (1998). International human resource management in developing and<br />
transitional economy countries: A breed apart? Human Resource Management Review, 8(1),<br />
39–77.<br />
Parsowith, B. S. (1995). Fundamentals of quality auditing. ASQC Quality Press.<br />
Patton, M. Q. (1987). How to use qualitative methods in evaluation. SAGE Publications, Inc.<br />
Pötsch, H. D. (2011, May 19). Volkswagen – driving forward. Deutsche Bank German and<br />
Austrian Corporate Conference. Frankfurt, Germany.<br />
Qualitätsmanagement in der Automobilindustrie: Prozessaudit. (2010, June). Verband der Automobilindustrie<br />
(VDA Band 6 Teil 3). (2. vollständig überarbeitete Auflage)<br />
163
Reichheld, F. F. (2003). The one number you need to grow. Harvard Business School Publishing<br />
Corporation.<br />
The road ahead. (2011). Automotive Industries Team – U.S. Department of Commerce.<br />
Robert B. Austenfeld, J. (2006). Toyota and why it is so successful. Papers of the Research Society<br />
of Commerce and Economics, 47(1), 100 – 173.<br />
Scullion, H., Collings, D. G., & Gunnigle, P. (2007). International human resource management<br />
in the 21st century: Emerging themes and contemporary debates. Human Resource Management<br />
Journal, 17(4), 309 – 319.<br />
Sinha, P. (2010). Speech processing in embedded systems. New York: Springer Science+Business<br />
Media.<br />
Spekman, R. E., Kamauff, J., & Spear, J. (1999). Towards more effective sourcing and supplier<br />
management. European Journal of Purchasing & Supply Management, 5, 103–116.<br />
Stroescu-Dabu, M. (2008). Assessment of the assessment method: a study of the effectiveness<br />
of supplier process audits to improve quality performance of suppliers in the automotive<br />
industry. Guided Research Project Report, <strong>Jacobs</strong> <strong>University</strong> Bremen. Bremen, Germany.<br />
Suthikarnnarunai, N. (2008). Automotive supply chain and logistics management. Proceedings of<br />
the International MultiConference of Engineers and Computer Scientists, 2, 19–21.<br />
van Weele, A. J. (2010). Purchasing and supply chain management: Analysis, strategy, planning<br />
and practice (fifth ed.). Hampshire, United Kingdom: Cengage Learning EMEA.<br />
Veloso, F., & Kumar, R. (2002, January). The automotive supply chain: Global trends and asian<br />
perspectives (Tech. Rep.). Asian Development Bank. (ERD Working Paper Series No. 3,<br />
Economic and Research Department)<br />
Venables, W. N., Smith, D. M., & the R Development Core Team. (2010). An introduction to<br />
R notes on R: A programming environment for data analysis and graphics version 2.13.0<br />
(2011-04-13). R Development Core Team.<br />
Verzani, J. (2002). simpleR – using R for introductory statistics. (Accessed in April, 2011)<br />
Volkswagen Internal Reports. (2012). Group Quality Assurance Supplier Audit, Volkswagen AG.<br />
Wolfsburg, Germany.<br />
Wealleans, D. (2005). The quality audit for ISO 9001:2000 (second ed.). Gower.<br />
Wittmann, J., & Bergholz, W. (2006). Wert der bewertung: Empirische untersuchung der wirksamkeit<br />
von lieferantenaudits. Qualität und Zuverlässigkeit, Carl Hansen Verlag, Munich,<br />
51, 38–42.<br />
World motor vehicle production by country - OICA correspondents survey (multiple editions).<br />
(1997 – 2010). International Organization of Motor Vehicle Manufacturers (OICA).<br />
164