MITIGATING RISK IN PHYSICAL AND NEW PRODUCT SUPPLY CHAINS
By
Richard Leland Kraude

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
Business Administration - Operations and Sourcing Management - Doctor of Philosophy
2017

ABSTRACT
MITIGATING RISK IN PHYSICAL AND NEW PRODUCT SUPPLY CHAINS
By
Richard Leland Kraude
Supply chain risk mitigation involves investing in strategic activities that minimize the
financial impact of a disruption to the flow of goods through a supply chain or to processes
within a firm. Strategic implementation of risk mitigation occurs along the supply chain with a
variety of activities such as collaboration, options contracts, and buffer capacity just to name a
few. These risk mitigation strategies require significant coordination and investment making it
imperative to efficiently allocate resources in order to achieve a positive return on investment.
Yet, the effectiveness of many of these strategies is not well known. This dissertation consists of
three essays that investigate the effectiveness of risk mitigation strategies, methods for
performance evaluation, and contextual factors that influence the interpretation and application
of them. Each essay investigates risk mitigation at a different level of strategic decision making,
beginning with the highest level, strategic, and proceeding to a more granular level with a focus
on tactical and then operational.
The first essay explores how strategic alliances influences new drug development project
failures. Developing new products is a complex process that is inherently risky because of the
enormous investment needed to develop them and the uncertainty of the product’s potential. This
risk is most prevalent for pharmaceutical drug development because of the strict requirements set
by the Food and Drug Administration and high cost of development. Many organizations are
adopting strategies aimed at mitigating this risk, including dispersing cost by alliancing with
other organizations. Yet, alliance partners bring additional resources and capabilities to bear on

new product development that may also reduce the likelihood of failure. I show that in early
clinical trials, exploratory alliances require alliance partners with diverse new product
development expertise to reduce the rate of failures, while in late stages exploitation alliance
reduce failure rate if their alliance partners have technological diversity.
Managing supply chain risk requires assessing the exposure a firm faces to various risks and
allocating resources to reduce the impact or likelihood of a disruption occurring. Unfortunately,
risk is often difficult to accurately quantify, exposing the process to bias by contextual factors. In
the second essay, I investigate how these biases may reduce the resilience of supply chains since
managing risk requires coordination across business partners. Moreover, this coordination
requires consistent terminology of the types of strategies being applied. This study applies the
behavior theory of the firm to show that a consistent taxonomy of supply chain risk mitigation
strategies can be developed based on their mechanisms of slack. I develop measures of flexibility
and redundancy in the context of risk management and address measurement challenges when
dealing with psychometric measures related to uncertainty. Using confirmatory factor analysis
and generalized structural equation modeling, I show that Japan has a significantly higher level
of perceived supply chain risk and significantly lower application of risk mitigation strategies
than two western culture countries, the USA and Australia.
The final chapter of the dissertation assesses risk mitigation investments on an efficiency
frontier by considering both the cost of the strategies and interdependencies among the different
stages of the supply chain, an aspect past work in risk mitigation has ignored. Risk mitigation
strategies were measured at the supplier, process, and customer segments of the supply chain and
their performance was evaluated simultaneously using a network data envelopment analysis.

This thesis is dedicated to my mother.
Thank you for providing the opportunity and motivation to achieve this goal.

iv

ACKNOWLEDGEMENTS

It is an interesting experience reaching the objective of a goal. Looking back at the time
when the goal was set, the end seemed distant and terminal. Yet, achieving a goal is simply
another step along our path and ultimately, a short period on the timeline of life. We never fully
grasp this until the previously set end point is surpassed. Navigating the journey to a goal
requires important guidance since we are unexperienced when we set forth. I am very grateful to
all of those whom have helped me reach this major life achievement. The first person I must
thank of those individuals is Sriram Narayanan. He was always very encouraging and one I
relied on the most to get here. I will be forever grateful for the effort and time he put in helping
me become a better researcher in this field. He always encouraged me to work harder and push
myself. Sriram led by example and is the hardest working person I know. I always found his
philosophy of work ethic interesting because as scholars of operations management, the
foundation of our field is to achieve the greatest output with least cost. I suppose you cannot tell
your doctoral students to publish as many high quality research papers as possible, but do not
work too hard to do so. I laugh and shudder when I think about my first paper I sent him. He
encouraged me to write a term paper for his methods seminar on dynamic panel regression
models because it aligned with a project we were beginning to work on. I laugh because at the
time I did not even what OLS stood for and shudder because of the implications that had on the
quality of the paper. Sriram’s patience is obviously one of his strongest traits and am thankful
above all else, for his patience with me.
Next, I must thank Srinivas Talluri for his role in accepting me into the doctoral program,
guidance on my dissertation, and providing a role model for what to become as a professor.

v

Srinivas’s expertise was instrumental in forming my dissertation topic and getting me to the
point I needed to graduate. I will never forget his seminar, primarily because he conducted his
entire lecture using an overhead projector in a time when they were long since obsolete. Scott
Duhadway was my closest friend while working on my Ph.D. and I am thankful that fate had us
begin the journey through the program together. There is not another person in this world that
has a personality as well aligned with mine to help each other survive the trials and tribulations
of getting a Ph.D. in the Supply Chain Management department at Michigan State University. I
thank Scott for matching me on enthusiasm and for always challenging me. We always pushed
the discussion in our lectures beyond where most Ph.D. students would because we both lacked a
filter and refused to accept anything until proven to us. The most memorable experiences I had
were with Scott, including transporting our newly bought white board on top of his car with our
hands out of the windows while it hailed. It was the largest whiteboard in any office and allowed
us to discuss ideas, to keep track of the goals we scored in soccer and the number of advanced
words we would sneak into our presentations for Dr. Narasimhan’s lecture.
I must also thank the rest of the faculty at Michigan State. Ram Narasimhan helped shape my
writing and thinking perhaps more than any other individual. His precision and critical analysis
always lead me to work harder. Brian Jacobs easily became my favorite faculty member because
he was kind, helpful, and the only one who appeared to be actually human. My other two
committee members, Tomas Hult and Vallabh Sambamurthy, provided the financial support
necessary to collect the data for this dissertation. Steve Melnyk, well, I cannot begin to describe
the influence he had on me. Anand Nair, Roger Calantone, John Hollenbeck, Linn Van Dyne are
other faculty members that were instrumental in my development as a scholar and whom I am
very grateful to for all the time they spent with me during my doctoral study.

vi

Finally, I would like to thank my entire family. My family was always supportive, even
though they had little idea of what I was trying to achieve. Without their financial and emotional
support, I would not have been able to buy my house or survive the low points I had to
overcome. As I think everyone would agree, the most thanks must go to my wife. Moving from
San Diego to Lansing was a sacrifice few could have made. While I will always argue that I
lived well on my doctoral stipend, having a physician for a wife who made ten times what I did,
definitely made life much nicer. However, the financial support was only minor compared to the
joy she brought me every day. Spending time in our garden, making dinner, and going on small
vacations provided immense emotional support. Finally, I have to thank Lily for always
spreading happiness to me, and everyone on the 4th floor when she was brought to work.

vii

TABLE OF CONTENTS

LIST OF TABLES ........................................................................................................................ x
LIST OF FIGURES ..................................................................................................................... xi
Chapter 1 – Introduction................................................................................................................ 1
Chapter 2 - Mitigating New Product Development Risk in the Pharmaceutical Industry ............ 6
1. Introduction ........................................................................................................................... 6
2. Drug Development Process ................................................................................................... 9
3. Literature Review ................................................................................................................ 12
3.1 Strategic Alliances in NPD ............................................................................................... 12
3.1.1 Exploration alliances .................................................................................................. 15
3.1.2 Exploitation alliances ................................................................................................. 17
3.2 Alliance Partner Characteristics ...................................................................................... 18
3.2.1 Exploration Alliance Diversity .................................................................................. 19
3.2.2 Exploitation Alliance Diversity ................................................................................. 22
4. Research Design ................................................................................................................... 24
4.1. Measures ........................................................................................................................... 25
4.1.1. Product Failure .......................................................................................................... 25
4.1.2. Risk Mitigation .......................................................................................................... 26
4.1.3. Alliance Diversity....................................................................................................... 26
4.1.4. Controls ...................................................................................................................... 28
4.2. Methods ............................................................................................................................. 31
5. Results ................................................................................................................................... 32
5.1. Robustness......................................................................................................................... 36
5.1.1. Endogeneity ................................................................................................................ 37
6. Conclusion ............................................................................................................................ 37
Chapter 3 - The Impact of Culture on Supply Chain Risk Perception and Mitigation ............... 41
1. Introduction ......................................................................................................................... 41
2. Theoretical Development .................................................................................................... 46
2.1. Slack............................................................................................................................... 46
2.1.1. Available Slack ...................................................................................................... 47
2.1.2. Unavailable Slack .................................................................................................. 49
2.2. Supply Chain Risk........................................................................................................ 50
2.3. Cultural Influences on Risk Mitigation ...................................................................... 52
3. Methods ................................................................................................................................ 54
3.1. Constructs ..................................................................................................................... 57
3.2. Validity .......................................................................................................................... 58
3.3. Data analysis ................................................................................................................. 62
3.3.1. Measurement Invariance...................................................................................... 62
3.3.2. Robustness ............................................................................................................. 70

viii

4.

Discussion ............................................................................................................................. 70
4.1. Limitations .................................................................................................................... 73

Chapter 4 - Evaluating the Performance of Risk Mitigation Systems: A Frontier Approach ..... 75
1. Introduction ......................................................................................................................... 75
2. Literature Review ................................................................................................................ 78
2.1. Performance Measurement ......................................................................................... 80
2.1.1. Supply risk mitigation .......................................................................................... 82
2.1.2. Process risk mitigation ......................................................................................... 83
2.1.3. Demand risk mitigation ........................................................................................ 85
3. Methods ................................................................................................................................ 87
3.1. Measures ....................................................................................................................... 88
3.2. Data Analysis ................................................................................................................ 93
3.2.1. Modeling Efficiency of Risk Mitigation in a Network Context ........................ 93
3.2.2. Linear Aggregation Methods ............................................................................... 96
3.3. Results ........................................................................................................................... 96
3.4. Robustness Check....................................................................................................... 105
4. Discussion ........................................................................................................................... 107
4.1. Limitations & Managerial Implications ................................................................... 108
Chapter 5 - Conclusion .............................................................................................................. 111
APPENDIX ................................................................................................................................. 113
REFERENCES ........................................................................................................................... 119

ix

LIST OF TABLES

Table 1. Data Summary ................................................................................................................ 30
Table 2. Weibull survival regression model results for Phase 1 & 2 clinical trials ...................... 33
Table 3. Weibull survival regression model results for Phase 3 clinical trials ............................. 35
Table 4. Summary of Hypotheses ................................................................................................. 36
Table 5. Risk Measurement Summary .......................................................................................... 55
Table 6. Demographic information ............................................................................................... 56
Table 7. Summary Statistics ......................................................................................................... 58
Table 8. Comparison of factor loadings and reliability for risk mitigation measures .................. 60
Table 9. Factor loadings and reliability for risk mitigation measures .......................................... 61
Table 10. Comparison of factor loadings and reliability for risk measures .................................. 62
Table 11 - Measurement invariance: Flexibility ........................................................................... 65
Table 12 - Measurement invariance: Redundancy........................................................................ 65
Table 13. Measurement invariance: Risk...................................................................................... 66
Table 14. GSEM Construct Between Country Comparison ......................................................... 68
Table 15. MANOVA results ......................................................................................................... 69
Table 16. Demographic information ............................................................................................. 90
Table 17. Summary of Measures .................................................................................................. 91
Table 18. Comparison between aggregation methods .................................................................. 91
Table 19. Correlation table (Kendall's tau) ................................................................................... 92
Table 20. OLS regression - Continuous...................................................................................... 106
Table 21. OLS regression - Categorical ...................................................................................... 107

x

LIST OF FIGURES

Figure 1. Drug Development Process ........................................................................................... 11
Figure 2. Moderation effects on Hazard rate for Early Phases (left) and Late Phase (right)........ 34
Figure 3. Cultural Differences ...................................................................................................... 54
Figure 4. Risk Mitigation Comparison ......................................................................................... 68
Figure 5. Perceived Risk Comparison .......................................................................................... 69
Figure 6. Network DEA ................................................................................................................ 94
Figure 7. Flexibility distributions ................................................................................................. 98
Figure 8. Redundancy distributions .............................................................................................. 99
Figure 9. Differences in Inputs (top) and Outputs (bottom) for NDEA vs. BCC efficient firms 102
Figure 10. Raw data comparison between firms ......................................................................... 103

xi

Chapter 1 – Introduction
“Il n'est pas certain que tout soit incertain.”
- Blaise Pascal, Pascal's Pensees
Evolution has ingrained in mankind a risk adverse nature which permeates every dimension
of our lives. In fact, risk is so fundamental to human beings that in striving to better understand
the risk of death and the afterlife, an entire field of mathematics was born. Statistics has
advanced tremendously since its inception and is the one of the most fundamental tools in
science. Before statistics was established, mankind already had begun developing complex risk
mitigation strategies that resembled modern insurance to distribute risk. The importance of
managing risk has only grown with advancements in civilization and statistics. While various
types of insurance were adopted before the Common Era, insurance became more sophisticated
and widespread practice during the enlightenment with the growth of maritime trade. Today,
international trade connects the world and the operations of countless firms, requiring continual
improvement in risk mitigation strategies. Developing a better understanding of the effectiveness
of these strategies is paramount in advancing supply chain and operations management. This
dissertation focuses on this objective by addressing the research question: Why do organizations
adopt various risk mitigation strategies and how effective are they in reducing the impact of
disruptions to operations?
This dissertation addresses this research question by investigating risk mitigation strategies at
different levels of strategic decision making. Supply chain risk is not isolated to any specific
strategic level, be it strategic, tactical, or operational. Additionally, strategies implemented at one
level, may have consequences at others. Thus, to have a more thorough understanding of risk
mitigation strategies, each level must be studied. To achieve this aim, the first essay investigates

1

strategic alliances in the context of new product development to show how alliance may reduce
the rate of failure in new product development projects. At the tactical level, the second essay
builds constructs of redundancy and flexibility based risk mitigation and shows how they differ
across regions. Finally, the last essay develops a performance measurement system for these
strategies to improve their application at the operational level.
Managing supply chain risks fall into four stages: identification, assessment, mitigation, and
response (Chopra and Sodhi, 2004; Manuj and Mentzer, 2008). Identification is the first step
because no action can be taken unless there is a realization that action is necessary. The risks that
were identified then must be assessed. Assessment is the stage in which the severity and
likelihood of risks that affect the supply chain are established (Zsidisin et al., 2004). Once the
potential impact and likelihood of various risks are known, a firm then must decide how to
mitigate them. Mitigation is a critical step in the supply chain risk management process because
it is the stage that firms execute strategies based on information from the previous two steps. The
last stage is response, and occurs after a disruption has happened. All firms must respond to
disruptions, but firms that have performed the previous three stages are better equipped to handle
disruptions.
I argue that mitigation is the most valuable stage in supply chain risk management and is the
focus of this dissertation. While identification and assessment are key components of mitigation,
they do nothing to directly reduce the negative impact of disruptions (Kern et al., 2012). Risk
mitigation is defined as “Supply risk mitigation comprises the actions used to eliminate,
diminish, or counteract supply risks” (Hoffmann et al., 2013). The final step, response, is simply
a necessity after a disruptive event occurs. Responding to a supply chain risk will be much easier
with effective mitigation tools in place. Response can also be a form of a reactive mitigation

2

strategy (Thun et al., 2011). The effectiveness of the response is directly tied to the mitigation
strategies that are in place. Additionally, mitigation strategies cannot be successfully
implemented if supply risks have not been identified and assessed. Therefore, risk mitigation
implies the inclusion of both identification and assessment.
In line with the organization of this dissertation, I study the effectiveness of risk mitigation
strategies to risks that threaten each level of an organization since mitigation strategies must
match the risk they are intended to mitigate. For instance, creating product portfolios presents
risks at a strategic level, thus mitigation strategies such as engaging in alliance partnerships are
the appropriate level of analysis and the topic of my first essay. At the tactical level, having
redundant suppliers mitigates risks associated with natural disasters, political instability, quality
or capacity issues, and any other problems arising at the supplier’s site. However, tactical
decisions such these must account for the increase in governance costs and economies of scale
that result from investing in these strategies. Operationally, ensuring continuity in production
requires reducing the many risk factors stemming from suppliers, quality control, demand
distortion, among others. Operational risk mitigation strategies firms can use are multiple supply
sources or safety stock. Carrying excess inventory and having multiple supplier poses tradeoffs.
For instance, safety stock can buffer disruptions from the supplier, but if the product becomes
obsolete, losses from holding too much inventory can quickly erode the value of mitigating
supply uncertainty. Thus, an effective performance measurement system is necessary to
efficiently allocate resource among the many types of mitigation strategies available and is the
focus of the final essay in this dissertation.
The first essay from my dissertation seeks to understand how firms mitigate the risk of new
product development (NPD) failure. In general, past research has been focused on strategies to

3

improve new product success, but avoidance of project failure is a fundamentally different
perspective that is highly relevant to current practice. An important risk mitigation strategy used
in NPD is to engage in strategic alliances in order to share the cost of project failures. However,
cost is only one of the variables in determining risk and the effectiveness of alliances at
mitigating risk is incomplete without a better understanding of how they influence the likelihood
of failure. In my first essay of the dissertation, I assembled a proprietary dataset of almost 9000
unique projects in the pharmaceutical industry to study how alliance partners reduce the
likelihood of failures in clinical trials. Using survival analysis, I test the effectiveness of
exploration and exploitation partnerships as well as the characteristics of alliance partners at
mitigating project failures across the different phases of clinical trials. The results of this study
not only inform the pharmaceutical industry about the effectiveness of strategies they employ,
but it contributes to the literature on how to minimize the risk of premature project failure.
In the second essay of my dissertation, I explore how culture influences the perception of risk
and the ways firms mitigate them. Based on the foundational work of the psychologists
Kahneman and Tversky, I expect risk perception to partially be a function of culture. For this
study, I apply the organizational theory of slack to differentiate two types of risk mitigation
strategies: flexibility and redundancy. I develop scales for redundancy and apply previously
develop scales of flexibility in the context of risk management. Further, I examine the
differences in scales across cultures by testing the differences between the strategies using a
multinational survey with data drawn from Australia, USA and Japan. I address measurement
issues by testing whether the scales across the different countries are invariant and then test for
these differences by applying a generalized structural equation model.

4

In the third essay, I extend the previous two by exploring the trade-off between the cost and
effectiveness of risk mitigation in supply chains. There is general agreement that applying risk
mitigation strategies is good practice, however, the trade-off between the investment in these
strategies and how much they reduce the impact of disruptions is unclear. I explore the relative
effectiveness of these two strategies in terms of their cost/performance trade-off. In examining
this trade-off, I contribute to the literature by first measuring risk mitigation across the different
echelons of the supply chain: supplier, operations and customer. Beyond enumerating the tradeoffs, mine is the first studies to use a network data envelopment analysis (DEA) methodology in
this context to understand the performance of supply chains in a holistic sense with
interrelationships among network partners.
Across an organization, many practices are used to mitigate different risks. This dissertation
investigates several of them, narrowing the scope of risk management to mitigation practices at
different strategic levels. Many scholars focus on a specific domain or strategic level to narrow
their research topic. Yet, much can be learned by relating strategies across an organization. My
research covers multiple levels of strategic decision making in operations and sourcing
management by exploring the effectiveness of mitigation strategies in new product development
and supply chain management. Managing uncertainty is one of the principle goals of any firm
and has important implications for firm profit and financial sustainability. This dissertation adds
to the body of knowledge of risk management by focusing on the strategies firms use to mitigate
them, and investigating how, why, and when they are effective at reducing risk.

5

Chapter 2 - Mitigating New Product Development Risk in the Pharmaceutical Industry
1. Introduction
Pharmaceutical drug sales have reached record levels, but so has the cost of developing them.
The estimated cost of developing a single drug today is almost $2.6 billion dollars (DiMasi,
2014). The main contributor to the astronomical price tag of drug development is drug failure.
Specifically, failed drugs and invested capital account for more than half of the total cost of
bringing a drug to market (DiMasi, 2014). In the clinical trial process, DiMasi (2014) finds that,
depending on the molecule, the chances of a drug moving from the initial phase, i.e., phase I, to
approval was between 13% to 32%. Failures not only hurt throughput in the drug pipeline of the
organizations, but also lock in financial capital putting organization’s performance at risk
(Schachter and Ramoni, 2007). Thus, it is critical that organizations mitigate the risk of failures
in drug development projects. To mitigate these persistent failures, organizations pursue many
different strategies to reduce the risk and impact of drug failure, dominant among them is
engaging in alliances (Das and Teng, 1998; Rettig, 2000; Khanna, 2012; Gunasekaran et al.,
2015).
Alliances in drug development projects serve multiple purposes. First, alliances facilitate
sharing of development costs across multiple stakeholders. Not only can partners share the cost
of development, they also lower development costs by leveraging their scale and expertise
(Rettig, 2000). Second, as organizations look to reduce failures, alliances bring in external
competencies to bear on projects. These competencies can be in the form of labor and knowledge
(Rettig, 2000). These advantages, in terms of gaining additional resources and engaging external
partners, have been shown to improve innovativeness in the new product development
environment (Ahuja, 2000; Petersen et al., 2005; Sampson, 2007). The knowledge alliance

6

partners bring into projects, may not only be related to innovation, but also related to mitigating
risks. In the context of drug development, against the backdrop of the number of failures that
organizations face, alliances might help the organization reduce the cost of development and the
risks of project failure.
An integral part of success and failure in drug development projects is the examination of
risks that organizations face in this context. In assessing risks, both likelihood and cost of an
event are important elements (Haimes, 1998). In this study, I focus on project risk in the context
of new drug development projects. While success is an often studied metric in the context of
product development relationships (e.g., Rothaermel and Deeds, 2004; Sampson, 2007) failures
are equally important in the context of drug development, given the magnitude of failures that
organizations face is significant in this domain. Depending on the molecule, failure rates can
vary between 68% to 85% (DiMasi et al., 2010). In order to improve this rate either the number
of successes needs to be increased or number of failures decreased. Past research on alliances has
primarily focused on increasing successes and has shown that they are associated with attributes
of the alliance partners in product development projects (e.g., Sampson, 2007), yet their
influence on failures is unexamined. Given that the failure rates are significantly higher than the
success rates, critical information can be garnered from studying the underlying role alliances
play in reducing the risk of failures in the drug development process. Thus, studies exploring the
theoretical mechanisms behind understanding failures can contribute to a more holistic
understanding of the New Product Development (NPD) process. In this study, I examine this
critical aspect by understanding the role of alliances in mitigating project risk in drug
development projects.

7

Focusing on alliances, this study builds on the theoretical framework of explorationexploitation espoused by March (1991). This framework is appropriate in the context of the
pharmaceutical industry because of the progression of the clinical trial process and has been used
in previous studies (e.g., Rothaermel and Deeds, 2004). The beginning of the clinical trial
process is knowledge intensive where activities involve significant exploration. For instance,
steps required at the beginning of the clinical trial process such as population identification, data
collection, and study design among others contribute to project specific knowledge. Furthermore,
subsequent activities such as the scaling up of trials, selecting site locations and coordinating
activities require exploitation of existing knowledge on how trials are conducted. Given that
failure rates are systematically different across the different phases of drug development; the role
of external knowledge from alliance partners may well differ across the different phases of the
development process. Thus, it is important to consider how the impact of external alliances on
failures differ systematically by the phase of the trials. In support of this observation, studies,
such as Arrowsmith (2011), find that success rates for projects in phase II are the lowest and
have fallen from 28% to about 18%. Furthermore, the success rates of trials in phase III is the
highest. The differences in success rates across the phases of clinical trials further suggests that
the impact of alliances may differ depending on the phase of the drug development process.
Accordingly, I examine the impact of alliances across the various phases of clinical trials using
the exploration-exploitation framework. Specifically, I define an exploration alliance as an
alliance with a partner in early stage environments and exploitation alliance as one with an
alliance partner in later stage of the clinical trial environment.
The underlying idea of examining risks within the phases of clinical trials also resonates with
the NPD literature. In examining the individual phases of the NPD process, managers make

8

decisions that can minimize risks within individual project phases. Similarly, in order to better
understand this phenomenon in the current context, I focus on the phases of the clinical trial
process to examine the impact of project based alliances on project failure. This has important
implications for managers in crafting strategies that mitigate these risks based on the specific
phase of development.
Finally, in mitigating project risks using exploration and exploitation alliances, it is important
for firms to be able to appropriately utilize the knowledge and resources of their partners.
Partners in complex NPD projects bring in multiple forms of complementary resources that can
impact the success or failure of the projects. Past studies in the innovation context have
suggested that when partners differ in their knowledge from their focal firm, they are more likely
to yield greater levels of innovation (Vasudeva and Anand 2011). Drawing from this stream of
research, I focus on understanding how resource diversity that partners bring to bear in drug
development projects can impact the risk of failure. This impact may differ depending on the
phases of the clinical trial process (early or late) and the nature of alliances (exploration or
exploitation). Accordingly, I examine the moderating role of partner diversity on the impact of
exploration and exploitation alliances on risks of project failure.
2. Drug Development Process
NPD is a sequence of processes that transforms a market opportunity into a product available for
sale (Krishnan and Ulrich, 2001). Developing new drugs, in many ways, is similar to the NPD
processes for other classes of products. A key differentiation is that, in the pharmaceutical
industry, the process is strictly regulated. Figure 1 shows the process along with the primary
organizations involved in the various steps. The process begins in a laboratory, typically with
high throughput screening to identify potential drug candidates. Once the molecule has been

9

‘discovered,’ it is then tested for biological or chemical activity within a specific disease
mechanism or pathway. After the molecule shows effectiveness for a specific disease in vitro it is
tested in animal models (in vivo). This process can take anywhere between 3.5 years to upwards
of 9 years (DiMasi et al., 2014). Once a drug candidate is successfully filtered from the
thousands of potential molecules, an Investigational New Drug (IND) application is filed with
the Food and Drug Administration (FDA).
Human studies are conducted in association with hospitals and begin with healthy volunteers
in phase I clinical trials. Healthy volunteers are chosen because the purpose of the first phase is
to determine toxicity in humans. Depending on the drug, phase I trials may last anywhere from a
few months to a couple of years and require approximately 20 to 100 subjects. Phase II of
clinical trials begins once toxicity is shown to be acceptably low. Phase II trials require a greater
number of subjects than phase I and the subject must have the disease. On average, there are
between 100 to 300 volunteers recruited for this phase of clinical trials. During this time, the
most effective dosage is determined. Once the dosage has been established, the drug advances to
the phase III.
Phase III is the largest and most expensive step in the development process; it is also the
most critical because it is the final development step before the drug can be approved for
marketing. At least two critical trials are conducted, requiring between 300-3,000 patients
depending on the condition treated and the drug. At this point, any side effects related to the drug
must be discovered along with a continued display of effectiveness in treating the intended
disease. If other drugs are currently available to treat the disease, the drug being developed is
required to show benefits beyond what is currently available. This phase takes approximately 3
years, but can last twice that long for chronic diseases.

10

Figure 1. Drug Development Process

The process as described may seem rather simple, but in reality it can deviate substantially
because it is merely the minimum requirements set by the FDA. For instance, phase I trials can
be designed to give early indications of efficacy rather than just safety. Phase II trials can
occasionally be skipped completely or even serve as a final registration trial in special cases.
Further complications can arise when the same drug is developed for multiple indications. For
each indication, separate phase II and III studies need to be conducted, but the phase 1 trials can
be bypassed. In addition, earlier phase trials may be initiated after later phase trials have begun
(represented Figure 1 by the + sign after the Greek symbol that represents that phase). For
example, a drug may be currently under study in a phase III trial and an organization may initiate
a phase I study to investigate the safety of drug-drug interactions for combination therapies or to
demonstrate a reduction in the variability of response either between or within patients. These

11

deviations from the standard clinical trial process allow for organizations to have the flexibility
necessary to adopt strategies to minimize risks to patients.
The drug development process requires managers to optimize their product development
resources in a very complex environment with a high degree of risk that changes as development
progresses. In terms of failures, 84% of drug failures are attributable to efficacy and toxicity
(Arrowhead and Miller, 2013). The remaining 16% are related to strategic, commercial, and
operational structure, such as subject recruitment, infeasible testing procedures, documentation,
incidents while testing, portfolio complexity, incorrect market assessment, and lack of
experience (Kennedy, 1997; DiMasi, 2001). Theoretically, toxicity and lack of efficacy are
factors inherent to the drug being studied, but the design of the study can have implications for
discovering these two vital parameters. Medical and laboratory research is notoriously complex
and false-negatives or false-positives often occur (Bounansegna et al., 2014). These incorrect
assessments of a drug’s efficacy and toxicity demonstrate how broadly managerial decisions may
impact project failures. This highlights the importance of knowledge and capabilities being
applied to reduce the likelihood of project failure in an industry that is one of the most
challenging to create new products. Alliances may play a critical role in reducing many of these
risks because alliance partners can critically evaluate study design and contribute their expertise
to improving clinical trials. Efficient utilization of knowledge resources may reduce the number
and rate of failures.
3. Literature Review
3.1 Strategic Alliances in NPD
Strategic alliances are common in pharmaceutical NPD. Firms enter strategic alliances for
various reasons, such as differences in resource endowments (Park et al., 2002) or

12

complimentary capabilities (Lorenzoni and Lipparini, 1999). Firms also enter strategic alliances
to share the cost of development and reduce their risk exposure. Engaging in alliances reduces
the NPD risk by reducing the amount each party invests in development. This allows
organizations to invest in products they otherwise would not be able to on their own or to
disperse their investment to multiple projects. It also allows access to an expanded network of
specialized firms, and their skilled personnel. The benefits of this network has been extensively
studied at the firm level from the perspective of positive outcomes, such as innovation (Ahuja,
2000), network strength (Kandemir et al., 2006), and knowledge exchange (Oxley and Sampson,
2004). The gap in alliance literature lies in a lack of research investigating the role alliances play
at minimizing the risk of the project under joint development.
The gap may be a result of the assumption that strategies associated with positive outcomes
are linearly and inversely proportional to negative outcomes, i.e. a success implies not failing
and vice versa. Under closer scrutiny, this assumption may not be valid as strategies directed at
achieving positive outcomes may not directly minimize risk. For example, investing heavily in
advanced laboratory equipment may facilitate discovery of new drugs and as a result improve the
innovativeness of a pharmaceutical organization, but does not necessarily influence the risk of a
research project failing. While a project success is the exact opposite of a failure, it is important
to understand how investments in strategies aimed at avoiding failure or achieving success may
vary together, but are not necessarily required to.
The high cost and likelihood of NPD failure is pressuring organizations to adopt NPD risk
mitigation strategies. To reduce risk of failures, organizations may choose partners based on
similar risk preferences or risk mitigation capabilities. For instance, Lilly is known for its focus
on risk management and went as far as hosting ‘failure parties’ to facilitate learning from failure

13

(Burton, 2004). Additionally, firms may use partner’s knowhow across the different phases of
the clinical trial process to minimize the risk of failure within individual phases of the project.
Despite the focus by pharmaceutical organizations on avoiding failures, existing studies in the
industry primarily focus on methods to improve the number of successes (e.g., Fang, 2011;
Rothaermel and Deeds, 2004; Stuart, 2000). Specifically, research has shown the benefit of
strategic alliances, citing linkages to innovativeness (Kotabe and Scott Swan, 1995) and product
development (Rothaermel and Deeds, 2004). Thus, shedding light on the relationship between
strategic alliance and failure reduction would contribute to a more comprehensive understanding
of the NPD process within this industry.
Focusing on the theoretical basis, past studies primarily use the seminal framework of
exploration and exploitation introduced by March (1991). From the perspective of the focal firm,
exploration and exploitation alliances differ in their learning orientations (Koza and Lewin,
1998). Organizations seek exploration alliances partners in new product development in order to
learn new knowledge from their alliance partners (Rothaermel and Deeds, 2004). In
pharmaceutical NPD, it could be argued that discovering something new occurs prior to the
clinical trial stage. Yet, clinical trials are still fundamentally concerned with discovery,
particularly the early stage trials. The important difference is the distinction between discovering
a potentially new product and updating a previously discovered product during the development
process with new information. Consistent with March (1991), Baum et al. (2000, p.768) define
exploration as “learning gained through processes of concerned variation, planned
experimentation, and play.” This definition introduces uncertainty in outcomes to exploration. In
pre-clinical and clinical trials, this is a key activity. Specifically, early phases (I and II) of the
trial process involve significant learning, and possible search of new ideas that are important for

14

the organization to learn about the product, and are characterized by a high degree of uncertainty
in outcomes.
March (1991, p.85) defines exploitation as, “refinement and extension of existing
competencies, technologies, and paradigms,” which suggests that exploration is relevant for
preclinical trials whereas exploitation is critical during clinical trials. In particular, in later stages
of clinical trial process, scaling the clinical trial process is a key element to successful execution.
The uncertainty in outcomes is significantly reduced at this stage and the value of alliance
partners lies in being able to leverage the knowledge gained in earlier phases. In a similar
context, Rothaermel and Deeds (2004) find that exploratory alliances are associated with a
greater number of products in the pipeline, while exploitation alliances are associated with a
greater number of products on the market. Other scholars have also used the exploration and
exploitation framework in the context of pharmaceutical NPD process (e.g., Hoang and
Rothaermel, 2010; Nielsen and Gudergan, 2012). I also adopt the exploration-exploitation
framework to study alliances as a risk mitigation strategy because of the importance these two
learning orientations may have for accessing and leveraging knowledge relevant to risk
mitigation in different phases of development.
3.1.1 Exploration alliances
Exploratory alliances are primarily intended to facilitate learning from partners (Baum et al.,
2000). Learning is necessary for risk mitigation in new product development and testing for
several reasons. First, learning improves the assessment of risk with updated information which
can improve decision making. In supply chain risk management literature, learning orientation
has been shown to indirectly improve risk mitigation in supply chains (Braunscheidel and
Suresh, 2009). In order for risk mitigation to be effective, risks must be accurately identified

15

(Kern et al., 2012). Alliance Partners (APs) improve access to diverse knowledge and
information resources (Ahuja, 2000), which may improve the methods of product development
and help solve problems with novel solutions. For example, false-positive outcomes are a
common source of failure in clinical trials and may be avoided through better study design and
identification of issues (Buonansegna et al., 2014). Exploratory alliances may result in improved
study design through access to superior information and resultant utilization of information.
As development continues through clinical trials, the impact of exploration alliances will
likely vary. In phases I and II of clinical trials, uncertainty about the outcome of the product is at
the highest level of all of the phases. Several key activities constitute this uncertainty. These
include assessment of drug safety and tolerability, response to dosages, its pharmacokinetic
profile, gaining of knowledge related to the mechanism of action and building an understanding
of side effects of the product (Buonansegna et al., 2014). A deep understanding of the product is
necessary in order for an organization to reduce their chances of failure in these early stages.
Exploration alliances are founded on the desire to create knowledge through learning, and that
knowledge will be specific to the development project. As development progresses to phase III,
the product approaches commercialization and existing knowledge may be better exploited since
lower uncertainty will permit greater generalizability of that existing knowledge toward problem
solving. In contrast to earlier stages, exploration activity may be less valuable in later stages,
since a majority of uncertainty is resolved within the drug development projects. As March
(1991) notes, organizations need to balance exploration and exploitation within their
development efforts. Such a balance is necessitated as management needs to balance both
resources and attention. As an activity, exploration is extremely suited to early stages of
development projects where uncertainty level is the highest. In contrast to early stages, firms that

16

pursue exploration alliances in later stages may increase chances of failure. Specifically,
exploration alliances in later stages are likely incompatible with the product development process
since they may be pursued at the expense of other strategies (exploitation). Furthermore, in later
stages, exploration likely takes resources away from focusing on planning for commercialization
of the products. Thus, I posit a reduction in risk for early phase of clinical trials, but an increase
in risk of failure in later stages of the development process:
HYPOTHESIS 1a: Exploration alliances reduce the probability of product failure in Phases
I & II of clinical trials.
HYPOTHESIS 1b: Exploration alliances increase the probability of product failure in Phase
III of clinical trials.
3.1.2 Exploitation alliances
Just as was the case with exploration, exploitation has a variety of definitions, but they differ to a
greater extent than those for exploration (Gupta et al., 2006). Despite the differences, the concept
of existing knowledge is generally shared within most of the definitions. Benner and Tushman
(2002) define exploitation technology as, “involve improvement in existing components.”
According to Vermeulen and Barkema (2001), exploitation is defined as “ongoing use of a firm’s
knowledge base.” These definitions were chosen because they exemplify the similarity in
conceptualization and represent two alternative perspectives on organizational learning. Both
specify the trait of the exploited resource as existing, but the first definition retains the same
learning element as exploration while the later simply specifies use. How then do exploitation
alliances influence NPD risk?
Past research has traditionally associated the type of alliance with the function it serves
(Lavie and Rosenkopf, 2006). For example, exploitation alliances have been shown to be more

17

influential on product success at later stages of the NPD process (Rothaermel and Deeds, 2004).
When applied for risk mitigation, that functional distinction and associated temporal precedent
appears less defined. Whereas exploration alliances reduce risk through learning, exploitation
alliances rely on utilizing existing knowledge. Exploitation alliances allow for information to be
combined and scrutinized by multiple stakeholders in order to create NPD processes with the
least risk. Exploitation alliances may be sought to assist in the successful production, sale, and
distribution of that product. While these capabilities are targeted toward the late stages of
product development, the formation of and, knowledge leveraged from exploitation alliances
may be implemented at any stage of the development process because it already exists.
Specifically, steps detailed in one of the stage of the development process to mitigate the product
development risks may also find value in mitigating risks in other stages. When forming an
alliance, organization thoroughly analyze the feasibility of the project and may contribute input
beyond their role that may enhance the risk profile of the project. In environments that require
speed in discovery of knowledge, such as pharmaceutical drug development, exploitation of
knowledge is essential for survival (Miller et al., 2006). Therefore, adopting exploitation alliance
can likely benefit from existing knowledge at any stage in development.
HYPOTHESIS 2a: Exploitation alliances reduce the probability of product failure in Phases
I & II of clinical trials.
HYPOTHESIS 2b: Exploitation alliances reduce the probability of product failure in Phase
III of clinical trials.
3.2 Alliance Partner Characteristics
In order to derive the maximum intended benefits from alliances, the specific traits of each
alliance partner in relation to one’s own are of critical importance. The term diversity is often

18

used in the literature when studying the impact of alliance partner characteristics and diversity in
resources, knowledge, and capabilities has been shown to improve alliance innovation
performance (Dell'Era and Verganti, 2010; Faems et al., 2005; Phelps, 2010; Srivastava and
Gnyawali, 2011). Despite key differences in the timing and objectives of exploration and
exploitation alliances, there has not been a thorough investigation to when and which type of
diversity is important for each of them. For example, Lane and Lubatkin (1998) note that it is
easier for partners to find and exchange knowledge when they have similarity of knowledge
profiles. Such similarities can help partners to better exchange knowledge and ideas in each
phase of the project resulting in their ability to control the risk outcomes from the project. In
contrast, dissimilarities in knowledge profiles can allow for greater learning abilities by pooling
resources of the diverse partner base. These pooled knowledge resources can result in greater
likelihood of problem identification, and consequent risk mitigation efforts by implementing
novel solutions. Postrel (2002), for example, notes that the overall payoff from external
knowledge depends on the relative learning costs and the marginal pay-off that organizations
expect to gain. Diversity of knowledge between focal firms and alliance partners is a critical
component of how organizations can expect to leverage external knowledge. I propose that
diversity in partner knowledge characteristics is likely to have an impact on how organization is
able to leverage exploration and exploitation alliances to reduce product risk.
3.2.1 Exploration Alliance Diversity
Exploration alliances as previously defined are sought with the purpose of generating new
knowledge through learning. As such, the diversity in alliance partners should reflect traits that
are specific to the goal. For exploration alliances, knowledge is created through the codevelopment process. Therefore, diverse NPD experiences of partners are likely an important

19

indicator of knowledge that can be used to explore new knowledge domains. For example, Wu et
al. (2015) found that differences in technological profiles of partners, is necessary to improve the
design and development of new products. This reflects the primary objective of exploration
alliances, generating new knowledge in the form of intellectual property. Past research has
examined the differences in partner patent profiles on innovation performance (Vasudeva and
Anand, 2011). It is likely that these differences in partner knowledge profiles reduce risks in
projects as the differing knowledge bases among the alliance partners allow them to combine
their knowledge to search for new ways to minimize failures in projects. Vasudeva and Anand
(2011) propose that diversity consists of latitudinal and longitudinal components, referring to
diverse and distant partner knowledge, respectfully. I adopt this logic and capture AP’s diversity
with two measures: NPD complementarity (latitudinal) and partner innovativeness (longitudinal).
I propose that when alliance partner innovativeness is high, it is likely that exploration
alliances will be more effective in reducing project risks. While past research has not directly
proposed a relationship between partner innovativeness and NPD risk reduction, many of the
arguments proposed can be extended to the domain of risk. For instance, Baum et al. (2000)
showed that innovativeness of alliance partners improves the success of biotech startups. The
authors suggested that alliance partners “…provide efficient access to diverse information and
capabilities…[and] provide more opportunity for learning…” (Baum et al., 2000, p. 268).
Innovative organizations have experience managing NPD risk, and partnering with them may
improve the collection and analysis of information necessary to minimize NPD risk. Specifically,
exploratory alliances are characterized by joint learning, discovery and co-development efforts.
In pursuing exploratory activities, greater diversity of knowledge in partners is likely to help the
firm discover novel information that can be utilized to reduce unseen risks that organizations

20

may face through the process. This can be attributed to greater degree of pooled domain
knowledge between the alliance partners. This pooled knowledge can facilitate better
identification of risks in the development process. In addition, it is likely that superior
identification of risks accompanies search and implementation of superior solutions to mitigate
the risks as well. I quantify knowledge diversity with a citation weighted patent diversity
measure and refer to it as Partner Innovativeness (P-Inn.). Thus, I hypothesize:
HYPOTHESIS 3a: Partner Innovativeness (P-Inn.) negatively moderates the relationship
between exploration alliances and product failure such that when alliance partners have high PInn., exploration alliances have a lower probability of product failure than at low P-Inn.
Beyond partner innovativeness, I expect that the relative experience partners have in
developing products in individual phases will also influence risk. Diversity of knowledge
domains reflects differences in fundamental understanding across the knowledge domains of
activities and processes. Katila and Ahuja (2002) reflect this in their study examining innovation
performance after acquisitions. The authors argued that innovation performance of an
organization is contingent on the knowledge characteristics between the acquiring and acquired
firm. Likewise, differences in process knowledge is likely to promote greater awareness of risk
mitigation strategies, more so given the criticality of processes in complex NPD projects. This is
likely true in processes such as clinical trials. Consider Dilts et al. (2010) who did an in-depth
analysis of phase III cancer trials. These trials involve an in-depth understanding of the process
of conducting the trial. Specifically, the process itself had several hundred steps requiring several
transactions that span regulatory authorities, patients, and other partners. Similarly, Adams and
Branter (2006) note that process related uncertainties in the clinical trial process are a key
contributor to the development cost of drugs. Several critical functions span this process

21

including protocol development, feasibility assessment, study contract processes for each stage
among others require deep execution knowledge (see Dilts et al., 2013 for an overview of the
steps).
When partners possess diverse process knowledge the combination of knowledge related to
NPD can reduce the risk of failure as they can better recognize the process risks that may
underpin the different phases. Organizations that have most of their products in early stages of
development will have a substantial amount of knowledge relating to early development projects.
The same argument can be made for organizations with the majority of their products in market.
Difference is experience with NPD process between organizations represents complementarity of
experience between the partners. For exploration alliances in which firms seek to discover
something new, I expect NPD complementarity to improve the knowledge generation required to
uncover and avoid NPD risks. Specifically, the diversity of collective process experience among
partners is more likely to help in resolving uncertainty involved in exploratory activities. This
can facilitate resolution of process related problems more quickly and efficaciously.
Accordingly, I hypothesize:
HYPOTHESIS 3b: NPD complementarity negatively moderates the relationship between
exploration alliances and product failure such that when alliance partners have high NPD
complementarity, exploration alliances have a lower probability of product failure than at low
NPD complementarity.
3.2.2 Exploitation Alliance Diversity
Exploitation alliances are driven by the opportunity to leverage alliance partner strengths in order
to improve the marketing of a product. Therefore, exploitation alliance will likely be influenced
by organizational level variables that represent existing knowledge domains that may be

22

exploited. Such variables that have been argued to moderate how alliances impact NPD
performance outcomes are power dependence (Álvarez Gil and Gonzålez de la Fe, 1999) and
firm size difference (Kalaignanam et al., 2007). To capture the two components of diversity for
exploitation alliances, I utilize partner size (longitudinal) and technological diversity
(latitudinal). Large pharmaceutical organizations have extensive experience with downstream
development activities (Rothaermel and Hess, 2010). They possess substantial capabilities to
exploit opportunities and leverage their experience for the production and distribution process
(Pisano, 1996). Accordingly, it is likely that large organizations are better able to improvise
downstream activities in the value chain and make adjustments to reduce project risks.
Therefore, size would influence the degree that exploitation alliances reduce project risk.
Therefore, I hypothesize:
HYPOTHESIS 4a: Size negatively moderates the relationship between exploitation
alliances and product failure such that when alliance partners are large, exploitation alliances
have a lower probability of product failure than when alliance partners are small.
From an organizational perspective, NPD knowledge domains can be observed based on the
lines of business they operate in. Organizations operating in a particular business segment are
more likely to possess significant knowledge of that industry. Haeussler et al. (2012) argue that
internal technological capabilities impact the upstream and downstream horizontal alliances.
Exploitation alliances in particular help the organization identify downstream risks and examine
countermeasures for the identified risks to reduce failures in the trial process. Further, as argued
before, it is likely that greater diversity of knowledge helps identify risks and consequent
countermeasures. As organizations broaden the scope of their knowledge, they are more likely to
find unseen risks. In order for knowledge assets to be leveraged however, they cannot be

23

completely redundant (Sampson, 2007). Thus, when alliances have greater degree of
technological diversity, exploitation alliances are more likely to help in identifying risks early
since the greater diversity of knowledge domains embedded within the alliance facilitates
superior identification of risks. In line with this argument, Koza and Lewin (1998), demonstrate
that exploitation alliances are effective when competencies across organizations complement
each other (Koza and Lewin, 1998). Thus, I hypothesize the following:
HYPOTHESIS 4b: Technological diversity positively moderates the relationship between
exploitation alliances and product failure such that when alliance partners have high
Technological diversity, exploitation alliances have a lower probability of product failure than
at low Technological diversity.
4. Research Design
I utilized a mixed method approach, combining qualitative interviews with secondary data to use
for statistical analysis. I chose the pharmaceutical industry from 1992-2015 as the setting
because of the highly regulated NPD process and the risk involved in developing new drugs. The
legal requirements of pharmaceutical drug development provide control for the statistical
analysis since I am trying to uncover how alliances influence project failure at various stages in
development. Managers involved in various aspects of pharmaceutical NPD were recruited for
involvement in the study. Interviews were conducted over the phone and lasted between 30
minutes to an hour. The interviews were semi-structured and were used exclusively to ensure
validity of the quantitative study. The participants were asked how they view NPD risk
mitigation, how frequently they used alliances as mitigation strategies and how the
characteristics of their alliance partners influenced project outcomes. Each interview was

24

followed-up after the quantitative analysis was initiated to ensure that the analysis captured risk
mitigation without omission of important items.
In order to test the hypotheses, both project level and firm level data was obtained from
Medtrack and the US patent office (USPTO). Medtrack is a database that compiles information
on pharmaceutical drug development progress, as well as patent, financial, or other related data.
Both public and private firms are covered by the database through compiling information from
public announcements and other proprietary sources. As a result, over 130,000 products and
30,000 companies are listed in the database. I restrict the sample to products that are currently in
development or have failed in phases I-III of clinical trials. Limiting the sample to the three
phases of clinical trials minimized sample selection bias since all drugs must submit a New Drug
Application (NDA) with the FDA before commencing clinical trials. The final sample included
in empirical analyses consists of 8,740 projects in 10 different therapeutic areas over the three
phases of clinical development. Therapeutic areas with less than 1% of the total sample were
dropped from the analysis. The quality of the final data was checked against previously reported
values of the key constructs, alliances and failures. Approximately 70% of the drugs in the
sample were self-originated (those being actively developed by the organization who discovered
the drug), matching the reported percentage of alliances in a study by DiMasi et al. (2010).
4.1. Measures
4.1.1. Product Failure
I examine the relative effectiveness of various risk mitigation strategies in the three phases of
clinical trials. In order to test the hypotheses, I utilize survival analysis since the time it takes a
project to fail is the dependent variable. Time to failure is quantified as the total time in days
from the beginning of a specific phase of clinical trials and the date it is discontinued. Given the

25

complexities of clinical trials, I use the earliest date of the first trial in each phase as the initial
time. The time of failure is the date that the product has been discontinued, suspended, or
withdrawn. Medtrack also classifies products as “no active development” if there has not been
any reported development of a drug after two years in order to capture unannounced
discontinuations. These instances are given a time to failure as two years from the last report,
plus the time from that report to initiation.
4.1.2. Risk Mitigation
I explore alliances as risk mitigation strategies in NPD. Following past studies that investigated
the relationship between alliances and innovation, I adopt March’s (1991) organizational
learning classification to differentiate exploratory and exploitation alliances. Each project in the
database contains the names of each company involved in development along with their role.
Following Rothaermel and Deeds (2004), a binary variable was created for projects that involved
companies involved in co-development as an exploratory alliance. A binary variable was created
to represent exploitation alliances for projects that involved companies in sales, marketing, or
production.
4.1.3. Alliance Diversity
The effectiveness of risk mitigation through collaboration likely depends on the characteristics of
both the focal firm and those of the alliance partner. In order to capture this, I quantify the
relative differences in technological diversity, size, innovativeness, and NPD complementarity.
Technological diversity: Technological diversity (Tech. Div.) is defined as the number of
subfields the APs operate in and is quantified as a count variable (Rothaermel and Deeds, 2004;
Shan et al., 1994). The number of technological subfields that are shared by the focal firm and
those of their alliance partners are subtracted from this measure so that a large overlap of

26

subfields is not misrepresented as high Tech. Div. For example, if an organization operates in
three subfields: Generics, Pharmaceuticals, Stem Cell Therapy and its AP operates in
Biotechnology, Generics, Pharmaceuticals, Tech. Div. would equal 1 since the AP operates in 3
subfields, but two of them are shared.
Partner Size: Partner Size (P-Size) is recorded for each AP as the total number of products in
their pipeline. Partner size is quantified as the average of size for all APs.
NPD complementarity: The measure of NPD complementarity (NPD Comp.) captures the
average degree alliance partners specialized in different phases of NPD than the focal firm. It is
quantified as a percentage difference of products in four stages of development: pre-clinical
research, clinical trials, marketed, and discontinued.
1
𝑗

∑𝑗𝑖=1 ∑𝑛𝑖=1 |

𝑥𝑛
𝑋

−

𝑥𝑗𝑛
𝑋𝑗

|

where xn is the number of products in development stage n and X is the total number of unique
products for the focal organization. Subscript j represents the alliance partners. Absolute values
of the differences were summed because the direction of difference is not theoretically relevant
and also to ensure that positive and negative differences would not cancel each other out. To
illustrate, consider a sponsor company that has 2 APs (P1 and P2) and 10 drugs in development.
The sponsor has 1 drug in pre-clinical research, 5 in clinical trials, 2 marketed, and 2
discontinued. P1 has 5 drugs, all in pre-clinical research. P2 has no drug in pre-clinical research,
23 in clinical trials, 12 marketed, and 10 discontinued for a total of 45 drugs in various stages of
development. The knowledge that the sponsor can leverage from its two APs depends on their
expertise in drug development which is different than its own. The measure is calculated as the
average percentage difference between the stages of development. P1 focuses on pre-clinical
1

5

9

research, therefore its technological diversity would be:|10 − 5| + 10. P2 is more focused on
27

1

5

23

2

12

clinical trials with several products in the market and discontinued: 10 + |10 − 45| + |10 − 45| +
2

10

|10 − 45|. The average of these two values is the final measure of NPD complementarity.
Partner Innovativeness: Partner Innovativeness (AP Innovativeness) is captured for each
organization as a five year citation weighted measure from the date of project initiation (see
below for additional details). An average of this measure is taken for all of the alliance partners
involved in co-development.
4.1.4. Controls
I include several controls that have been shown to influence pharmaceutical drug failures.
Size: Since the sample consists of both public and private firms, common size measures such as
employees or assets were not available for each observation. Therefore, I use the number of
products a firm has in development and in the market as the measure of firm size since it is also
an asset that varies with the size of the organization. Using a subsample of public firms from the
dataset, I verified this assertion by calculating the correlation between the measure of size and
the number of employees (𝑟𝑆𝑖𝑧𝑒∙𝐸𝑚𝑝𝑙𝑜𝑦𝑒𝑒𝑠 = 0.835, p<.001, n=7242).
Indications: Many drugs are developed for more than one disease. Clinical trial data from one
project may be used to minimize uncertainty of other projects. Past research has shown that the
more indications a drug is developed to treat reduces the risk of product failure (DiMasi and
Faden, 2011). This measure is quantified as a count variable, calculated as the sum of all drugs
being developed or have completed development by the sponsor organization that has the same
active compound as the drug under analysis.
Number of Companies involved: Since the alliance variables are binary, I also account for the
number of firms involved in the process from the beginning of development. I do so because as
the number of companies involved in development increases, the potential access to knowledge

28

and resources widens while the difficulty in project coordination may also increase. This variable
(Co.) is quantified as a count variable to account for these potential effects on failure rate.
Licensing: Beyond alliances, purchasing of intellectual property can also be accomplished via
in-licensing. To account for the difference between self-originated compounds and the products
being developed jointly, I control for licensing contracts by including a binary indicator equal to
1 if the compound was in-licensed and 0 otherwise.
Innovativeness: More innovative organizations may fail at NPD less because they are simply
better at NPD than other organizations. I adopt the measure of innovativeness used by Dutta and
Weiss (1997), which is an industry adjusted five-year trailing citation weighted measure. For
every project, patent citations were summed five years prior to the first date of project initiation
and divided by the average five year patent citation count for that organization’s industry based
on MedTrack’s classification of industry. The USPTO database was used to develop this
measure to limit differences in patent systems (Nagaoka et al., 2010) and because the majority
the companies in the sample are located in the United States.

29

Table 1. Data Summary
Variable
Development
time
Phase 1
Phase 2
Phase 3

Phase 1
Phase 2
Phase 3

Description
N
Time in days from the beginning of clinical trial
to discontinuation of product development
11
39
28
73
55
7
Time in days from the beginning of clinical
trial to censored date
11
35
20
71
94
7

Sponsor
Characteristics
Clinical

Number of drugs in clinical trials

Marketed

Number of marketed drugs

Discontinued
Drugs

Number of drugs that have been
discontinued
Total number of drugs developed

Innovativeness

5 year citation weighted patent count

Project
Characteristics
Indications
Exploration
Alliances
Exploitation
Alliances
Licensing
agreement
# Companies
involved
Technological
Diversity
NPD
Complementar
ity

87
20
87
20
87
20
87
20
87
19

Median

Mean

σ

Min

Max

851

1053.
82
1270.
67
1417.
68

678.85

21

4749

791.96

10

5536

961.08

11

5791

906.3
3
1091.
27
1030.
70

816.69

3

5148

953.29

2

7867

898.80

1

5550

13

70.49

107.93

0

382

3

109.8
8
83.62

194.91

0

929

140.71

0

581

290.6
2
4.31

452.20

1

1719

6.97

0

83

1095
1227

619
814
750

10
48
0.77

Number of diseases that the same
compound is being developed to treat
Indicator of co-development in product
development
Indicator of sales, marketing or
manufacturing partnerships in product
development
Indicator that rights to develop a technology
were obtained through a licensing
agreement
Number of companies involved in
development

87
22
87
22
87
22

2

3.75

4.33

1

38

0

0.23

0.42

0

1

0

0.04

0.20

0

1

87
22

0

0.09

0.29

0

1

87
22

1

1.60

0.86

1

9

Total number of sub-fields partner firms
operate in
Degree of difference in NPD expertise
between partner firms and sponsor firm

36
22
36
22

3

4.94

4.53

0

33

0.89

0.90

0.50

0

2.65

30

4.2. Methods
I estimate the effectiveness of the various NPD risk mitigation strategies previously described to
advance pharmaceutical products through clinical trials using a shared frailty Weibull survival
regression analysis (Cleves et al., 2016). Survival analysis is necessary to obtain consistent and
efficient estimates of the variables because many of the projects in the sample have not made it
to the next phase of research or have been discontinued and are therefore censored (Hanagal,
2006). I test the hypotheses in each of the three phases of clinical trials. Each is estimated
separately to determine the effectiveness of the mitigation strategies throughout the development
process.
The duration of time it takes for each stage is the dependent variable and modeled as a log
linear function of the covariates according to equation (1).
log(tj) = xjβ + zj

(1)

Where xj is a vector of covariates and β is their corresponding vector of regression coefficients.
The error term, zj, is modeled with a Weibull probability distribution. I utilize the Weibull model
because it allows for hazard rates to increase or decrease exponentially with time rather than the
more common Cox regression which assumes that the hazard function is constant. This
assumption is likely not valid in the case of pharmaceutical drugs due to the change in
technology and policy over time. The Weibull hazard function takes the form:
ℎ(𝑡) = 𝑝𝜆𝑗 𝑡𝑗𝑝−1

(2)

𝜆𝑗 = e−𝑝𝐱𝒋 𝛃

(3)

where p is the estimated shape parameter; the hazard rate, represented by equation (2) is
increasing if p is greater than 1 and decreasing if p is less than one. The scale parameter
displayed in equation (3), Îťj, allows for estimation of the impact of covariates on the hazard rate.

31

In addition, I account for potential correlation in error terms using the gamma shared frailty
model. The frailty was defined to be shared by year to account for differences in policy,
technology and economic health over time.
5. Results
I present the results in Tables 2 and 3. The results are presented in a hierarchical order, with the
controls in model 1, the addition of main effects in model 2, and the interaction effects in model
3. Early and late phases of clinical trials were estimated separately to determine the effectiveness
of alliances at the two stages in the development process. Phase I and II were combined because
they were hypothesized in the same direction and many products are allowed to bypass phase I if
the same compound has passed the safety requirements for a different drug. A dummy variable
accounting for the differences in failure rate of drugs in the first phase was included in the early
regression analyses. Combining them also aligns with the theoretical motivation of exploration
versus exploitation and has precedence in literature in the pharmaceutical industry (Urbig et al.,
2013). Across the phases, each of the control variables exhibits a similar pattern of effect sizes
and significance levels. Innovativeness was not significantly related to failure likelihood. An
increase in number of diseases that a drug is being developed for reduces the risk of drug failure,
confirming results from past research (DiMasi, 2013). Size as measured by the total number of
products in the company’s pipeline is positive and significant, indicating an increased likelihood
of failure for larger companies. An increase in the number of companies involved in
development reduces failure likelihood, supporting the general notion that a greater knowledge
base reduces risk. In each model, the Weibull shape parameter is greater than 1, indicating that
the hazard rate is increasing as the project remains in development at a decreasing rate.

32

Table 2. Weibull survival regression model results for Phase 1 & 2 clinical trials
Model 1
Model 2
Model 3
Early
β
se
β
se
β
se
Intercept
-12.970
Phase 1
0.171
Indications
-0.014
Size
0.018
Innovativeness
0.025
Co.
-0.151
License
0.065
Exploration
Exploitation
Tech. Div.
P-Size
NPD Comp.
AP. Innovativeness
Exploitation Interactions
Tech. Div.
P-Size
Exploration Interactions
NPD Comp.
AP. Innovativeness
Weibull shape parameter
1.761
Frailty Parameter
0.566
N
7239
AIC
11927.3
Log lik.
-5929.9

**
***
***
***
*
**
*

(0.200) -13.111
(0.037)
0.169
(0.004) -0.014
(0.003)
0.018
(0.022)
0.022
(0.019) -0.107
(0.068)
0.106
0.079
-0.173
-0.012
0.007
0.002
-0.094

**
***
***
***
*
**
*

†
†

*

(0.201)
(0.037)
(0.004)
(0.003)
(0.022)
(0.032)
(0.074)
(0.057)
(0.099)
(0.007)
(0.008)
(0.021)
(0.042)

-13.097
0.167
-0.014
0.018
0.022
-0.138
0.082
0.270
-0.362
-0.009
0.001
0.030
-0.031

**
***
***
***
*
**
*
**
**

0.026
0.056

**
*

(0.013)

1.762
**
0.566
*
7239
11924.1
-5922.7

-0.205
-0.047
1.760
(0.013)
0.567
7239
11920.8
-5917.2

(0.201)
(0.037)
(0.004)
(0.003)
(0.022)
(0.035)
(0.075)
(0.092)
(0.135)
(0.007)
(0.008)
(0.036)
(0.051)
(0.027)
(0.038)

*

**
*

(0.092)
(0.045)
(0.013)

Note: Standard errors in parenthesis. Fixed effects were included as controls in each model
for every therapeutic area treated by the drug in development, but are excluded from the table
for clarity.
†
p<0.10,*p<0.05, **p<0.01, ***p<0.001

33

Figure 2. Moderation effects on Hazard rate for Early Phases (left) and Late Phase (right)

Turning to the effect of the covariates of interest, early phase results are presented in Table 2.
Model 2 displays the main effects of the different types of alliances. The coefficient of
exploration alliance is not significant, rejecting the first hypothesis (1a). Supporting hypotheses
2a, exploitation alliances are moderately significantly associated with a 16.2% (β=-0.177,
p<0.10) reduction in the failure likelihood. When adding the interaction terms, the significance
of exploitation alliances increases along with the effect size (β =-0.376, p<0.01), while the main
effect of exploration alliances become significant (β =.278, p<0.01). However, the direction of
this significant main effect is counter to hypothesis 1a. In model 3, there is only one significant
interaction between exploration alliances and NPD complementarity (β =-0.209, p<0.05).
Comparing two exploratory alliances while holding all other variables constant, an increase of
one unit in NPD complementarity yields an 18.9% reduction in failure likelihood. I find here that
the impact of exploratory alliances on product development varies depending on the differential
expertise of the partners. Specifically, exploratory alliances by themselves result in increased
product failures, but when partner expertise differential is high, exploratory alliances seem to
reduce product failures. While these results display an inverse relationship than the main effect

34

posited in hypotheses 1a, it supports the moderating hypothesis 3b and does not support
hypotheses 3a.
Table 3. Weibull survival regression model results for Phase 3 clinical trials
Model 1
Model 2
Model 3
Late
β
-12.280
-0.026
0.007
-0.075
-0.192
0.530

Intercept
Indications
Size
Innovativeness
Co.
License
Exploration
Exploitation
Tech. Div.
P-Size
NPD Comp.
AP Innovativeness
Exploitation Interactions
Tech. Div.
P-Size
Exploration Interactions
NPD Comp.
AP Innovativeness
Weibull shape parameter
1.613
Frailty Parameter
0.478
N
1507
AIC
2040.1
Log lik.
-997.0

**
**

**
*
**

se
β
(0.491) -12.400
(0.011) -0.022
(0.009)
0.013
(0.064) -0.060
(0.046) -0.170
(0.181)
0.724
0.284
-0.649
0.011
0.013
-0.030
-0.294

**
**

*
**
*
**
*

**

se
β
(0.494) -12.380
(0.010) -0.022
(0.009)
0.012
(0.068) -0.052
(0.079) -0.228
(0.218)
0.694
(0.173)
0.472
(0.177) -0.378
(0.019)
0.033
(0.024)
0.004
(0.047)
0.026
(0.114) -0.249
-0.100
0.009

**
*

(0.035)

1.634
0.491
1507
2026.0
-984.0

**
*

-0.195
-0.100
1.639
(0.035)
0.494
1507
2028.0
-981.0

**
**

**
**

†

†

*

se
(0.497)
(0.011)
(0.009)
(0.070)
(0.086)
(0.227)
(0.249)
(0.263)
(0.021)
(0.027)
(0.078)
(0.147)
(0.051)
(0.057)
(0.242)
(0.098)

**
*

(0.035)

Note: Standard errors in parenthesis. Fixed effects were included as controls in each model
for every therapeutic area treated by the drug in development, but are excluded from the table
for clarity. †p<0.10,*p<0.05, **p<0.01, ***p<0.001
In the final phase of clinical trials shown in Table 3, the main effects of exploitation alliances
are significant and negative, supporting hypothesis 2b (β =-0.649, p<0.001). When adding the
moderating alliance variables in model 3, the main effect of exploitation becomes insignificant,
while the main effect of exploration alliances becomes moderately significant and positive (β
=0.472, p<0.10). The moderating effect of technological diversity is significant and negative (β
=-0.100, p<0.05). The moderating effect when two exploitation alliances are compared while

35

holding all other variables constant yields a 9.5% reduction in failure rate with an increase of one
unit of technological diversity. This result provides support for hypothesis 4a.
Table 4. Summary of Hypotheses
Hypotheses
1a
1b
2a
2b
3a
3b
4a
4b

Variable
Exploration Alliances
Exploitation Alliances
Partner Innovativeness
NPD complementarity
Partner Size
Technological diversity

Phase
I-II
III
I-II
III
I,II,III
I,II,III
I,II,III
I,II,III

p-value (<)
0.01
0.10
0.01
n.s.
n.s.
0.05
n.s.
0.05

Direction Supported
+
N
+
N
Y
N
N
Y
+
N
Y

5.1. Robustness
While a Weibull distribution with shared frailty was used to estimate the model, there are other
models and distributions that could have been applied. First, I assessed the impact of different
distributions on the results. The full model (model 3 in each phase) was estimated using the
exponential, gamma, log-normal, Gompertz, and loglogistic distributions. The estimates from
each model were consistent regardless of distribution used. AIC values from each model were
also compared to ensure the best distribution was selected. The AIC value for the Weibull was
slightly higher than the log-normal, but not enough to override the theoretical justification for
utilizing Weibull distribution.
Cox regression was also performed and showed remarkably similar results. One of the
primary assumptions of Cox regression is that the covariates have proportional hazards. This
assumption was tested by interacting each covariate with the survival time and including it in the
original regression (Cleves, 2008). Each interaction was insignificant, supporting the assumption
of proportional hazards. Outliers were detected by obtaining the Nelson-Aalen cumulative hazard
rate for each regression and plotting it against the residuals from the model. Dropping outliers
and rerunning the model did not significantly change the results for each model. To assess the
36

accuracy of functional form, martingale residuals were visually inspected. No significant linear
trend or asymmetry was observed.
5.1.1. Endogeneity
While it is difficult to know with certainty if a project is going to succeed, APs will pursue
collaboration on projects they expect to succeed. Since time to failure is the dependent variable
in the model, this may lead to a sample selection issue that can result in overestimation
(Heckman, 1979). Although, I do not expect this to be a serious problem because all projects are
pursed with the expectation they may succeed regardless if alliance partners are involved or not.
To minimize the concern of selection bias, I use a two-step estimation procedure to ensure that
endogeneity is not biasing the results. APs may decide to become involved in a project because
of the attributes of the company they are partnering with and the characteristics of the drug in
development. Thus, an indicator of alliance formation is regressed in the first stage on the
sponsor company’s innovativeness and its size, along with the project characteristics: total
number of companies involved in the project, the therapeutic area, and the molecule type. The
inverse mills ratio was estimated from the first stage and included as a control in the original
model specification. The mills ratio is insignificant and the results remain consistent. Summary
of robustness results are presented in the appendix.
6. Conclusion
Simply engaging in exploration alliances in clinical trials is not enough to reduce risk, and unless
alliance partners are chosen carefully based on the capabilities they possess, may increase project
failure. The two early development stages, phase I & II, require partners with unique expertise in
developing products to reduce risk. A greater differential in NPD complementarity reduces the
rate of failure, but at low levels there is an increase in risk. This result suggests that

37

organizational learning from exploratory alliances is accomplished more effectively with alliance
partners who have a different perspective on the development process. This interpretation is
important because it suggests that the impact alliances can have is much stronger than suggested
in past literature. Not only can they improve successes, they also reduce failures which further
strengthens the motivation to engage in strategic alliances. It also indicates that the most fertile
combination of knowledge is from diverse and intimate knowledge of processes. As drug
development progresses to phase III, exploration alliances do not reduce the risk of project
failure, even with alliance partners that have high NPD complementarity.
In the final phase, technological diversity of exploitation alliances becomes important in
reducing the risk of project failure. This further supports the importance that knowledge diversity
plays in cultivating risk mitigation while demonstrating how the type of diversity that should be
sought is dependent on the stage of development. Differences in knowledge domains gained
through expertise in different industry subfields can be combined to minimize risk. Without
considering the attributes of exploitation partners, exploitation alliances reduce NPD risk in and
of themselves. When technological diversity of the exploitation alliance partners is included,
greater diversity is shown to be the primary factor driving the risk reducing impact of the
alliances.
Exploitation alliances not only reduce risk in the last phase of NPD, but also in earlier phases
as well. This result highlights the importance of understanding how alliance partners sought for
exploitation may have a significant effect on the development of a project rather than just on the
final characteristics of the product. This may be a result of the reality of the drug development
process. When the entire product development stage is considered, clinical trials consist of the
latter half of development. Thus, even in phase I and II, exploitation can benefit development

38

processes. Additionally, alliance formation is a process that not only takes time, but also must be
initiated before the product is ready to be exploited. The experiential knowledge of exploitation
partners may be leveraged during the formation process to reduce the risk of the product failing
before it reaches the point that exploitation APs take the lead in marketing or producing the
product.
Despite the extensive application and research on the impact APs have on innovation
outcomes, their role in risk management has been almost ignored. Considering how widely
alliances are applied to mitigate project level risks, surprisingly little is known about their
effectiveness in reducing risk beyond simply sharing cost. The results presented provide several
interesting and useful insights that take one step toward a more comprehensive understanding of
the role of alliance in risk management. When seeking alliance partners, managers should
consider alliance partners for both their capabilities as innovators and as joint risk managers. The
risk reducing effect of alliances was also shown to vary by the type of alliance and the attributes
of alliance partners in relation to one’s one throughout the drug development process.
By understanding the role that alliances have in risk management, the valuation and selection
of APs may be improved. I showed that the characteristics of APs in relation to the sponsor
company are a key factor in determining if and to what extend APs reduce risk. Considering the
astronomical cost of drug development, even small improvements in failure rate can have large
consequences. This study provides an alternative perspective from the most common risk
management tools used in NPD which are borrowed from financial analysis (Doctor et al., 2001)
and include net present value (NPV), decision trees, discounted cash flow, and Monte Carlo
simulation (Kleczyk, 2008). All of these methods are used to measure risk, not mitigate it. Other
scholars have also stressed the importance of risk mitigation for NPD in biotechnology

39

(Vanderbyl and Kobelak, 2008), but none have quantified the impact mitigation strategies have.
This study contributes to the project risk management literature by motivating and quantifying
the impact alliances have to mitigate risk.
It is imperative to develop a better understanding of project risk and actions that can be taken
to reduce them. Pharmaceutical companies must mitigate the risk of project failure which
requires an understanding of the effectiveness of the strategies they employ. The complexity of
this decision making process coupled with the fact that only 1% of compounds make it through
the NPD process, results in imprecise methods for determining which path to take (Gino and
Pisano, 2008). Many trends in pharmaceutical strategy can be seen from a risk management
perspective, including alliance formation. My findings suggest that there is not a pure temporal
separation between the influence of exploration and exploitation alliances as risk mitigation
strategies. Product development is a continuous process and mitigating risk requires preemptive
action. Both types of alliance allow organizations to leverage the experiential knowledge of AP.
However, alliance partner attributes specific to the type of alliance sought need to be considered
when forming alliance.

40

Chapter 3 - The Impact of Culture on Supply Chain Risk Perception and Mitigation
1. Introduction
Making investment decisions on mitigating supply chain risk is a complicated process. Supply
chain managers are faced with many sources of risk and have several ways to mitigate them. For
instance, supplier quality control, equipment malfunction, and demand uncertainty each threaten
the continuity of a supply chain, but require different types of mitigation strategies to combat
them. Additionally, these risks are interdependent and firms must rely on supply chain partners
whom often cross international borders. Effectively mitigating supply chain risk requires
coordination (Kleindorfer and Saad, 2005), but past research has shown that coordination of
global operations can be challenging (Prater and Ghosh, 2006), especially since cultural and
geographical differences may impact the type of operational routines used by firms (Kull and
Wacker, 2010). This is particularly relevant in supply chain risk management because of the
global interconnectedness of supply chains and vast differences between cultures that may
influence the adoption of risk mitigation strategies. Research has shown that several differences
exist between countries that may systematically influence how risks are mitigated (Lytle et al.,
1995). To ensure effective risk mitigation, these cultural differences must be acknowledged to
ensure that interfirm relationships support risk mitigation. Despite the known cultural differences
between countries (Hofstede, 2001) and the reliance on international supply partners for building
a resilient supply chain (Pettit et al., 2010), the supply chain risk management literature has yet
to study these cultural implications. This chapter investigates how culture influences the
perception of supply chain risk and the types of strategies used to mitigate them. I do this in
multiple stages. First, I develop measures of risk mitigation across the supply chain, test whether

41

these measures are invariant, and finally, examine the how these measures differ across countries
using cross sectional data.
Risk mitigation is one of the key steps of the risk management process and involves the
preemptive implementation of strategies to either reduce the probability and/or impact of a
disruption (Kern et al., 2012). Culture influences the behavior of a firm’s employees (Schein,
1985), which likely results in different in risk mitigation patterns. Making decisions involving
risk is particularly exposed to being influenced by culture, requiring better explication of how
cultural differences influence the perception of risk. Hofstede’s (1980a, 1980b, 1984, 2001)
seminal work on culture is widely used and identifies six primary dimensions that vary between
cultures: Power Distance, Individualism-Collectivism, Masculinity, Uncertainty Avoidance,
Long Term Orientation, and Indulgence.
Of the cultural dimensions, Power Distance, Uncertainty Avoidance, Individualism, Long
Term Orientation are directly related to how decisions are reached on mitigating risk. Power
distance is the distance between subordinates and their bosses in their ability to make decisions
(Hofstede, 1984). When power distance is high, there is a lack in autonomy to make risk
mitigation decisions and the strength of voice that subordinates may exhibit (Brockner et al.,
2001). Since much of the tacit knowledge relevant for optimizing risk management is contained
in employees involved in everyday operations whom have low power, accounting for power
distance in designing strategies may be important. Uncertainty avoidance refers to the level of
comfort individuals have in making decisions with high degrees of uncertainty. Since adopting
risk mitigation strategies is highly uncertain, countries with high uncertainty avoidance may be
less flexible in the choices they make since they are more likely to abide by strict rules of
conduct (Hofstede, 1980, 2001). The degree to which people prefer to work individually or

42

within a group is referred to as individualism (or equivalently its inverse, collectivism). Since
employees act as agents for a firm, higher levels of individualism (lower degree of collectivism)
may result in the adoption of strategies that may take more of an individual’s opinions or
experiences into account. Finally, long term orientation refers to the length of time horizon that
individuals prefer for planning. Thus, people in a high long term orientation society highly value
planning and are more likely to accept short-term losses (Ashkanasy et al., 2004). This may also
influence risk mitigation decisions because they are preemptive and may require a long-term
view. Consistent with this idea, Chopra and Sodhi (2004) found that managers focus on low
impact reoccurring events rather than long-term disruptive events. Despite the importance of
cultural dimensions on risk mitigation, studies in the current literature have not examined this
aspect.
Risk mitigation is undoubtedly a vital component of a high performance supply chain, but the
extent and type of risk mitigation strategies to implement is not as clear because of the many
ways in which risk can be mitigated. Despite the variety in methods for mitigating risk, they all
function by providing a firm with slack. Cyert and March (1963) asserted the importance of slack
in determining organizational behavior when they proposed a behavior theory of the firm. Slack
is “…that cushion of actual or potential resources which allows an organization to adapt
successfully to internal pressures for change in policy, as well as to initiate changes in strategy
with respect to the external environment” (Bourgeois, 1981, p. 30). Further, organizational
perceptions of slack can also be influenced by culture, and consequently influence organizational
response (Bansal, 2003). This suggests that culture is an important attribute to consider in the
context of risk management. Slack is not unidimensional and differs in its availability for use.
Firms keep slack in different ways to combat uncertainty. For instance, unabsorbed slack is kept

43

to be used immediately while absorbed slack can be appropriated when necessary (Tan and Peng,
2003). While several authors have studied the importance of slack for innovation (Nohria and
Gulati, 1996) or financial performance (Bromiley, 1991), I focus on developing a more thorough
understanding of the role slack plays in mitigating supply chain risk, and test for cultural
differences in perceptions of the slack.
Focusing on slack, two specific types have been proposed in the literature that are relevant
for risk mitigation: absorbed or unabsorbed (Sharfman, 1988). Absorbed slack is derived from
resources already committed to current operations. Managers may have less discretion in
modifying this form of slack. In contrast, unabsorbed slack is derived from resources that are yet
uncommitted (Lee and Wu, 2015). Within the domain of supply chain risk, I relate these two
types of slacks to flexibility and redundancy, two common supply chain risk mitigation
strategies. I argue that the theory of slack provides a consistent basis for categorizing risk
mitigation practices and a taxonomy based on the buffering mechanisms of these strategies.
There are many different strategies that managers can use to mitigate supply chain risks which
have the characteristics of absorbed or unabsorbed slack. Further, managerial perceptions about
these can differ based on their cultural vantage points and business practices (Newman and
Nollen, 1996).
Focusing on risk mitigation strategies, the supply chain literature has argued that flexibility
and redundancy are important primary strategies for mitigating risk. For example, Chopra and
Sodhi (2004) argue that flexibility and redundancy are among the most fundamental risk
reduction strategies of the eight proposed. Similarly, Sheffi and Rice (2005, p. 41) note: “An
organization’s ability to recover from disruptions quickly can be improved by building
redundancy and flexibility into the supply chain.” Sodhi and Tang (2012) describe flexibility and

44

redundancy as two of three risk mitigation categories. While other studies have developed
alternative classifications of risk mitigation strategies based on their observations (Johnson et al.,
2013; Scholten et al., 2014), flexibility and redundancy are reoccurring classifications used
across the literature. Finally, both flexibility and redundancy, as I will discuss, are measures that
can be anchored on the foundations of slack. Accordingly, I focus on these two dimensions in
this study.
Finally, it is important to recognize that supply chains are not monolithic entities. Rather,
supply chains are composed of different parts in the overall firm’s value chain – namely supply,
process and demand. Both flexibility and redundancy are implemented across the different stages
of the supply chain and their implementation in individual stages can have an influence on the
overall chain. It is likely that managerial approach to managing flexibility and redundancy are
culturally distinct in each of these stages. In studying cultural differences, and developing
measures for supply chain risk, this study focuses on better explicating these differences across
the overall supply chain.
Overall this study makes several important contributions to the literature. First, I develop a
measure of redundancy and expands scales of flexibility to the domain of risk with the
consideration that these strategies differ across the value chain. Measures that have already been
developed for supply chain risk have not yet captured these elements of mitigation and are
usually operationalized at a higher or lower level of analysis; refer to Table 5. Secondly, I apply
the theory of slack to supply chain risk mitigation strategies so that the different typologies
proposed in the literature can be related to one another by their theoretical mechanisms. Lastly, I
uncover the differences culture makes in risk perception and mitigation. I achieve this by

45

creating a survey instrument, collecting data with it in three countries, and testing the invariance
between the scales.
2. Theoretical Development
The literature on supply chain risk management is continuing to grow and yield valuable insight
for managers and researchers. However, there are still several gaps in the literature, including a
dearth of empirical studies (Sodhi et al., 2012). I contribute to filling that gap by developing
measures of redundancy and expanding the context of flexibility. The following section builds on
past research to create a theoretical understanding of the differences between the two types of
strategy based on organizational theory of slack.
2.1. Slack
The forms slack can take include financial reserves (George, 2005) excess labor (Steele and
Papke-Shields, 1993), or any other resources that are in excess of what is required for business
operations. In reality, maintaining zero slack is often impossible and the challenge for managers
is not deciding to have slack or not, but determining how much slack is optimal. According to
lean manufacturing principles, slack is a waste and can lead to inefficient and poor performance
(Inman and Mehra, 1993). On the other hand, slack is necessary to buffer against variability in
lead times, production delays, uncertain demand, as well as other environmental fluctuations.
Therefore, determining the appropriate types and levels of slack is necessary for managing
supply chain risks (Chopra and Sodhi, 2004; Kleindorfer and Saad, 2005).
The supply chain risk management literature describes many different practices that are used
to mitigate supply chain disruptions which have the properties of slack. For instance, extra
capacity (Bourland and Yano, 1994) and inventory (Hendricks et al., 2009) are both well-known
supply chain risk mitigation strategies and have been studied under the lens of slack resources.

46

These types of slack have been termed operational and differ in accessibility from more general
financial measures of slack (Bourgeois 1981). Bourgeois (1981) argued that operational slack is
embedded in production, and therefore less available than more discretionary organizational
slack, such as financial reserves. Considering the variety of operational practices used for
mitigating supply chain risks, it would be incorrect to assume that all forms of operational slack
have the same degree of discretion. Huang and Chen (2010) show that operational slack can be
more easily deployed than past research would suggest. Rather than subsuming all operational
practices into a broad classification of operational slack, I seek to understand two types of slack I
define as redundancy and flexibility. These two types of operational slack differ by the
mechanism in which they buffer against shocks and their availability for deployment.
2.1.1. Available Slack
Redundancies are a form of slack by definition. Operations managers use redundancy to ensure
resilience to disruptions both within and external to their firm. The function of redundancies is
best understood from a systems or network perspective. In graph theory, scale free networks
have been shown to be resistant to random attacks to the network (Albert and Barabasi, 2002).
This phenomenon can be generalized to supply chain networks since they have been shown to
have the characteristics of scale free networks (Sun and Wu, 2005). Resistance in graph theory
refers to the structure of the network maintaining its structural characteristics after disruptions
occur in the form of ties or nodes being removed from the network. This resilience is largely a
result of redundancy. Similar to when a supplier were to be removed from a supply chain, when
a node (firm) is removed, all of its vertices (connections) are destroyed as well. However, the
connectivity of the network is maintained as a result of a substantial number of redundant
vertices linking the network together.

47

From a systems engineering perspective, redundancies are a key parameter in determining
the reliability of a system. Supply chains can be viewed as systems since they have a similar
structure to mixed systems that contain both a series and parallel structure for the movement of
goods and information. Optimization techniques can be implemented to determine the necessary
number of redundancies to achieve certain levels of reliability in systems (e.g., Chern, 1992;
Ghare and Taylor, 1969). Similarly to supply chains, redundancies also have a cost associated
with them which needs to be considered when determining optimal levels of redundancy. By
understanding how redundancies operate in systems and networks, it is apparent that they are
deployed immediately in the event of a disruption since they replace the resource they are a
duplicate of. This basic function is the same for improving business continuity in supply chains.
Therefore, the classic assumption that all operational slack is unavailable may be invalid and
limit the understanding of slack.
Most studies in supply chain management focuses on only one type of redundancy, primarily
in manufacturing. For instance, Hendricks (1992) sought to determine optimal buffer allocation
in serial production lines to maximize throughput and minimize variability. Other examples
include buffer capacity (e.g., Enginalar et al., 2002) and inventory stocking levels (e.g., Ha,
1997). Similar to research on internal redundancies used for manufacturing, most research on
upstream or downstream risk mitigation strategies only investigates a single practice. For
instance, Yu et al. (2009) model single versus dual sourcing on supply chain performance with
disruptions. They argue that single sourcing has advantages, such as cost and relational
efficiencies, but can be risky in the case of a major disruption and show that the choice of
sourcing decision depends on magnitude of disruption probability. Bartezzaghi and Verganti
(1995) explore the risk mitigation benefits of over-planning under demand uncertainty by

48

constantly adjusting slack. These studies share a common theme, a focus on cost reduction rather
than risk minimization. I build on these studies by developing a construct of redundancies that
are kept to minimize risk.
2.1.2. Unavailable Slack
Ironically, flexibility has been used in a wide variety of settings, representing many different
business practices. For the purpose of this study, I focus on supply chain flexibility used to
mitigate risk which incorporates internal, upstream, and downstream practices. Slack can also be
thought of in terms of flexibility. Levinthal (1997) described the mechanism of slack in
navigating rugged landscapes by the ability to adapt. Adaptation is the fundamental property of
strategies classified as flexible that afford risk reduction. Based on this description, flexibility
provides the potential to adapt to change so is a form of slack, but since it functions by
adaptation, it is less discretionary.
Studies testing the influence of flexibility on supply chain risk have been few, and primarily
analytical. One exception was a study by Lee and Makhija (2009) that showed the effect of
strategic flexibility, as measured by international investments, on firm value during an economic
crisis. Tang and Tomlin (2008) show the value of flexibility in mitigating supply chain risks
using five stylized models of multiple suppliers, flexible supply contracts, flexible manufacturing
processes, postponement, and responsive pricing. They concluded that relatively low levels of
flexibility significantly reduce risk and that there are decreasing returns to scale for investments
in flexibility. While flexibility has been studied extensively, it has been done so primarily in the
area of manufacturing flexibility. Slack (1983) provided a foundation for the literature to build
from through an examination of flexibility as a manufacturing objective that provides completive
advantage. Upton (1995) revisited manufacturing flexibility by highlighting the many challenges

49

managers face in implementing flexibility. Although, a broader perspective on flexibility needs
to be adopted to account for all of the risks firms are exposed to. Swafford et al. (2006)
addressed this issue when they developed measures for flexibility, separating flexibility into
three functional areas: procurement/sourcing, manufacturing, and distribution/logistics.
Sharfman (1988) clarified the definition of slack by distinguishing it from waste and other
buffers. The importance of this distinction is that flexibility must be purposely created for
mitigating risk. Consider two pieces of manufacturing machinery, one can only produce one
product, while the other can produce that same product as well as other products when necessary.
The equipment that can produce multiple products fits the classification as flexible, but only if
the additional function is purposely reserved to buffer risk. Although, flexibility is not purely an
excess resource which will affect how easily it can be deployed. Continuing with the previous
example, a piece of manufacturing equipment requires time to change over to producing a new
product and cannot perform its primary role when being applied to mitigate risk. The additional
time and opportunity cost render flexibility less discretionary than redundancy as a form of slack.
Therefore, it can be thought of as a potential or unavailable slack. In one study that compared
these two types of slacks, Talluri et al. (2013) show in a simulation that flexibility is a more
efficient strategy to adopt than redundancy. The reason for this finding is that despite the slack
being less discretionary, is still functions as a buffer without sitting idle as available slack does.
2.2. Supply Chain Risk
Reviewing the literature exposes another critical distinction that is missing in studies on slack:
the source of uncertainty that is buffered. Previous literature on slack has predominantly
investigated the benefits of slack on general uncertainty, without matching the types of slack to
specific sources of uncertainty. A few exceptions do exist. For instance, Caputo (1996) discussed

50

the difference buffers play in internal and external uncertainty through illustrative examples.
Cheng and Kesner (1997) went further to hypothesize and test the different roles internal and
external slack play in resource allocation. However, the effectiveness of slack under different
sources remains underdeveloped, especially when focusing on supply chain risk.
The classification of risk in supply chains varies substantially in the literature. Based on a
review on the literature, Ho et al. (2015) divide supply chain risk into Macro and Micro risks.
This classification aligns closely with those that other scholars have proposed, such as
catastrophic/operational (Sodhi et al., 2012) and disruption /operational (Tang, 2006). Ho et al.
(2015) further divide micro risks into four categories: supply, manufacturing, demand, and
infrastructural. Christopher and Peck (2004) categorize the sources of supply chain risk similarly
into five elements: supply, process, control, demand, and environmental. Environmental risk
refers to the inherent risk all firms face, while two of these risks are external to the firm, the
other two internal. Internally, control and process risks are those that threaten the sequence of
value-adding steps undertaken by the firm and procedures that govern them. Demand and supply
risks are the two external sources of risk that threaten product, material, or information flows
either down or upstream of the firm. External sources of risk are harder to detect (Christopher
and Peck, 2004), and as a consequence harder to dedicate the appropriate level of slack towards.
Rather than risk, Bode and Wagner (2015) show that three types of complexity, horizontal,
vertical and spatial, are related to an increase in supply chain disruptions. These same authors
also studied the antecedents of three types of supply chain risk, demand-side, supply-side, and
catastrophic (Wagner and Bode, 2006). Other types of risk that threaten the continuity of supply
chains have been labeled infrastructural and include transportation, information technologies,
and financial systems (Ho et al., 2015). In proceeding, I focus the study on micro supply chain

51

risks and classify them as internal (process, manufacturing and control) and external risk (supply
and demand).
2.3. Cultural Influences on Risk Mitigation
Deciding which type of slack to apply to mitigate internal or external risk may be a function of
culture as well as organizational requirements. One of the main challenges in measuring supply
chain risk is that context influences risk perceptions between organizations (Desai, 2008). The
risk management literature suggests that buffers are chosen based on the exposure to risk, but
has not shown how culture influences the allocation of these buffers.
The main difficulty in optimally mitigating any risk is that they are uncertain. Thus,
effectively mitigating risk requires utilizing as much information as possible. In countries where
power distance is high, managers make decisions and their subordinates do not conflict with their
decision. In countries with low power distance, managers’ decisions are more likely to be
challenged if their decisions do not align with their subordinates. While disagreement can lead to
conflict, conflict may be a source for discourse that may lead to better decisions (Amason, 1996).
Based on the impact power distance may have on decision making, I expect differences in risk
mitigation adoption between countries that have varying levels of power distance.
Collectivism “…requires that individuals sacrifice themselves for the alleged interests of the
collective” (Realo et al. 2008, p.448). Thus, collectivism may influence the choice of slack
applied to mitigate risk because of the inherent desire to ensure the safety of the group. Power et
al. (2010) argue that collectivism influences operational goal orientation and that high
collectivism is associated with greater goal congruence. Therefore, collective goals such as
mitigating firm risk are a priority over individual achievements that may lead to suboptimal
decision (Ramamoorthy et al., 2007). In combination with power distance, a focus on collective

52

goals suggests the power distance enhances collectivism so that managers will engage with their
subordinates in order to achieve goal congruence. This may lead to a greater application of long
term strategies that are best suited to mitigate risk comprehensively rather than addressing minor
reoccurring problems. Potential slack requires greater investment and planning than available
slack, consistent with the goal of achieving the greatest level of risk mitigation in the long term.
Additionally, flexibility, often associated with potential slack, has been shown to be more
effective at mitigating risk than redundancy, another term for available slack (Talluri et al. 2013).
Two other dimensions of Hofstede’s cultural thesis that are highly relevant for this study are
uncertainty avoidance and long term orientation. Uncertainty Avoidance is “…the reliance on
norms and procedures to alleviate unpredictability’ (Kull and Wacher 2010, p. 228). Relying on
norms and procedures is associated with a greater application of fact-based decision-making and
scientific methods (Naor et al., 2010). However, Weber and Hsee (1998) show that the
perception of risk was the key difference in risk behavior between cultures. Thus, I expect
countries that are high in these two will have a higher degree of perceived risk. Western
countries have lower uncertainty avoidance and long term orientation scores than do Confucian
Asian countries. Developing a survey to study risk management should cover a broad
international region since the scope of supply chains is worldwide. Accounting for these
potential cultural influences are necessary when designing a survey based on the previous
discussion. In the following sections, I seek to uncover how these differences influence
measuring risk and mitigation strategies.

53

Figure 3. Cultural Differences

3. Methods
To develop the measures, I collected data using a cross-sectional survey instrument. The survey
was administered in three first world countries: the United States of America, Australia, and
Japan. These countries were chosen to provide a strong contrast in cultural values while
providing as close to a geographical control as possible. The USA and Australia have similar
cultural dimensions, both with a stark contrast to Japan. Meanwhile, Australia and Japan are
located in the south pacific across the Pacific Ocean from the USA. Supply chain strategies and
configuration may be a result of location as well as culture. Since I am trying to investigate the
role culture plays in risk perception and mitigation, I chose the closest country to Japan that has
similar cultural dimensions as the USA in order to provide the best control group available. I
created the survey instrument by following standard development techniques (Churchill Jr, 1979;
Dillman, 2006). The survey was carefully crafted to ensure high quality responses by minimizing
cognitive burden and maximizing interest in the survey through question ordering. The survey

54

was pretested by several academics and practitioners before administration, resulting in a few
minor structural and grammatical adjustments to the instrument. Anonymity was ensured to limit
social desirability, and were not given a detailed introduction to the study, limiting response bias.
Furthermore, measures were scaled to specific rates of use whenever feasible or anchored to a
specific situation to limit variance in measurement due to perception (Doty and Glick, 1998).
Table 5. Risk Measurement Summary
Articles
Wagner and Bode, (2008)
Weiss and Maher, (2009)
Ellis et al., (2010)
Bode et al., (2011)
Blackhurst et al., (2011)
Kern et al., (2012)
Wagner and Neshat , (2012)
Grotsch et al., (2013)
Hoffmann et al., (2013)
Ambulkar et al., (2015)

Operationalization of risk mitigation
Risk and Risk management
Operational Hedging in the Airline industry
Alternative suppliers
Supply chain disruption orientation
Global Supply Resiliency
Multi-item measure of risk mitigation
Risk management, risk planning and supply chain performance
Organizational characteristics and SCRM pro-activeness
Formative measure of SCRM
Firm Resilience

Supply chain, procurement, and logistics managers were the target respondents and they were
limited to working for manufacturing firms that have experience managing supply chain risks.
Third parties were contracted to administer the internet based survey to managers in the US and
Australia. I relied on research partnerships to assist in data collection in Japan. Japan Institute of
Logistics Systems (JILS) reviewed the survey, and sponsored it by allowing it to be sent to their
member base. The survey was translated into Japanese by a professional translator and compared
to the English version of the survey by a bilingual expert in supply chain management. In order
to ensure compatibility, several minor corrections in terminology and language were necessary
after review. Collectively, I received a total of 209 complete high quality responses. The
response rates varied by country, but were only available for the USA. The USA survey was sent
to a panel of 7280 potential respondents, of which 803 responded (11% response rate). This is

55

comparable to other studies containing respondents with similar job titles. Many different
industries are represented, with a median firm age of 1950 in the American sample, 1980 for the
Australian sample, and 1915 for the Japanese sample. The majority of the sample respondents,
82.5%, were very experienced, with 10 years of cumulative experience, 24.8% of them being at
their current position for that same length of time. Non-response bias was tested by comparing
the average size and revenue of the firms in the final sample to those of the non-respondent.
Non-response bias was assessed with the USA sample and t-tests for both size and revenue were
insignificant (size: p=0.53; revenue: p=0.84); refer to Table 6.
Table 6. Demographic information
Firm characteristics
Age (Years)

Risk management
program (years)

Size (employees)

Sales ($)
Respondent
characteristics
Experience in firm

Total experience

Japan
USA
Australia
Japan
USA
Australia
Japan
USA
Australia
Japan
USA
Australia

N
55
98
44
46
101
44
58
110
44
57
110
44

Median
101
66
35
5
2
5
250 to 499
5,000 to 9,999
10 to 49
1 bill to 10 bill
500 mill to 1 bill
2 to 5 mill

Min
1
1
1
0
0
0
2 to 4
6 to 9
1
<50,000
<50,000
<50,000

Max
124
179
115
25
50
15
>25,000
>25,000
>25,000
>10 bill
>10 bill
>10 bill

Japan
USA
Australia
Japan
USA
Australia

N
57
109
44
58
101
44

Median
1 to 3 years
6 to 10 years
6 to 10 years
>10 years
>10 years
>10 years

Min
< 1 year
< 1 year
< 1 year
1 to 3 years
1 to 3 years
< 1 year

Max
1 to 3 years
>10 years
>10 years
>10 years
>10 years
>10 years

56

3.1. Constructs
Flexibility was measured using a scale developed by Swafford et al. (2006) that included three
dimensions of flexibility: procurement, manufacturing, and logistics. I renamed these dimensions
in the study as supply, process and demand respectively. This measure was chosen over other
similar measures of flexibility because the items consisted of practices that have been considered
risk mitigation in past studies and it captures flexibility at each stage along the supply chain.
Redundancy was measured by creating a scale that mirrored the one used for flexibility. Items
for the redundancy measures were created from practices classified in the literature as
redundancy. This led to a list of 16 practices which were subject to pretesting by five graduate
students. This led to a reduction of 4 items per stage in the supply chain. Further pretesting of the
survey instrument was conducted with both managers and academics to establish face validity
and ensure appropriate wording of the items. The feedback from the pretest resulted in the
rewording of several items to align with current practices, improvement of clarity, and removal
of doubled barreled items; refer to Table 8, 9 & 10 for a full list of measures.
Wagner and Bode’s (2008) measures of risk were used for the study, but adapted to measure
both probability and magnitude. Risk is commonly quantified in practice based on the impact
and likelihood of an event. Measuring only one dimension, as originally done by Wagner and
Bode (2008), would not capture risk as it is defined, potentially limiting the usefulness of the
measure. Thus, I collect data on both probability and magnitude for each item and multiply them
together to create each measure. All items were measured on a 7-point Likert scale.

57

Table 7. Summary Statistics
Items N Mean St.Dev.
a) Absorbed Slack
11 211 4.156 1.138
b) Unabsorbed Slack 11 211 3.681 1.108
c) Internal Risk
4 209 10.804 7.614
d) External Risk
7 209 13.126 6.460
*Note: Composite reliabilities (CR) on diagonal

a

b

0.893
0.747 * 0.890
-0.028 -0.038
0.042 0.015

c

d

0.848
0.600 * 0.897

3.2. Validity
A measurement model (confirmatory factor analysis, CFA) was created for each of the survey
items by mapping them onto their respective latent constructs. Estimation was conducted using
maximum likelihood in Stata 14. Since risk and flexibility were adaptations of past measures, I
begin validity testing by comparing the CFA results with published statistics. First, comparing
the measures of flexibility to Swafford et al.’s (2006) measures shows that each of the measures
exhibited better composite reliability (CR) and average variance explained (AVE) than the
original; refer to Table 8. This suggests that the survey adequately reproduced the variables of
interest. The measure of redundancy did not have a benchmark to compare to, but is above
accepted levels for new constructs; refer to Table 9. Since redundancy is a new measure, only
items with a factor loading above .5 were retained, leading to the removal of one item from the
process stage.
Next, the measures of risk were compared to the original values reported by Wagner and
Bode (2006); refer to Table 10. Supply risk displays similar CR and AVE, process risk has
slightly higher CR and AVE, and demand risk has slightly lower CR and AVE. The variance in
reliability statistics is likely due to the difference in operationalization of the measures. Note, in
this study, supply and demand risks are combined into a single construct of external risk and the
individual comparisons were conducted for validity purposes only.

58

The presence of common method bias was tested using the marker variable technique.
Richardson et al. (2009) show that the CFA marker variable is the best of the available methods,
but still needs to be viewed with caution. The marker variable is a measure of process technology
usage within a company and was chosen because the target respondents should have strong
knowledge of these types of technologies and the variance in their use should be associated with
the products companies produce rather than the risk mitigation strategies adopted or risk
exposure. To begin, a CFA was conducted with all of the constructs, including the unconstrained
marker variable. The unstandardized path loadings were obtained, and used as constraints in a
baseline model along with constraining the correlation between the marker variable and all
others to zero. The second model was identical to the baseline, with the addition of paths from
the marker latent construct to each item. The third model constrained the paths added in the
previous step to be equal. Finally, the correlations between latent constructs from the
unconstrained model were added as constraints to the second model. The chi-square difference
tests between each stepwise model were statistically significant, suggesting common method bias
is present. While this is not definitive proof of common method bias and may be due to the lack
of a perfect marker variable (Richardson et al., 2009), I proceed with caution in using the data.
Fortunately, rather than relating variables to one another using regression techniques that would
pose endogeneity, my research aim only requires I differentiate between groups responses.

59

Table 8. Comparison of factor loadings and reliability for risk mitigation measures
Risk Mitigation scales and associated items

Supply Flexibility
Extent to which your supplier lead-time can be
expedited/changed
Extent to which your supplier short-term capacity can be
adjusted
Extent of flexibility (options) within supplier contracts
Extent to which share of supplied components can be
reallocated between suppliers
Process Flexibility
Range of products manufactured by your firms’ plants
Average number of products that each manufacturing facility
can produce
Number of product changeovers per manufacturing facility
made each month
Range of production volumes across which manufacturing can
accommodate
Demand Flexibility
Average number of items handled by each distribution facility
Average number of items per order handled by each
distribution facility
Average number of customers supported by each distribution
facility
Number of carriers
Number of distribution facilities

60

Risk mitigation
Îťi
(CR=0.841;
AVE=0.571)

Swafford et al.
(2006)
Îťi
(CR=0.600;
AVE=0.313)

0.771

0.560

0.783
0.805

0.870
0.300

0.655
(CR=0.835;
AVE=0.560)
0.777

0.300
(CR=0.606;
AVE=0.342)
0.560

0.847

0.960

0.666

0.330

0.689
(CR=0.913;
AVE=0.779)
0.942

0.160
(CR=0.815;
AVE=0.493)
0.830

0.869

0.870

0.832
-

0.830
0.360
0.450

Table 9. Factor loadings and reliability for risk mitigation measures
Risk mitigation
Îťi
(CR=0.760; AVE=0.448)
0.722

Risk Mitigation scales and associated items

Supply Redundancy
Priority in stocking inventory from suppliers
Amount of capacity reserved by your suppliers in case of
changes in requirements
0.731
Number of alternative/backup suppliers selected on a per
component basis
0.617
Suppliers’ ability to produce components not currently
sourced from them
0.596
Process Redundancy
(CR=0.677; AVE=0.414)
Amount of capacity reserved in case of sudden changes
in demand
0.731
Average number of manufacturing plants that can
produce each product
0.571
Amount of equipment stored in each plant to produce
components not currently produced in that plant
0.617
Demand Redundancy
(CR=0.771; AVE=0.460)
Amount of inventory held by your warehouses to meet
demand requirements
0.575
Amount of space reserved in warehouses in case of
sudden changes in demand
0.774
Extent to which demand at a specific location can be
reassigned to a different location
0.728
Number of worldwide storage/distribution facilities
0.618

61

Table 10. Comparison of factor loadings and reliability for risk measures
Risk scales and associated items

Supply Risk*
Poor logistics performance of suppliers (delivery dependability,
order fill capacity)
Sudden default of a supplier (e.g., due to bankruptcy)
Supplier quality problems
Poor logistics performance of logistics service providers
Capacity fluctuation or shortages on the supply markets
Demand Risk*
Unanticipated or very volatile customer demand
Insufficient or distorted information from your customers’ quantities
about orders or demand

Impact*Probabil Wagner & Bode
ity
(2008)
Îťi
Îťi
(CR=0.848;
(CR=0.828;
AVE=0.445)
AVE=0.497)
0.639
0.659
0.646
0.667
0.732

0.585
0.731
(CR=0.897;
AVE=0.685)

0.824
0.711
0.797
0.636
0.513
(CR=0.833;
AVE=0.713)
0.877
0.811
(CR=0.817;
AVE=0.529)

Internal Risk**
Downtime or loss of own production capacity due to local
disruptions (e.g., fire, industrial accidents, etc.)
0.793
0.82
Perturbation or breakdown of internal IT infrastructure (e.g., caused
by computer viruses, software bugs)
0.853
0.724
Loss of own production capacity due to technical reasons (e.g.,
machine deterioration, poor quality)
0.798
0.705
Perturbation or breakdown of external IT infrastructure
0.864
0.649
Note: *Supply and Demand risk are combined as a single construct, external risk, in this study. The
collective composite reliability and average variance extracted are shown. The comparative values are
Supply: CR=0.802; AVE=0.448 and Demand: CR=0.607; AVE=0.438
**Wagner and Bode (2008) originally titled this risk ‘infrastructure’ but was renamed for consistency
in this paper.

3.3. Data analysis
3.3.1. Measurement Invariance
The next step of my analysis is to show that the meaning of constructs does not vary between
countries. To accomplish this, I tested for measurement invariance across the three countries;
refer to Table 11, 12 & 13. Since the objective of the analysis is to compare the means of the
constructs between countries, full metric invariance is not required, nor likely considering the
strictness of the invariance requirements (Horn, 1983). Metric and scalar invariance of at least
two items per construct is required to make defensible quantitative group comparisons of means

62

(Byrne et al., 1989; Meredith, 1993). I conduct each step in the measurement invariance process
while focusing on the two forms of invariance required to test differences in means of the
constructs between countries.
Measurement invariance is a stepwise process which progressively adds stricter constraints to
the measurement model. The first step is to estimate a CFA including a country level latent
variable with all parameters freely estimated. Parameters are constrained to be equal across
countries in the following order: factor loadings, intercepts, residuals, factor means, and
variances (Gregorich, 2006). Invariance is supported when the chi-square test between models is
insignificant (i.e. not rejecting the null hypothesis, Ho: measurement is not different between
groups) or the stepwise CFI increments is less than .01 (Vandenberg and Lance, 2000).
The null hypothesis was rejected for every step in the invariance test procedure for the
measure of flexibility. Thus, flexibility did not display measurement invariance and pooling the
data without accounting for country differences would result in biased estimates. Conducting a
post-hoc inspection of modification indices shows most factor loading are not invariant across
groups. The required change in the hypothesized model would be too drastic to selectively relax
constraints and pursue partial metric invariance. For the other two constructs, redundancy and
risk, metric invariance was displayed, but only one of the stricter tests of invariance were
supported. Metric invariance refers to an insignificant difference of factor loading between
countries, suggesting that the meaning of the constructs is the same across countries. For the
redundancy measures, modification indices indicated that two items’ intercepts may be
responsible for the lack of scalar (strong) invariance. The intercept constraints of the first items
for demand and supply were relaxed based on the modification indices, resulting in a significant
improvement in model fit in comparison to the original scalar invariance model (∆χ2 (4) 105.18,

63

p<0.001) and an insignificant fit in comparison to the metric invariance model (∆χ2 (12) 5.12,
p>0.1). Thus, partial scalar invariance was supported for redundancy and the means of the latent
constructs can be compared meaningfully across countries using the partial invariant model.
Turning to the risk measures, partial scalar invariance may also be achieved if a couple of the
constraints on item intercepts are relaxed. Upon inspection, modification indices suggest that the
last external risk item is primarily responsible for the poor fit. Also, the fifth external risk item
for Japan appears significantly different than both Australia and USA. Therefore, the constraint
that the fifth external risk intercept be constant for Australia and USA is retained, while the other
two intercepts are relaxed. Comparing the fit statistics of the partial model to the metric
invariance model provided support for partial scalar invariance (∆χ2 (15) 18.05, p>0.1) while
showing a marked improvement over the original model (∆χ2 (3) 30.85, p<0.001). Both chisquare difference and CFI difference test provided the same evidence of null hypothesis rejection
for every test conducted.
The only stricter form on invariance that was observed for risk and redundancy was full
measurement invariance, despite each step leading up to it being significant. This indicates that
the variances of the items are invariant, but the residuals and means are not. Thus, pooling the
data is inappropriate for conducting statistical tests such as path analysis. Since this is beyond the
scope of this paper, I do not pursue testing for partial invariance at stricter levels of measurement
invariance. I note however, that this result has significant implications for researchers conducting
international survey studies on supply chain risk management. Pooling data must be done with
caution and may lead to erroneous conclusions if done without measurement invariance.

64

Table 11. Measurement invariance: Flexibility
Model

Hypothesis

Test

1

Base

2

Metric invariance

2/1

3

Strong invariance

3/2

4

Strict invariance

4/3

5
6

-

Strict invariance plus
factor means
Strict invariance plus
factor means &
variances

5/4
6/5

Description of Hypotheses

χ2 (d.f.)

∆χ2
(d.f.)

All parameters free

462.40
(132)

Factor loadings are constrained to be equal across
groups
Factor loadings and intercepts are constrained to be
equal across groups
Loadings, intercepts and residuals are constrained to be
equal across groups
Loadings, intercepts, residuals and factor means are
constrained to be equal across groups

511.65
(148)
564.92
(164)
603.87
(180)
645.92
(186)

49.25
(16)
53.27
(16)
38.95
(16)
42.05
(6)

668.15
(198)

22.23
(12)

χ2 (d.f.)

∆χ2
(d.f.)

Loadings, intercepts, residuals, factor means and
variances are constrained to be equal across groups

p

CFI

∆CFI

0.755
0.000

0.731

0.024

0.000

0.703

0.028

0.001

0.686

0.017

0.000

0.659

0.027

0.035

0.652

0.007

p

CFI

∆CFI

Table 12. Measurement invariance: Redundancy
Model
1

Hypothesis
Base

Test
-

2

Metric invariance

2/1

3

Strong invariance

3/2

4

Strict invariance

4/3

5
6

Strict invariance plus
factor means
Strict invariance plus
factor means &
variances

5/4
6/5

Description of Hypotheses
All parameters free

750.47
(132)

Factor loadings are constrained to be equal across
groups
Factor loadings and intercepts are constrained to be
equal across groups
Loadings, intercepts and residuals are constrained to be
equal across groups
Loadings, intercepts, residuals and factor means are
constrained to be equal across groups

767.61
(148)
877.91
(164)
904.72
(180)
919.55
(186)

17.14
(16)
110.30
(16)
26.81
(16)
14.83
(6)

937.60
(198)

18.05
(12)

Loadings, intercepts, residuals, factor means and
variances are constrained to be equal across groups

65

0.447
0.377

0.446

0.001

0.000

0.361

0.085

0.044

0.352

0.009

0.022

0.344

0.008

0.114

0.333

0.011

Table 13. Measurement invariance: Risk
Model

Hypothesis

Test

1

Configurational
Invariance

2

Metric invariance

2/1

3

Scalar invariance

3/2

4

Strict invariance

4/3

5
6

Strict invariance plus
factor means
Strict invariance plus
factor means &
variances

-

5/4
6/5

Description of Hypotheses

χ2 (d.f.)

∆χ2
(d.f.)

All parameters free

739.13
(135)

Factor loadings are constrained to be equal across
groups
Factor loadings and intercepts are constrained to be
equal across groups
Loadings, intercepts and residuals are constrained to be
equal across groups
Loadings, intercepts, residuals and factor means are
constrained to be equal across groups

753.03
(153)
801.93
(171)
847.41
(189)
864.52
(193)

13.9
(18)
48.90
(18)
45.48
(18)
17.11
(4)

866.33
(199)

1.81
(6)

Loadings, intercepts, residuals, factor means and
variances are constrained to be equal across groups

66

p

CFI

∆CFI

0.593
0.736

0.588

0.005

0.000

0.557

0.031

0.000

0.543

0.014

0.002

0.532

0.011

0.936

0.529

0.003

Full metric invariance along with partial scalar invariance was observed for the measures of
redundancy and risk, allowing us to proceed to the next step in testing the differences in
constructs between countries. Note, while flexibility did not display measurement invariance, I
still include it in the further analyses for completeness with the understanding that differences in
mean values are likely a result of the latent variables having different meanings across countries
and not to culture. I test the differences in latent means by including a path from an observed
country construct to the latent variables (Dimitrov, 2006). Since country is a categorical variable,
I estimate the means using a generalized structural equation model (GSEM) in Stata 14. With the
USA as the reference group, the path from Australia to both internal and external risk is
insignificant while for Japan, external risk and internal risk are both positive and significant
̅̅̅̅̅̅̅̅̅𝑗𝑎𝑝 = 1.88, p <0.10 & 𝑅𝑖𝑠𝑘𝑖𝑛
̅̅̅̅̅̅̅̅̅𝑗𝑎𝑝 = 3.48, p <0.01); see Table 14. To compare Australia
(𝑅𝑖𝑠𝑘𝑒𝑥
and Japan, Australia was held as the reference group. For internal risk, Japan and Australia
displayed no significant difference while external risk was significantly higher in Japan
̅̅̅̅̅̅̅̅̅𝑗𝑎𝑝 = 2.87, p <0.05). Analysis of the difference in redundancy reveals a similar pattern
(𝑅𝑖𝑠𝑘𝑒𝑥
but in the opposite direction. Redundancy was significantly lower in Japan than the USA
̅̅̅̅̅̅̅̅̅̅𝑗𝑎𝑝 = -0.73, p <0.001; 𝑅𝑒𝑑𝑝𝑟𝑜
̅̅̅̅̅̅̅̅̅̅𝑗𝑎𝑝 = -0.78, p <0.001; 𝑅𝑒𝑑𝑑𝑒𝑚
̅̅̅̅̅̅̅̅̅̅̅𝑗𝑎𝑝 = -0.73, p<0.001).
(𝑅𝑒𝑑𝑠𝑢𝑝
Process redundancy was significant and negative when comparing Australia to the USA
̅̅̅̅̅̅̅̅̅̅𝑎𝑢𝑠 = -0.36, p<0.05). With Australia as the reference group, all three dimensions of
(𝑅𝑒𝑑𝑝𝑟𝑜
̅̅̅̅̅̅̅̅̅̅𝑗𝑎𝑝 = -0.48, p <0.05; ̅̅̅̅̅̅̅̅̅̅
redundancy for Japan were significant and negative (𝑅𝑒𝑑𝑠𝑢𝑝
𝑅𝑒𝑑𝑝𝑟𝑜𝑗𝑎𝑝 =
-0.42, p <0.10; ̅̅̅̅̅̅̅̅̅̅̅
𝑅𝑒𝑑𝑑𝑒𝑚𝑗𝑎𝑝 = -0.49, p <0.05). For flexibility, Japan and Australia are not
significantly different, while the USA has positive and significantly greater supply and process
̅̅̅̅̅̅̅̅̅̅𝑈𝑆𝐴 = 0.53, p <0.01; ̅̅̅̅̅̅̅̅̅̅
flexibility than Australia (𝑅𝑒𝑑𝑠𝑢𝑝
𝑅𝑒𝑑𝑝𝑟𝑜𝑈𝑆𝐴 = 0.58, p <0.05). With the
USA as the reference group, Japan has a significant and negative level of flexibility for each

67

̅̅̅̅̅̅̅̅̅̅𝑗𝑎𝑝 = -0.82, p <0.001; 𝑅𝑒𝑑𝑝𝑟𝑜
̅̅̅̅̅̅̅̅̅̅𝑗𝑎𝑝 = -1.00, p <0.001; 𝑅𝑒𝑑𝑑𝑒𝑚
̅̅̅̅̅̅̅̅̅̅̅𝑗𝑎𝑝 = -0.76, p
dimension (𝑅𝑒𝑑𝑠𝑢𝑝
<0.01).
Table 14. GSEM Construct Between Country Comparison
Japan-USA

Îťi
External Risk
1.882+
Internal Risk
3.480**
Supply Flex
-0.817**
Process Flex
-0.995**
Demand Flex
-0.764**
Supply Red
-0.731**
Process Red
-0.778**
Demand Red
-0.735**
Note: +p<0.1, *p<0.05, **p<0.01

se
(1.117)
(1.101)
(0.181)
(0.231)
(0.276)
(0.191)
(0.186)
(0.167)

Japan-Aus.

Îťi
2.024
2.866*
-0.285
-0.414
-0.297
-0.476*
-0.416+
-0.493*

se
(1.371)
(1.340)
(0.216)
(0.270)
(0.339)
(0.231)
(0.225)
(0.192)

Aus.-USA

Îťi
-0.141
0.614
-0.532**
-0.581*
-0.468
-0.255
-0.362+
-0.241

se
(1.213)
(1.173)
(0.193)
(0.235)
(0.300)
(0.204)
(0.185)
(0.163)

Figure 4. Risk Mitigation Comparison

Risk Mitigation Comparison
5
4.5
4
3.5
3
2.5

Japan
USA

2

Australia
1.5
1
0.5
0
Supply
Mitigation Flexibility

Process
Mitigation Flexibility

Demand
Mitigation Flexibility

Supply
Mitigation Redundancy

68

Process
Mitigation Redundancy

Demand
Mitigation Redundancy

Figure 5. Perceived Risk Comparison

Preceived Risk Comparison
18
16
14
12
10

Japan
USA

8

Australia

6

4
2
0
Internal Risk

External Risk

Table 15. MANOVA results
Japan

USA

Australia

MANOVA

Variable

N

Mean

σ

N

Mean

σ

N

Mean

σ

External Risk

56

15.392

6.525

109

13.288

7.985

44

12.368

7.163

2.77

0.010

3.91

0.001

1.31

0.248

Internal Risk

56

13.661

6.819

109

9.580

7.389

44

10.199

8.311

3.40

0.011

3.96

0.005

1.21

0.307

Absorbed Slack

57

3.626

1.386

110

4.464

1.108

44

3.919

1.433

6.26

0.000

2.38

0.010

1.44

0.155

Unabsorbed Slack

56

3.165

1.199

110

3.988

1.039

44

3.686

1.260

3.94

0.000

1.28

0.247

2.28

0.011

69

Japan-US

Japan-Aus.

US-Aus.

3.3.2. Robustness
CFA accounts for the multidimensionality of reflective latent constructs, but the measures may
also be thought of as formative since they consist of practices firms use to create redundancy and
flexibility in their supply chains. Since the distinction between reflective and formative measures
is not always perfectly clear, I use a MANOVA to show the differences between the constructs,
further supporting my analyses. Wilks’ criteria, F tests, and their associated p value for each of
the country differences are displayed in Table 15. The MANOVAs were conducted for each of
the three pairs of countries in the sample. Perceived internal and external risks are significantly
different between both US and Japan and Australia and Japan. There is not a significant
difference in perceived risk between US and Australia, providing further support that the
difference is due to cultural dimensions. Despite the higher perceived risk in Japan, the level of
risk mitigation is significantly lower in Japan than USA for both flexibility and redundancy.
Meanwhile, there is no significant difference in flexibility between the US and Australia, while
there is a significant difference between Australia and Japan. For redundancy, there is a
significant difference between the USA and Australia, but not one between Japan and Australia.
This suggests that the difference in flexibility may be cultural while the difference in redundancy
could be due to regional or other constraints in the supply chain.
4. Discussion
This study develops a measure of risk mitigation and investigates differences in both the
perception of supply chain risk and the usage of risk mitigation strategies in different countries. I
advance the theoretical distinction between two types of risk mitigation strategies based on how
they function and refine their measurement. This is an important contribution to the literature
because without consistent language and proper measurement, research cannot advance a field

70

(Meredith, 1993). By leveraging theories on slack, flexibility and redundancy are argued to have
unique characteristics that mitigation risk in fundamentally different ways. The challenges in
measuring risk were also explored and I show that care must be taken when measuring risk over
broad regions since differences in culture may affect how people respond to questions regarding
risk.
Flexibility and redundancy are both effective methods for reducing risk and they should not
be viewed as being implemented independent of the other, but rather as different options that
may be used to combat risk in a unique ways. The two risk mitigation strategies may be used in
combination or independently. Reiterating claims made in the literature on slack, I emphasize
that flexibility and redundancy need to be applied based on the threats a firm is exposed to
coordinating them across the supply chain is vital for building resilience. I contribute to the
literature by further advancing the importance of flexibility and redundancy as risk mitigation
strategies while developing a measure for redundancy. I argue that these forms of slack are
unique in classifying strategies firms implement to buffer against uncertainty. Additionally, I
encourage future studies to clearly articulate which stage of the risk management process they
are focusing on and to account for the application of slack across the supply chain.
In contrast to the neoclassical economic view, slack is not always a waste that needs to be
optimized out of a system, but a necessary parameter in maximizing long term financial
sustainability. The types of slack applied in different countries were shown to vary significantly.
Japan, despite their high perceived risk, implements lower levels of both flexibility and
redundancy than the two western culture countries. This is important for supply chain risk
managers to realize as it may impact the types and levels of strategies they apply internally.

71

Understanding the differences between cultures may improve communication between supply
partners, which in turn may improve the resilience of a supply chain.
Flexibility did not exhibit measurement invariance. While this result did not allow for the
comparisons between countries to be associated to cultural differences, it suggests that how
flexibility is understood varies between countries. This may be a result of the complexity of
unavailable slack. Flexibility is a more complex form of slack than redundancy since
redundancies are easily quantifiable and more universal in operations management. The degree
and location of unavailable slack can vary substantially within the operations of a firm,
potentially influencing how each respondent across country or firm boundaries understands the
construct. Extra precaution must be taken in administering surveys on flexibility for mitigating
risk.
Supply chain risks extend beyond firm boundaries, requiring careful coordination between
suppliers, distributors and retailers to ensure adequate risk mitigation strategies are in place.
However, suppliers may not share the same perspective on risk which may result is
misappropriated resources to mitigate them. In this study I show that the perception of risk and
the degree to which risks are mitigated varies significantly across cultural borders.
Communication is necessary for a resilient supply chain (Chopra and Sodhi, 2004; Stevenson
and Spring, 2007). Without an understanding that suppliers from regions with well-defined
cultural differences have systematic different understanding of what their exposure to risk is,
coordinating risk management may be ineffective. I hope that this study informs managers of
how cultural differences manifest themselves in the context of risk management so that
communication and resource allocation is improved.

72

4.1. Limitations
Unfortunately, this study suffers from several drawbacks that limit its empirical potential. While
redundancy and risk both display partial scalar invariance, they are not invariant at the stricter
levels required to pool the data and conduct path analysis. Requiring respondents with the
specific profiles necessary for this study made it difficult to meet the high observation
requirements of SEM. As a result, I only received 109 usable surveys from the USA, 58 from
Japan and 44 from Australia. Even with a simple model, using the appropriate methodology for
reflective psychometric constructs, SEM, only permits estimating between 10 and 43 free
parameters using the recommended 20:1 to 5:1 ratio for sample size to free parameters. In order
to conduct path analysis with data from different countries that are not strictly invariant, GSEM
is the recommended methodology to account for higher order country effects. The number of
free parameters required to estimate higher order GSEM models increases by a multiple of the
number of groups in comparison to traditional SEM, increasing the sample size requirements
considerably. I attempted to reduce many of the drawbacks of the survey methodology and in the
future would like to pursue alternative methods for measuring risk mitigation to further limit the
challenges of survey research.
Managerially, understanding these difference is important since supply chain are
international and working with supply chain partners with different perceptions of risk and risk
mitigation practices can be problematic for several reasons. When creating supply contracts,
difference is risk perceptions may lead to different valuations of a contract and inhibit
negotiation. Alternatively, parties from different countries may be negotiated on different terms,
leading to a false understanding that may lead to conflict. Risk may also be transmitted through a
supply chain based on the practices adopted by supply chain partners. A firm may be impacted

73

by the misalignment in risk mitigation practices between their supply partners. For example, a
supply partner may focus on unavailable slack in the form of machine flexibility, meanwhile
have high capacity utilization with the expectation that the downstream partners utilize available
slack, such as keeping an adequate about of safety stock. Once an unforeseen spike in demand
occurs, the risk may be amplified downstream.

74

Chapter 4 - Evaluating the Performance of Risk Mitigation Systems: A Frontier Approach
1. Introduction
Global supply chains are fraught with risk and managing these risks is an integral part of any
organization. The risk management literature stresses the strategic importance of supply chains
and that they be resilient to disruptions in the flow of materials and information (Narasimhan and
Talluri, 2009; Sodhi and Tang, 2012). In the quest to build resiliency, most firms invest in risk
mitigation strategies to limit the cost and likelihood of a disruption taking place. Yet, these
investments may counter the fundamental premise of efficiency with which the supply chain
operates. Thus, simply implementing a system of risk mitigation policies, procedures, and
activities is not always effective. Tony Hayward, former CEO of British Petroleum (BP), focused
on improving risk management within the organization when he was hired in 2007. Despite his
strategic orientation toward managing risks, the catastrophic Deepwater Horizon oil spill in the
Gulf of Mexico happened while he was in charge of the company. Not only did the loss of the oil
rig disrupt BP’s supply chain, it resulted in fines and restoration efforts that cost BP billions of
dollars. The root cause of the incident was identified by an U.S. government committee as a
failure in“…the ability of individuals involved to identify the risks they faced and to properly
evaluate, communicate, and address them.” (Kaplan and Mikes, 2012, p. 48). Evaluating risk and
their mitigation strategies could have helped BP avoid the disaster or reduce its impact.
Evaluation is a prerequisite for the other two steps highlighted in the investigation,
communication and addressing risk, since information from the evaluation can be used as a
benchmark for improvement and is easily communicated. It is also important to evaluate the
returns on investments of risk mitigation strategies across the supply chain. Assessing the
effectiveness of implementing these strategies also aligns with the supply chain’s fundamental
goal of improving efficiency. This issue forms the core research question of this study: from a
75

central planner’s perspective, how can firms most effectively evaluate the performance of their
risk mitigation strategies?
Evaluating supply chain risk mitigation strategies presents multiple challenges. First, each
firm faces a unique set of risks and has many choices in the strategies to adopt that may mitigate
those risks. This presents a situation of equifinality; there are multiple paths to take to achieve a
specific goal (Doty and Glick, 1994). Second, risk and the mitigation strategies adopted to
address them are often linked. Supply chains are not monolithic entities and as the flow of goods
are dependent on each linkage in the chain, so too are risk mitigation strategies. For instance,
O’Riely (2014) notes that implementing dual sourcing as a risk mitigation strategy may not be
effective without considering reliability of downstream activities. In another example, the textile
industry is known for its lack of flexible supply capacity so contracts are fixed far in advance,
increasing the risk of stockouts and excess inventory (Eppen and Iyer, 1997). In order to mitigate
this risk, backup agreements can be used to increase the flexibility of obtaining goods from a
supplier, but this may have implications for other risks a firm is exposed to and the strategies
used to mitigate them. Specifically, the type, quantity and location of inventory may be impacted
as well as the modes of transportation of the goods to the market. In contrast, semiconductors
have a long production run times and short technology cycles. This environment requires
different approaches to mitigating risk than the textile industry, such as flexibility in plant
design, number of backup suppliers, and new product development processes. Thus, a
combination of the diverse risk mitigation methods throughout the echelons of the supply chain
requires firms to coordinate the different approaches of mitigating risks. Lastly, resources are
limited in any firm, so the cost of risk mitigation is a critical component of evaluation. Thus, it is
the job of a supply chain manager to critically assess the resources channeled to risk mitigation

76

effort and the consequent outcomes obtained from these efforts. In accounting for these
challenges, this study proposes a method for evaluating risk mitigation strategies that addresses
each of the aforementioned challenges.
Firms must balance potentially conflicting objectives of cost efficiency of strategies pursued
while minimizing supply chain risks (Lee et al., 2004). Whether it is holding extra capacity,
stocking extra inventory, developing options contracts, or establishing secondary supply sources,
implementing risk mitigation strategies come at a cost beyond what is necessary to deliver value
to customers. Many performance measurement techniques, such as the balanced scorecard
(Kaplan and Norton, 1996), fail to account for cost while only focusing on performance levels.
Evaluating the performance of risk mitigation strategies must account for the cost of them to
facilitate the appropriate allocation of resources among the strategies that provide the greatest
reduction in risk. Further complicating the decision making process, these costs are borne
regardless of a disruption occurring and the likelihood of a disruption is uncertain. O’Reilly
(2014) highlights the difficulty in balancing meeting demand and overinvesting. Part of the
reason is that managers view the cost of mitigation as an insurance premium and the return on
investment is difficult to quantify (Sheffi, 2001). Specifically, managers can invest significant
resources to minimize the risk, but that may not benefit the firm overall due to high costs. In this
regard, to create a cost-efficient supply chain that is resilient to disruptions, risk mitigation
strategies themselves must be implemented efficiently (Tang, 2006a). The varying cost and
effectiveness of different methods of mitigating risk demand efficient utilization of resources
applied to the appropriate sources of risk is a key objective of performance evaluation.
Literature on supply chain risk management is growing significantly and several facets have
been studied. However, measurement and evaluation literature in risk management is at a nascent

77

stage. This study aims to fill those gaps and makes multiple contributions to the literature. First,
most of the current risk management literature considers the supply chain as either a monolithic
entity or as individual and independent parts. I acknowledge the importance of each element in
the supply chain and develop a comprehensive framework of risk that considers each of these
entities and the interdependences between them using a Network Data Envelopment Analysis
(NDEA). It is likely that firms may gain differentially from investing in each aspect of their
supply chain and the NDEA analysis allows us to investigate how these elements help mitigate
risk. Second, Ambulkar et al. (2015) focus on resilience as a key component of supply chain risk.
However, building resilience is expensive and none of the current studies in the literature have
focused on measuring the costs of building resilience. I fill this gap by measuring the costs of
risk mitigation strategies in addition to the level of implementation. I note that this gap in the
literature is particularly relevant with respect to the orientation for efficiency within supply chain
management. To illustrate, I demonstrate the importance of including cost when evaluating the
performance of risk mitigation practices to better understand the cost benefit trade-offs in supply
chain risks through the use of a NDEA approach. NDEA uniquely allows identification of these
trade-offs. Specifically, cost benefit trade-offs may not just happen between individual strategies
(such as flexibility and redundancy), but across different entities in the supply chain. For
example, firms may more efficiently mitigate one component of supply chain risk as compared
to others, while another may be more efficient in process risk mitigation, and still be at the
frontier.
2. Literature Review
The literature most relevant to this study is that of supply disruptions and supply chain
performance evaluation. Within these literature streams, I focus the review on research that

78

quantifies risk mitigation practices for performance evaluation of supply chain risk. Mitigating
risk can be accomplished using various practices and is a vital component of the supply chain
risk management process (Kleindorfer and Saad, 2005). Kern et al. (2014, p.81) define risk
mitigation as a process that “…incorporates all activities concerned with deciding how to deal
with a certain risk and executing the chosen strategy.” The practices that can be used to mitigate
supply chain risk vary substantially, from collaboration to geographic dispersion; refer to
(Ceryno et al., 2015) for a comprehensive list. Mitigation strategies are used to manage many
types of risks, such as establishing improved supplier relationships to increase upstream visibility
or adopting revenue management policies to shift demand, thereby reducing demand
fluctuations. Thus, classification systems are often developed to distinguish the mitigation
strategies according to the risks they mitigate or the mechanism by which they function. While
several such classifications have been proposed, redundancy and flexibility are two of the most
common (Braunscheidel and Suresh, 2009; Chopra and Sodhi, 2004; Kern et al., 2012).
Accordingly, I focus on these two for this study. Redundant and flexible risk mitigation
strategies differ based on the mechanisms of risk mitigation. I define redundancy as, the
provision of duplicate resources to ensure business continuity in the case of a risk event.
Inventory, multiple suppliers of the same components, and multiple production facilities that
produce the same products are redundant mitigation strategies. Flexibility is defined as, the
ability of a firm to easily change or adapt resources to ensure business continuity in the case of a
risk event. Examples of flexibility are options contracts with suppliers or equipment that can be
used to produce many different products. Flexibility has been argued to be the better of the two
in mitigating risk (Talluri et al., 2013), but a comprehensive risk mitigation system likely

79

requires both. Further, a balance must be made with these strategies depending on the specific
environment a firm is in.
With many different sources of risk and methods to mitigate them, a one-size fits all
approach is inappropriate. Thus, a ‘tailored’ strategy is required (Chopra and Sodhi, 2004),
suggesting that there are multiple ways to achieve a cost-efficient low risk supply chain. This
situation of equifinality has been further categorized into 4 types, based on the degree of conflict
in functional demands and the latitude of structural options (Gresov and Drazin, 1997). Based on
the classification of Gresov and Drazin (1997), a risk mitigation system would be classified as
‘Configurational Equifinality’ since there are many strategies that can be adopted and many
types of risk to mitigate. Specifically, there is not an ideal set of practices that will result in the
highest performance, which needs to be accounted for during performance evaluation and
measurement. Under this form of equifinality, the performance of an organization, “…results
from identifying a subset of functional demands that minimize functional conflict, and matching
these demands with a set of appropriate structural features that are internally consistent” (Gresov
and Drazin 1997, p.416). This highlights two of the dilemmas in performance measurement of
risk mitigation: deciding which strategy to pursue and benchmarking. Specifically, effective
performance management (PM) provides a benchmark for strategy (Koufteros et al., 2014) and is
a critical link between strategy implementation and its financial effectiveness (Melnyk et al.,
2004). Yet, the literature on supply chain risks is scant on methods to measure the performance
of risk mitigation strategies.
2.1. Performance Measurement
Many scholars have developed frameworks and argued for the value of PM systems (Bititci et
al., 1997; Bititci and Turner, 2000; Neely et al., 2005; Gunasekaran et al., 2001; Estampe et al.,

80

2013). The importance of an effective PM system to assess risk mitigation is highlighted by a
firm’s resource limitations and the strategic importance of a firm’s supply chain. In the context
of a supply chain, measuring the performance of the supply chain as a whole has been shown to
be a key element of firm competitiveness (Gunasekaran et al., 2001). In order for a supply chain
to perform smoothly, it must be resilient to disruptions, suggesting that risk management should
be considered as an integral part of PM for a supply chain. While investing resources in risk
mitigation is important, how they are deployed is argued to be as important as which resources
are possessed by a firm (Hansen, 2004). Further, these resources need to be ‘orchestrated’ to
maximize their effectiveness (Sirmon et al., 2011). Koufteros et al. (2014) show that the
orchestration of resources is facilitated by PM. PM may improve supply chain performance by
improving decision making and facilitating effective strategy implementation (cf., Henri 2006).
Extending these arguments to risk mitigation is especially relevant since the resources invested
in risk mitigation may not be directly applied toward delivering value to customers, but rather act
as an insurance (Sheffi, 2001). Accordingly, these resources need to be allocated as efficiently as
possible.
As a rule, managers want to avoid any negative impact from the risks they face and adopting
flexibility and redundancy risk mitigation strategies is considered good practice (Sheffi and Rice
Jr, 2005). However, the extent and type of risk mitigation strategies that firms need to implement
to achieve specific outcomes is not as clear. Supply chain risk is unique in that risks are interconnected because supply chains are interdependent across its stages. This presents a situation
where investments in individual stages may have a differential influence on the efficiency of risk
mitigation. Ritchie and Bradley (2007) echo this argument as it is a key aspect in their
framework relating risk mitigation and performance measurement. Similarly, Lambert and

81

Pohlen (2001) argue that PM systems should reflect the organizational structures. Yet, current
studies relating to risk mitigation in supply chains do not take this inter-connected nature of
various supply chain stages into account. Accordingly, in this study, I categorize the interconnected risks and mitigation strategies as supply, process and demand since they represent the
flow of materials through a supply chain which accounts for the requirement of PM in operations
along the stages of the supply chain (Fortuin, 1988). Each stage of the supply chain has specific
risk mitigation practices associated with it and these practices are reviewed in the following
sections.
2.1.1. Supply risk mitigation
Despite the fact that supply chain risk involves more than just the supply side of a firm’s
operations, measuring supply risk mitigation strategies have been the focus of more empirical
research than either process or demand risk. Since this study is concerned with performance
measurement, I acknowledge omitting many important studies from the literature review, but I
focus on empirical articles that cover the measurement of risk mitigation strategies. The
measures used in the literature are primarily survey based and have not been very consistent. For
instance, Avittathur and Swamidass (2007) surveyed supplier flexibility and showed it had an
impact on growth, profitability, and inventory. However, this study was focused on positive
outcomes and did not directly test the relationship of plant flexibility to supply chain risk.
Wagner and Bode (2008) developed a broad six item measure of risk management that includes
supply risk mitigation strategies, including business continuity plans, supplier collaboration,
supplier monitoring, and insurance. Blome et al. (2012) show that supply chain agility fully
mediates the relationship between supply competence and operational performance. The authors

82

use a latent agility construct that captures the extent to which a firm can ‘adapt’, ‘react’, or
‘adjust’ to several supply and production activities.
Rather than mitigation, Groetsch et al. (2012) measure a similar construct of supply chain
risk management reactiveness using a three item construct developed using context analysis of a
set of interviews. Hoffman et al. (2013) provide the most direct measurement of risk mitigation
by identifying 22 risk mitigation activities through a discussion workshop. The authors then
refined their measure to 6 items that were used as a formative construct in an empirical model.
Kern et al. (2012) also measure a latent construct of supply risk mitigation, but do so as a
reflective, three item construct. Ellis et al. (2010) measure risk mitigation in their study on
supply disruption risk as a single survey item of ‘search for an alternative supplier’. In their
study on supply chain vulnerability, Wagner and Neshat (2012) developed a measure of supply
chain risk planning and use three psychometric items that contain risk mitigation strategies to
quantify it. While not intended to be exhaustive, these studies highlight the inconsistency of
measuring risk mitigation and the importance of assessing their performance since each study
relates risk mitigation to some form of performance.
2.1.2. Process risk mitigation
Mitigating disruptions due to demand or supply uncertainty are challenging because they involve
actors external to the firm while risks in production processes are internal and should be more
easily controlled. However, process risk still presents a difficult challenge for operations
managers and is also a key element of supply chain risk. Both redundancy and flexibility are
frequently studied as risk mitigation strategy in this context. For instance, smaller batch sizes
(Betts and Johnston, 2005) and postponement (Tang and Tomlin, 2008) are examples of
manufacturing activities that have been defined as flexible risk mitigation strategies.

83

Redundancies are very commonly applied in manufacturing to reduce risk and examples include
overcapacity in production (Thun et al., 2011) and inventory (Hendricks et al., 2009). From a
performance evaluation perspective, research on flexibility and redundancy have provided
several measures of each type of risk mitigation strategy, but the tradeoff between cost-risk and
other strategies still needs investigation (Vokurka and O'Leary-Kelly, 2000).
Several empirical measures of process flexibility have been developed using surveys, but not
in the context of supply chain risk. In addition to supply flexibility, Avittathur and Swamidass
(2007) used a survey to assess the impact of plant flexibility on growth, profitability, and
inventory by quantifying plant flexibility as a factor score of four items. Other scholars have
dissected plant flexibility into 5 dimensions: machine, material, product, labor and new product
development flexibility (Koste et al., 2004; Patel and Jayaram, 2014; Patel et al., 2012). While
there exists many measures of process flexibility and it is well established to improve
competitiveness, few studies have investigated the impact of process flexibility on supply chain
risk, despite the many qualitative and analytical studies supporting the relationship. In one
example of secondary data analysis, Weiss and Maher (2009) showed in the airline industry that
operational hedging mitigated the financial impact of the terrorist attacks on September 11, 2001.
They provided four measures of operational hedging including fleet standardization, utilization,
ownership and globalization, limiting the generalizability of the study to other industries. This
study builds on this literature stream by applying flexibility to a performance evaluation
framework for mitigating risks.
Arguably the most critical redundancy strategy used internally within a firm to buffer
uncertainty is inventory. Inventory is a common buffer against supply and demand fluctuation
since it is controlled internally and can be deployed immediately. Consequently, inventory is an

84

important input to mitigating risk. Raw inventory and inventory turnover are widely adopted
measures in practice and empirical research. For instance, Hendricks et al. (2009) show that days
of inventory has no impact on negative stock market returns during a disruption. Additionally,
inventory efficiency is positively associated with stock market returns and not significantly
related to stock market risk (Mishra et al., 2013). Accordingly, inventory is an element of
redundancy I examine as a process risk mitigation strategy. Further, I also account for the cost of
inventory in examining its efficiency. Overall, I focus on both flexibility and redundancy as risk
mitigating mechanisms internal to a firm in evaluating its overall supply chain risk mitigation
system.
2.1.3. Demand risk mitigation
In a review of demand risk mitigation techniques, Ho et al. (2015) found four clusters of
research: optimal-order placement, forecasting, contracts, and other. Although, most of that
research is analytical so does not involve measurement in the performance evaluation process.
Regardless, many of the studies in this domain highlight the key elements of mitigating demand
risk. Demand risk is “risk associated with the outbound logistic flows and product demand”
(Ceyrno et al. 2015, p.1150). Factors that contribute to product demand risk include inaccurate
demand forecast, demand variability, and information distortion (Ho et al., 2015). Appropriate
forecasting and flexibility in contracts are common methods of combating demand risk. For
example, risk-sharing contracts have been shown to reduce risk for both the manufacturer and
retailer (Chen, et al., 2006; Xiao and Yang, 2009). Quantity flexible contracts, for example, help
absorb demand fluctuations and reduce demand uncertainty (Kim, 2013). The forecasting models
that have been proposed as methods to mitigate product demand risk include the macro-

85

prediction market model (Guo et al., 2006) and the generalized autoregressive conditional
heteroskedastic model (Datta et al., 2007).
Logistic flow risks require different mitigation strategies than product demand uncertainty to
mitigate risk and research on product demand exceeds that of logistic flow significantly. Vehicle
routing is the most common method for mitigating logistics risk (Giaglis et al., 2004). Due to the
increased use of information technologies and advances in computing power, vehicle routing is
possible to be adjusted in real time, offering flexibility that could reduce risk. In the event of a
disruption, Beroggi and Wallace (1995) show that real-time tracking information can be used by
applied with decision support systems to minimize cost and risk associated with an unexpected
event. Zhang et al. (2009) developed a greedy upper bounding and Lagrangian relaxation
algorithm to show an optimal method for rerouting in the case of interruptions in the flow in a
manufacturing facility. Other methods for mitigating logistic flow risk are maintaining excess
distribution capacity and flexible distribution channels (Huchzemeier and Cohen 1996; Cohen
and Huchzemeier 1999; Novaes 2000). Huchzemeier and Cohen (1996) and Cohen and
Huchzemeier (1999) develop stochastic models that show how supply chain network design can
reduce logistic flow risk. Novaes (2000) present a case to display the importance of logistic flow
mitigation strategies.
While the studies presented above show that demand risk mitigation is effective at reducing
demand uncertainty, the empirical measurement and testing of these present a gap in the
literature. For instance, the two forecasting models developed by Datta et al. (2007) and Guo et
al. (2006) do not attempt to measure the degree of forecasting used to mitigate demand risk,
much less the effectiveness of these forecasting tools in practice nor the cost of implementing
them. The other studies cited in this section follow a similar pattern. In summary, the studies on

86

demand risk mitigation have not measured the strategies resulting in a lack of reliable methods to
evaluate the performance of a risk mitigation system holistically.
Overall, the collective review of demand, process and supply risk reveals the following. First,
supply side risk is the focus of most empirical research on supply chain risk and has been
primarily been conducted by surveys. This may be a result of supply chain research
fundamentally concerned with upstream suppliers and the difficulty is obtaining data on supply
chain practices across firms. Second, several multi-item measures have been developed to
measure supply risk mitigation strategies. However, these measures either measure flexibility or
flexibility and redundancy practices are grouped together in a single latent construct of risk
mitigation. From a theoretical and practitioner standpoint, more insight can be gained if the two
strategies are studied in relation to one another than separately or grouped as a single construct.
Furthermore, none of the current empirical studies measure the investment in risk mitigation
strategies, a critical component of the total costs that firms invest to manage supply chain risk.
This presents a conflict between the measurement and the supply chain objective of efficient
deployment of resources. Accordingly, this study measures both mitigation of risks and accounts
for the corresponding costs. I also measure mitigation strategies at each stage of the supply chain
to achieve a more comprehensive evaluation of risk mitigation.
3. Methods
To build and test the performance evaluation methodology of a firm’s risk mitigation system, I
used the data collected in Chapter 3. I only used the sample from the United States of America to
avoid issues due to cultural differences. See the previous chapter’s method section for details.
A third party was contracted to administer the internet based survey. Quality was ensured in
collaboration with the third party by removing a) quick responses (less than 6 minutes), b)

87

random or illogical responses, c) overuse of non-responses, and d) failed responses to the two
attention filter items. Of the 803 responses, 285 did not meet the requirements for the survey and
another 289 did not complete the survey. Collectively, I received a total of 229 complete
responses which were then subject to the quality control procedure previously described,
resulting in 109 usable responses. Non-response bias was tested by comparing the average size
and revenue of the firms in the final sample to those of the non-respondents. As mentioned in the
previous chapter, t-tests for both size and revenue were insignificant (size: p=0.53; revenue:
p=0.84). Many different industries are represented, with a median firm age of 66 years. The
majority of the sample was very experienced, with 10 years of cumulative experience, 35% of
them being at their current position for that same length of time.
3.1.

Measures

Flexibility was measured using a scale developed by Swafford et al. (2006) that included three
dimensions of flexibility: procurement, manufacturing, and logistics. I renamed these dimensions
in the study as supply, process, and demand respectively. This measure was chosen over other
similar measures of flexibility because the items consisted of practices rather than outcomes of
flexibility and it captures both the level (range) of flexibility and cost (adaptability) of the
mitigation strategies. Redundancy was measured by creating a scale that mirrored the one used
for flexibility, but replaced each item with practices classified in the literature as redundancy. For
example, a measure of process redundancy ‘Amount of capacity reserved in case of sudden
changes in demand’ reflects the process flexibility measure ‘Number of product changeovers per
manufacturing facility made each month’ by capturing the redundant mechanism used in
production to mitigate risk; refer to Table 17 for a full list of measures. Corresponding measures
of cost, measured both in terms of money and time, were collected for each measure using the

88

adaptability instrument developed by Swafford et al. (2006). In addition, a measure of financial
performance was obtained independent of the survey instrument by the survey agency for each
respondent and appended to the survey. This was done to limit common method bias, and is
quantified as firm revenue. Size was also captured as a control variable, quantified as the number
of employees that work for the firm.
To compare the proposed NDEA methodology to traditional approaches, I also operationalize
flexibility and redundancy using confirmatory factor analysis (CFA). The validity and reliability
of the measures was verified by creating two latent constructs of flexibility and redundancy,
covarying them, and estimating the factor loadings of each of their respective items using
maximum likelihood. The composite reliability of flexibility and redundancy were above
accepted levels (Flexibility = 0.724; Redundancy=0.806). Discriminant validity was tested using
two methods, the average variance extracted to squared error comparison and the HeterotraitMonotrait cutoff (HTMT) as recommended by Voorhes et al. (2015). The Heterotrait-Monotrait
ratio was less than the 0.85 cutoff (HTMT.85 =0.387) and the average variance extracted for each
measure was greater than the squared correlation between them (SE= 0.089; Flexibility AVE =
0.289 & Redundancy AVE = 0.222). The fit indices of the CFA were not ideal, but acceptable
considering the loadings are only being used to create a factor variable as a comparison against
the DEA based measure (χ2 = 420.515, p<0.01; RMSEA=0.083, 90% CI [0.096, 0.070]; CFI =
.805; SRMR= 0.110). Other descriptive statistics are shown in Table 17, 18 and 19.

89

Table 16. Demographic information
Firm characteristics
Age (Years)
Risk management program (years)
Size (employees)

N
98
101
110

Sales ($)

110

Respondent characteristics
Experience in firm
Total experience

N
109
101
Sent:
7280
%
15.09
7.55
11.32
7.55
4.72
2.83
10.38
1.89
15.09
4.72
7.55
1.89
4.72
3.77
0.94

Response Rate
Industry sector
Industrial Machinery
Electronics
Automotive
Chemicals
Metal Working
Pharmaceuticals
Consumer goods
Construction
General
Food, Beverages
Aerospace, defense
Telecommunication
Banking / Financial Services
Energy & Utilities / Oil & Gas
Healthcare / Medical

N
16
8
12
8
5
3
11
2
16
5
8
2
5
4
1

Median
Min
66
1
2
0
5,000 to 9,999
6 to 9
500 mill to 1
50,000
bill
Median
Min
6 to 10 years
< 1 year
>10 years
1 to 3 years
Taken:
Qualified:
738
109
Respondent function
Supply chain management
General management
Logistics
Purchasing
Others
Respondent job title
CEO/vice president
Director/department head
Manager
Others

90

Max
179
50
>25,000
>10 bill
Max
>10 years
>10 years

18
19
15
41
11

17.31
18.27
14.42
39.42
10.58

10
36
60
3

9.17
33.03
55.05
2.75

Table 17. Summary of Measures

Level

Cost

Risk
Mitigation
Flexibility

Echelon
Supply
Process
Demand
Redundancy Supply
Process
Demand
Flexibility
Supply
Process
Demand
Redundancy Supply
Process
Demand

Items Mean St.Dev.
4 4.665
0.863
4 4.805
0.992
4 4.759
1.030
4 4.342
0.894
4 4.317
1.032
4 4.489
1.110
8 4.109
0.641
8 4.071
0.679
8 3.898
0.851
8 4.079
0.639
8 4.131
0.688
8 4.002
0.715

Table 18. Comparison between aggregation methods
Risk Mitigation
Flexibility
Mean
St.Dev.
Min
Max
Redundancy Mean
St.Dev.
Min
Max

Equal
Weights
4.743
0.709
2.917
6.583
4.382
0.869
1.750
7.000

CFA
11.887
1.930
6.210
16.547
5.981
1.393
2.727
10.467

NDEA
0.542
0.213
0.281
1.000
0.516
0.248
0.247
1.000

91

DEA
0.711
0.176
0.403
1.000
0.641
0.169
0.370
1.000

Table 19. Correlation table (Kendall's tau)

Flex. (CFA)
Red. (CFA)
Flex. (E.W.)
Red. (E.W.)
Flex. (NDEA)
Red. (NDEA)
Flex. (DEA)
Red. (DEA)
Revenue
Size
Note: * p<0.05

Flex.
(CFA)
1.000
0.178
0.897
0.236
0.344
-0.003
0.429
0.069
0.080
0.068

Red.
(CFA)
*
*
*
*
*

1.000
0.210
0.552
-0.030
0.105
0.000
0.121
-0.159
-0.164

*
*

*
*

Flex.
(E.W.)

Red.
(E.W.)

1.000
0.265 *
0.315 *
-0.013
0.424 *
0.063
0.083
0.063

1.000
0.020
0.205 *
0.053
0.295 *
-0.047
-0.138 *

Flex.
(NDEA)

1.000
0.201
0.640
0.274
0.156
0.130

92

Red.
(NDEA)

*
*
*
*
*

1.000
0.175 *
0.642 *
-0.034
-0.058

Flex.
(DEA)

1.000
0.287 *
0.157 *
0.161 *

Red.
(DEA) Revenue

1.000
0.041
-0.020

Size

1.000
0.435 * 1.000

3.2.

Data Analysis

3.2.1. Modeling Efficiency of Risk Mitigation in a Network Context
Most research on risk mitigation strategies is focused purely on the level of mitigation strategy
and not the cost of implementing them. For instance, Brandon-Jones et al. (2014) survey supply
chain managers on the extent they use three risk mitigations: capacity, safety stock at the supplier
and plant. While the level of mitigation strategies implemented is important, they are
investments whose costs need to be considered in order to determine the financial impact they
have. Incorporating the cost of risk mitigation is challenging in empirical research because it is
often unavailable in typical secondary data sources. Accordingly, the input-output measures
come from the survey I conducted. Further, this study is concerned with the performance
measurement of risk mitigation, i.e., an efficient firm is one that maximizes risk mitigation at the
least possible investment in relation to other firms. Thus, it was deemed appropriate to use a
DEA methodology.
DEA is a mathematical programming method that calculates the efficiency of a decision
making unit (DMU) at minimizing multiple inputs while maximizing multiple outputs. In this
study, a DMU is a firm. Each firm receives a composite efficiency index ranging between zero
and 1, with efficient firms receiving a value of 1. Firms with scores less than 1 can be interpreted
as inefficient relative to the best firms in the sample. DEA assigns weights to all of the inputs
and outputs based on the optimization formulation shown in equation 1, without assuming that
the inputs and outputs are equally important or without subjectively assigning weights to them.
Specifically, being an input-output approach, it allows us to examine the gains of risk mitigation
vis-Ă -vis the cost of implementation.

93

In order to account for both the cost and interdependence of supply chain risk mitigation
strategies, I adopted a NDEA model. NDEA is an extension to the DEA model frequently used in
the management science literature that allows for a comprehensive optimization of a firm’s
productivity. A NDEA model relies on the same underlying optimization principles as DEA, but
links the inputs and outputs of different stages together (each stage is referred to as a division).
Thus, NDEA allows us to explicitly account for the interdependencies between inputs and
outputs across the different stages of the supply chain. Past studies on risk mitigation do not
explicitly consider these interdependencies across the various stages. This interdependency and
NDEA model is illustrated in Figure 1. Traditional DEA analysis assumes independence of
individual stages of the supply chain (and consequently risk mitigation strategies). This
assumption is likely invalid since organizations are integrated and the coordination between
divisions of the organization should have an impact on financial outcomes. Without accounting
explicitly for the interdependence between stages of supply chain management may lead to
inappropriate conclusions about the efficiency of individual risk mitigation practices, presenting
a limitation in assessing risk mitigation efficiency.
Figure 6. Network DEA

I apply a input oriented variable returns to scale method, incorporating the cost of risk
mitigation measured in both units of time and money as inputs and their respective degree of
mitigation level as outputs. The optimization function minimizes the cost of risk mitigation
relative to the level of risk mitigation. Thus, inefficient firms must minimize their inputs in order
to reach the efficiency frontier. Outputs from one division, in this case supply, are added as
inputs into the next stage, process, along with the original process inputs. This same dependence
is captured in the optimization calculation linking process outputs to demand as inputs; refer to
94

Figure 6. There are n DMUs (j = 1,…, n) consisting of K divisions (k= 1,…, K). Let mk and rk be
the number of inputs and outputs to division K. Linkages L from division k to h are denoted as
(k,h). The mathematical expression of the objective function and corresponding constraints are
shown below in equation 1.
1

𝑘−
𝑚 𝑠𝑖
𝑘 )]
𝑥𝑖𝑜

𝑘
𝑘
min𝜆𝑘,𝑠𝑘− ∑𝐾
𝑘=1 𝑤 [1 − 𝑚 (∑𝑖=1
𝑘

(1)
s.t.

𝒙𝑘𝑜 = ∑𝑛𝑗=1 𝒙𝑗𝑘 𝜆𝑗𝑘 + 𝑠𝑗𝑘−

∀𝑘

𝒚𝑘𝑜 = ∑𝑛𝑗=1 𝒚𝑘𝑗 𝜆𝑗𝑘 − 𝑠𝑗𝑘+

∀𝑘

∑𝑛𝑗=1 𝑒𝜆𝑗𝑘 = 1

∀𝑘

∑𝑛𝑗=1 𝑤𝑗𝑘 = 1

∀𝑘

∑𝑛𝑗=1 𝒛𝑗(𝑘,ℎ) 𝜆ℎ = ∑𝑛𝑗=1 𝒛𝑗(𝑘,ℎ) 𝜆𝑘 ∀𝑘, ℎ
𝜆𝑘 , 𝑠 𝑘− , 𝑠 𝑘+ , 𝑤 𝑘 ≥ 0,

∀𝑘

where 𝑠 𝑘− and 𝑠 𝑘+ are the input and output slack variables respectively; the input, output and link
𝑚

𝑟

(𝑘,ℎ)

vectors are represented as: 𝒙𝑗𝑘 ∈ 𝑅+ 𝑘 , 𝒚𝑘𝑗 =∈ 𝑅+𝑘 , 𝒛𝑗

𝑡

∈ 𝑅+(𝑘,ℎ) ; 𝑤 𝑘 is the relative weight of

Division k, i.e. supply, process, and demand, and is adapted from the weighted stochastic block
model (Cooper et al., 2007).
I also include a measure of flexibility and redundancy mitigation efficiency using a variable
returns to scale slack based DEA model for comparison (Banker et al., 1984). The variable returns
to scale model is identical to the original model developed by Charnes et al. (1978), but adds a
convexity constraint to account for the asymptotic efficiency changes in output relative to input.
The dual form of the linear program of the Banker, Charnes, and Cooper (BCC) model is
presented below:
+
−]
min 𝜃ℎ − 𝜀[∑𝑖 𝑆𝑖ℎ
+ 𝑆𝑟ℎ

𝑠. 𝑡.

+
∑𝑗 𝑥𝑖𝑗 𝜆𝑗 + 𝑆𝑖ℎ
= 𝜃ℎ 𝑥𝑖ℎ ∀𝑖
−
∑𝑗 𝑦𝑟𝑗 𝜆𝑗 − 𝑆𝑟ℎ
≥ 𝑦𝑟ℎ ∀𝑟

95

∑𝑗 𝜆𝑗 = 1
+
−
𝜆𝑗 , 𝑆𝑟ℎ
, 𝑆𝑖ℎ
≥0
+
−
where 𝑆𝑖ℎ
and 𝑆𝑟ℎ
are the input and output slack variables respectively. The ith input and rth

outputs are represented by 𝑥𝑖𝑗 and 𝑦𝑟𝑗 for each DMU j. The DMU under evaluation is indicated
by h. The dual variable is represented by 𝜆𝑗 , 𝜃ℎ is the efficiency score and 𝜀 is a small positive
scalar.
3.2.2. Linear Aggregation Methods
Next, to compare the NDEA operationalization of risk mitigation strategies, I apply two common
weighting schemes to the same data resulting in two additional measures. The first is an equal
weighting scheme commonly used to create factor scores of multi-item measures (Comrey and
Lee, 1992). Each item is assumed to contribute equally to the construct so a simple sum is used
to form an index of risk mitigation. This provides a baseline case for comparison. Many basic
performance measurement analyses used by organization often use such linear aggregation, for
example the balance scorecard (Kaplan and Norton, 1992). The second measure utilizes weights
derived from CFA analysis of the scales. The item loadings from the CFA are used as weights to
generate an index of risk mitigation (DiStefano et al., 2009). The formulation for these measures
is a weighted sum shown in equation 2 below.
Risk Mitigation scorej = ∑𝑚
𝑖=1 𝜌𝑖 𝑥𝑗𝑖

(2)

where xji denote the level of risk mitigation strategy i for firm j; 𝜌𝑖 is the weight for risk
mitigation strategy i and equals 1 for the equal case and the item loadings for the CFA case. Risk
mitigation scores were calculated for flexibility and redundancy.
3.3.

Results

NDEA not only provides a single measure of efficiency collectively for the supply chain as a
unit, but also scores for individual stages – supply, process and demand. Figure 7 and 8 show the
96

flexibility and redundancy histograms for the four operationalization’s of risk mitigation at the
supply, process, and demand levels. The differences between the distributions of these variables
are apparent. For both flexibility and redundancy, the linear aggregation based measures are bell
shaped and approximately normally distributed (the Shapiro-wilk test confirms this assertion for
a majority of the distributions). There is also not an apparent difference between the supply,
process, and demand levels for these measures. This is not the case for the NDEA scores. The
divisional scores account for the dependence between network components and reveal significant
differences between stages of the supply chain in their risk mitigation strategies. The supply
stage has a heavier distribution of lower NDEA scores than either the process or demand level
for both flexibility and redundancy. For flexibility, the percent of firms on the efficiency frontier
is also much less than process and demand. The DEA scores calculated individually for each
echelon display a similar distribution patter to the NDEA divisional scores. Despite the
similarity, the DEA scores are more evenly distributed than the NDEA scores for flexibility and
more heavily distributed in the middle of the range for redundancy.

97

Figure 7. Flexibility distributions

98

Figure 8. Redundancy distributions

Comparing the difference between distributions statistically supports the observations gained
from visually inspecting the histograms. Considering that the DEA based measures are nonnormal, I use tests that are robust against non-normality. First, I conducted six pairwise
comparisons of the medians and variance for both flexibility and redundancy (12 total). A nonparametric Wilcoxon matched-pairs signed rank test was used to compare the difference in
medians. Each test resulted in a rejection of the null hypothesis (at the 1% level), that both
distributions are the same. Therefore, the NDEA method creates different distributions for the
aggregate scores and DEA score for both flexibility and redundancy using the same data. Next, I

99

compare the equality of variances using the Levene’s test. Each pair is statistically significant at
the 1% level except the variance of the flexibility NDEA and DEA scores. Based on the results
of the Wilcoxon’s and Levene’s test, the distributions of each method to quantify the
performance of risk mitigation systems appear sensitive to the weighting schemes used. This
may result in different interpretations of performance and result in inaccurate risk mitigation
practices. For instance, a firm may use a simple aggregate performance metric and conclude that
their mitigation implementation is acceptable. However, without considering the cost and
interdependency among risk mitigation practices, they may be underperforming in reality. This
underestimation could even amplify the impact of a disruption by providing a false sense of
security.
Given the similarity in methods, comparing the two DEA based measures requires a slightly
more detailed investigation. Both provide a method for ascertaining efficiency, yet the NDEA
method allows for interdependencies to be explicitly modelled. This results in differences in
classification on the efficiency frontier. For instance, 21 firms have an efficiency score of 1 for
redundancy and 19 for flexibility using the NDEA score, while only 10 and 8 firms are on the
efficiency frontier based for the DEA model, respectively. Each of the firms on the efficiency
frontier from the DEA model are also on the efficiency frontier of the NDEA model. Of the firms
that are not on the flexibility efficiency frontier in the DEA model, several are among the lower
2/3rds of the sample if I rank order them according to their efficiency scores. This illustrates the
loss of information that occurs when interdependencies are ignored. Additionally, the required
inputs and corresponding outputs to achieve efficiency vary systematically between the two
methods; refer to Figure 9. Inputs are higher and outputs lower to achieve efficiency in the
NDEA model, reflecting how efficiency can be achieved by focusing on different aspects of the

100

supply chain. Therefore, the incremental impact of all these factors in a consistent direction does
have a significant impact on the efficiency evaluation. Given this, I make the case for
considering a holistic measure for supply chain efficiency by incorporating the interactions
between various stages via NDEA since traditional DEA underestimates the true nature of
efficiency resulting in possible over allocation of resources to improve efficiency and related
performance.
From a theoretical standpoint, NDEA permits the modeling of equifinality in achieving an
efficient risk mitigation system by considering the interdependencies between stages of the
supply chain. Specifically, firms can achieve the frontier by varying mix of investments in
supply, process, or demand stages of the supply chain. To illustrate the idea of equifinality, two
firms from the sample that both have efficiency scores of 1 are examined, Firm A and B. While
both firms had an efficiency score of 1 for both flexibility and redundancy, they achieved it in
different ways. This can be illustrated by comparing their raw survey scores to each other and the
median of the sample. I see that a firm with an efficiency score of 1 does not necessarily need to
have the lowest raw inputs or highest outputs in each domain of risk mitigation. For instance,
Firm B’s supply redundancy cost is above the sample median, but was able to achieve efficiency
compared to the sample by also having a high degree of supply redundancy. In comparison to
Firm A, Firm B has much lower focus on process and demand redundancy, but maintains
efficiency through low input cost of these strategies. For flexibility, Firm A has a weakness in
supply flexibility while Firm B is poor at converting process flexibility investments to process
flexibility. However, each of these firms has a high conversion of inputs to outputs in another
domain. Firm B has a higher than median level of supply flexibility with a relatively low
corresponding investment. Firm A has higher than median process flexibility and a low input

101

cost of process flexibility. Despite these differences, they are both efficient, displaying how
NDEA allows for equifinality in the performance evaluation of risk mitigation strategies. This is
important because it also highlights the fundamental premise of considering multiple stages of
the supply chain and its interdependencies. Thus, the proposed performance evaluation method
allows for the real-life flexibility that managers may have in achieving supply chain efficiency
by focusing on different facets of the supply chain.
Figure 9. Differences in Inputs (top) and Outputs (bottom) for NDEA vs. BCC efficient firms
7
6

Cost

5
4
3

NDEA Efficient

2

BBC Efficient

1
0

Flexibility

Redundancy

7

Mitigation

6

5
4
3

NDEA Efficient

2

BBC Efficient

1
0

Flexibility

Redundancy

102

Figure 10. Raw data comparison between firms

Furthermore, the results of the NDEA allows firms that are not on the efficiency frontier to
uncover which dimensions each firm may improve upon based on the items from the survey
instrument. Take for instance two firms that were not on the efficiency frontier, Firm X & Y.
Firm X has an overall score of Redundancy = 0.763 & Flexibility = 0.876 while Firm Y has an
overall score of Redundancy = 0.9158 and Flexibility = 0.7621. Not only does it show that Firm
X could improve on objectives such as the ‘number of product changeovers per manufacturing
facility made each month facility’ for process flexibility and ‘number of alternative/backup
suppliers’ for supply redundancy; NDEA provides projections for each input, displaying the
relative change required to reach efficiency. For instance, the projections show that Firm X must
increase the number of backup suppliers 15% or number of product changeovers by 68.8% if it
wants to become efficient. These two examples simply represent the questions on the survey
instrument, and the performance on any one of the other items may be insightful as well

103

depending on the firm. On the other hand, the NDEA results show that Firm Y could improve its
risk mitigation efficiency by reducing a few of its process redundancy costs, such as ‘produce the
same products at multiple manufacturing plants’ or ‘maintaining reserved equipment capacity for
adverse events.’ In fact, the projections even specify the required change of -2.5 % in time to
produce the same product at multiple plants and -30% in time to maintain reserved equipment
capacity. While these items were chosen to represent a higher order dimension of flexibility, they
are quantifiable activities that a firm could use if conducing a NDEA to evaluate their own risk
mitigation system. Items could also be adjusted to reflect the objections of a risk management
program tailored to a specific firm. The scores for each of the items are illustrated in Figure 10.
How do firms that are more efficient at mitigating risk perform relative to those that are less
efficient? To analyze the impact of using different measures to evaluate performance, I ran an
OLS regression on overall firm revenues as a measure of performance. Using revenues allows us
to account for the trade-off between the cost of risk mitigation and the impact they may have a
reducing risk. A firm that has experience significant disruptions because they did not have a
robust risk mitigation system in place would be expected to have lower revenue that a resilient
firm. In addition, the cost of investing in risk mitigation strategies may impact revenue
negatively. Since the amount of revenue will be higher for larger firms, I control for size and
estimate the impact of flexibility and redundancy on firm revenues for each performance
evaluation method, refer to Table 20. Size, as expected, is positive and significant in each model.
Flexibility quantified using NDEA is positive and significant, while redundancy is insignificant.
The CFA based measures are both moderately significant, with flexibility positive and
redundancy negative. The equal weight measures for both flexibility and redundancy are

104

insignificant. The DEA score of flexibility is moderately significant and positive while
redundancy is insignificant.
While admittedly limited, this analysis displays how differently managers may interpret the
same set of risk mitigation strategies simply based on how their performance is evaluated.
Revenue provides a financial measure of performance to compare the different quantifications
because the objective of any strategy is to improve the bottom line. From a financial performance
perspective, the regression results show that flexibility is a valuable strategy. Albeit, quantifying
flexibility using CFA or DEA resulted in only moderate significances between financial
performance and flexibility. In addition, using CFA may lead to conclude that redundancy
strategies reduce financial performance and should be limited. However, from an efficiency
perspective they do not have a significant relationship to financial performance and are likely
valuable for reducing financial risk if such a measure was available. Overall, the NDEA analysis
provides the most powerful model for managers to base decisions concerning resource allocation
of risk mitigation that may have an impact on financial performance.
3.4. Robustness Check
While the OLS results show discrimination between the measures, they assume a linear
relationship between the covariates and outcome. However, the DEA and NDEA scores are
highly skewed and past research has argued that when DEA scores are used as independent
variables they are more appropriate as categorical variable rather than continuous (Jacobs et al.,
2016; Swink et al., 2006). Thus, I expand upon the first analysis by classifying the firms into
three categories: Efficient, Tier 2, and inefficient. Firms were assigned to these groups by
quartile for the two linear aggregation measures and based on the efficiency frontier for the
NDEA measure following the method of Swink et al. (2006). Firms that were efficient, i.e. have

105

a score of one, were labeled as efficient and removed from the sample. The NDEA and DEA
optimizations were recalculated without the efficient firms and those firms who achieved the
efficiency frontier on this second analysis were label Tier 2 efficient. The rest of the firms were
assigned to the inefficient group. Table 21 present the results of this analysis. The results are
consistent for the NDEA measure; high flexibility is significantly positively associated with firm
revenue while redundancy is insignificant. Conducting an exploratory test between the impact of
efficient and tier 2 firms on performance is insignificant. The results from the continuous
analysis do not hold for the CFA measure when firms are categorized. The DEA score of
flexibility is again marginally significant and positive, but only for the efficient firms, not the
Tier 2 firms.
Table 20. OLS regression - Continuous

Revenue
Constant
Size
Flexibility
Redundancy
N
R2

Equal Weights
(E.W.)
Std.
β
Err.
+
3.531
2.010
**
0.666
0.089
0.473
0.382
-0.056
0.299
108
39.30%

CFA
β
5.067
0.634
0.229
-0.326
108

*
**
+
+

37.50%

DEA
Std.
Err.
2.091
0.092
0.117
0.173

β
4.499
0.648
2.296
-0.078
108
38.50%

106

**
**
*

Std.
Err.
1.314
0.090
1.002
0.956
36.45%

Std.
β
Err.
+
3.021
1.739
**
0.653
0.088
2.862 + 1.565
0.763
1.647

Table 21. OLS regression - Categorical
Equal Weights (E.W.)

β
5.381
0.634

β
5.244
0.672

0.808
0.887

0.738
0.747

1.118
0.725

0.735
0.677

1.127
1.352

Redundancy
Medium -0.412
High -0.351
N
108

0.691
0.764

-0.640
-0.857
108

0.720
0.702

-0.065
-0.323
108

R2

35.30%

**
**

**
**

36.20%

DEA

NDEA
Std.
Err.
1.243
0.095

Std. Err.
1.109
0.090

Revenue
Constant
Size
Flexibility
Medium
High

β
5.055
0.670

CFA

36.30%

**
**

*
*

Std.
Err.
1.146
0.092

β
4.594
0.647

**

0.521
0.656

1.136
1.488

+

0.729
0.571

-0.081
0.527
108

**

Std.
Err.
1.246
0.088
0.819
0.775
0.859
0.804

37.90%

4. Discussion
This study highlights the importance and difficulty in accounting for the cost of risk mitigation
and interdependency in risk mitigation strategies across the supply chain. This study is the first
empirical study to incorporate the cost and interdependency of supply chain risk mitigation
strategies when measuring and evaluating them. A NDEA method was proposed to address these
concerns and the financial implications were displayed in contrast to several other methods.
Regression analyses showed that NDEA provides a more powerful and comprehensive
perspective on risk mitigation performance across the supply chain. As an optimization model,
NDEA allows for the tradeoff between the cost and level of risk mitigation strategies to be
quantified explicitly. The study of risk mitigation using NDEA is particularly important because
it allows for equifinality in assessing a set of risk mitigation strategies. This follows several
scholars emphasis on the importance of accounting for specific exposures and capability of each
firm to mitigation risk (Chopra and Sodhi, 2004).
Flexibility and redundancy are both common methods for reducing risk occurring along the
supply chain. This study displayed the relative impact each have on financial performance.

107

Flexibility had a significant positive association with financial performance while redundancy
did not. When cost efficiency of the strategies was not considered, redundancies had a negative
relationship to financial performance. This result could be expected without considering cost
since they are an investment and thus effective only if a disruption occurs. However, when
resources are efficiently allocated toward redundant risk mitigation strategies, their negative
impact on financial performance disappears. Thus, it appears that flexibility has more than a risk
reducing impact, while redundancy is primarily an investment in case a disruption occurs. The
literature has argued for the advantages of each, and both are likely needed to obtain a resilient
supply chain. Although, flexibility appears to be the better of the two because of its positive
relationship to financial performance.
This study also stresses that risk mitigation practices should not be viewed in isolation since
the supply chain requires cohesiveness in order to ensure business continuity. A supply chain is
dependent on each element within it to operate smoothly. Comparing the DEA and NDEA scores
displays how the interdependent nature of risk mitigation strategies can significantly change the
perceived performance of them. Without linking the entire risk management program together,
interpretation of the program may be incorrect which could have harmful consequences for a
firm.
4.1. Limitations & Managerial Implications
This study was limited in several ways which presents opportunity for improvement in future
research. First, data was collected via a survey, limiting the scale of measurement and potentially
introducing measurement error. Secondary data across firms on levels and costs of specific risk
mitigation practices is not available as far as I am aware. However, surveys were deemed
appropriate since it was the only source of data available and are often used internally for

108

performance evaluation, for example using balance scorecards (Kaplan and Norton, 1996).
Secondly, NDEA analyses only provide relative scores so is sensitive to the addition of DMUs
that may be more efficient that those in the sample. While this is not a major concern empirically
in this study because I have a single sample, it may limit the generalizability of the results and
must be acknowledged if applying NDEA to other samples. Finally, using survey data as an
input is not ideal for DEA because the scales are ordinal. While I use a 7 point scale to make the
scale approximate continuous, ordinal linear optimization methods applied to NDEA may permit
improved optimization (Cook and Zhu, 2006). Unfortunately, there has not been a NDEA model
developed that permits the adaption of the method to ordinal data.
Evaluating the performance of strategic initiatives has been shown to be a key to achieving
financial performance (Ittner and Larcker, 2003). An important aspect of performance evaluation
is measurement and quantification of those measures since this data is what assists the strategic
decision making process. As most managers are aware, building risk mitigation systems is not
about just implanting risk mitigation strategies, but doing so efficiently. Thus, the measurement
of risk mitigation must assess both the degree of risk mitigation and the cost of implementing
them. Accounting for both of these dimensions can be accomplished using NDEA. In addition,
the entire risk mitigation program must be implemented holistically with the understanding that
the different components of operations and supply chain management are intimately linked. The
effectiveness and efficiency of a risk mitigation strategy in one area of supply chain risk
management may influence other areas. Acknowledging this interdependence can improve the
creation and adaptation of mitigating risks. I displayed a method of risk mitigation performance
evaluation that accounts both interdependences and investment costs through a novel NDEA
analysis.

109

NDEA is a versatile tool that allows managers to assess the efficiency of their risk mitigation
system in relation to other units, divisions, or firms while accounting for the equifinality in
achieving a resilient supply chain. While not necessary, weights can also be assigned to different
divisions in the optimization to allow for managers to provide additional input in to the model. I
displayed the NDEA method for quantifying the efficiency of two general types of risk
mitigation strategies: flexibility and redundancy. Flexibility and redundancy function in different
ways to reduce the impact of a disruption. According to the analysis, flexible risk mitigation
strategies may act as more than simply insurance since financial performance was shown to be
significantly related to them and not redundancy. Redundancies are necessary for most supply
chains, but due to the high cost of them are not associated with financial performance. I hope that
managers can take advantage of my statistical analysis and example of performance evaluation
using NDEA to improve their risk management programs and the resilience of their companies.

110

Chapter 5 - Conclusion
I examine the effectiveness of risk mitigation strategies in the three essays of this dissertation. In
the first, I show that exploration and exploitation alliance reduce the rate of failure at different
stages in new product development and that this rate is contingent on the diversity of alliance
partners. This essay emphasized the role external partners play in reducing risk, building to the
next essay of developing risk scales and exploring the differences culture plays in how risks are
mitigated and perceived. Finally, the last study applies a performance measurement system that
accounts for the interdependencies of risk mitigation strategies throughout the supply chain.
Overall, this dissertation investigates the role alliance partners have in new product development,
how to effectively leverage them to mitigate risk, and why it is imperative to understand the
various perspectives partners may have in managing risk.
Multiple methodologies were used in this dissertation. A survival analysis model was tested
empirically from a combination of secondary data sources. Results from this study may help
improve the structure and valuation of strategic alliances in the pharmaceutical industry. In the
second and third essays, a survey was developed and distributed to three countries. The second
essay compares survey data from three countries to explore how different risk mitigation
strategies are adopted and supply chain risks are perceived. Measurement invariance tests were
conducted using SEM, then country level differences were tested using GSEM. This study
enumerates differences in risk mitigation that may improve the coordination of risk management
between international firms. In the final essay, I used the data acquired from the USA to develop
a performance evaluation tool using NDEA. NDEA is a linear optimization technique that allows
for the interdependencies of risk mitigation and the cost of them to be accounted for. This study

111

may improve the performance measurement of risk mitigation strategies and the resource
allocation of them for manufacturing firms.
This dissertation integrates risk management at different levels of strategic decision making
and a broad perspective of the value chain to contribute to the literature on supply chain risk
management (SCRM). A review by Tang (2006) highlights the isolation of research across
different areas of SCRM. I attempt to integrate risk management research through a survey
instrument and performance measurement tool that account for upstream, internal and
downstream practices. Barnes (1984) contend that strategic decisions must account for various
biases. I observe that cultural biases are associated with differences in risk perception and
mitigation adoption. This result has implications for decision making at a tactical level and in
allocating resources between types of slack across the value chain. Lastly, alliance partnerships
are an important and necessary strategic decision (Das and Tang, 1998; Baum et al., 2000) that
have implications for risk management. I contribute to the literature by displaying in which
phase of development different forms of alliances are most effective and the types of alliance
partner characteristics are most strongly related to risk reduction in new product development.
Risk threatens supply chains in many different ways and not mitigating them effectively can
cost firms millions of dollars. Thus, it is vital that firms have a better understanding of the
methods they can use to mitigate risk and the importance of coordination with external supply
partners. This dissertation provides a step in improving our knowledge base on supply chain risk
mitigation, which will hopefully lead to more research in this domain and an improvement in
how risk is mitigated in practice.

112

APPENDIX

113

APPENDIX

Survey Instrument
1. Approximately how many employees work for the company? ____________________________
2. What were the company’s sales last year? (Millions) ____________________________total sales
3. How many product families does your firm produce? ___________________________________
4. How many manufacturing plants does your firm own?___________________________________
5. Do you have broad experience and knowledge of supply chain operations?_______Yes______No
6. What is your job title?____________________________________________________________
7. How many years have you been at your current position?___________________________(years)
8. How many years of total work experience do you have?____________________________(years)
9. How similar are your manufacturing plants in terms of:
a. Manufacturing technologies?
Very similar
1

2

3

Very different
4

5

6

7

b. Operational practices and protocols?
Very similar
1

2

3

Very different
4

5

6

7

10. How many regions are your firms manufacturing facilities located? ___# countries__# continents
11. How many outlets does your firm supply across all regions?______________________________
12. How similar are your customers in terms of:
a. Delivery speed requirements?
Very similar
1

2

3

Very different
4

5

6

7

b. Product quality requirements?
Very different
1

2

3

Very different
4

5

6

7

13. How many markets does your firm operate in? __________# countries___________# continents
14. How many suppliers are in your supply base (defined as suppliers that are actively managed through
contracts and the purchase of parts, materials and services)?_______________________
15. What percentage of your suppliers do you know actively manage supply chain risks?_________%

114

16. How similar are your suppliers in terms of:
a. Products that they produce?
Very similar
1

2

Very different
3

4

5

6

7

b. Operational practices and protocols?
Very different
1

2

Very different
3

4

5

6

7

17. How many locations does your firm source from? ________# countries___________# continents
Please indicate the average level of the following characteristics associated with your firm relative to
your competition (Circle a number for each factor with 1 standing for low and 7 standing for high)
Low
High
Extent to which your supplier lead-time can be expedited/changed
1 2 3 4 5 6 7
Extent to which your supplier short-term capacity can be increased or decreased
1 2 3 4 5 6 7
Extent of flexibility (options) within supplier contracts
1 2 3 4 5 6 7
Extent to which share of supplied components can be reallocated between suppliers
1 2 3 4 5 6 7
Range of products manufactured by your firms’ plants
1 2 3 4 5 6 7
Number of products that each manufacturing facility can produce (average)
1 2 3 4 5 6 7
Number of product changeovers per manufacturing facility made each month
1 2 3 4 5 6 7
Range of production volumes across which manufacturing can accommodate
1 2 3 4 5 6 7
Number of carriers used for each type of delivery mode, on average
1 2 3 4 5 6 7
Number of items handled by each distribution facility, on average
1 2 3 4 5 6 7
Number of items per order handled by each distribution facility on average
1 2 3 4 5 6 7
Number of customers supported by each distribution facility on average
1 2 3 4 5 6 7
Amount of inventory held by your manufacturing plants to manage supply risks
1 2 3 4 5 6 7
Priority in stocking inventory from suppliers
1 2 3 4 5 6 7
Amount of inventory held by your warehouses to meet demand requirements
1 2 3 4 5 6 7
Amount of capacity reserved by your suppliers in case of changes in requirements
1 2 3 4 5 6 7
Amount of capacity reserved in case of sudden changes in demand
1 2 3 4 5 6 7
Please answer this item with a one
1 2 3 4 5 6 7
Amount of space reserved in warehouses in case of sudden changes in demand
1 2 3 4 5 6 7
Number of alternative/backup suppliers selected on a per component basis
1 2 3 4 5 6 7
Number of manufacturing plants that can produce each product (average)
1 2 3 4 5 6 7
Extent to which demand at a specific location can be reassigned to a different location 1 2 3 4 5 6 7
Number of worldwide storage/distribution facilities
1 2 3 4 5 6 7
Suppliers’ ability to produce components not currently sourced from them
1 2 3 4 5 6 7
Amount of equipment stored in each plant to produce components not currently
1 2 3 4 5 6 7
produced in that plant
Please indicate the average level of cost and time associated with engaging in the following activities in
your business unit relative to your competition (Circle a number for each factor with 1 standing for low
and 7 standing for high)
Cost
Time
Influence supplier’s to increase or decrease short-term capacity
1234567 1234567
Change volume allocation among existing suppliers on a global basis 1 2 3 4 5 6 7 1 2 3 4 5 6 7
Change quantity of supplier’s order
1234567 1234567
Change delivery times of order placed with suppliers
1234567 1234567

115

Change manufacturing volume capacity when necessary
Accommodate changes in manufacturing mix as required
Change between producing different products
Adjust manufacturing processes
Fills customer orders from individual facilities
Reallocate logistics services to meet changing customer requirements
Adjust global delivery capacity
Reallocate finished product distribution among global storage facilities

Develop strong relationships with suppliers
Stock inventory in case of supply chain risk
Manage multiple sources of supply
Produce the same products at multiple manufacturing plants
Maintaining reserved equipment capacity for adverse events
Improve forecast accuracy
Reserve space in warehouses for inventory volatility
Develop options contracts with suppliers
Duplicate inventory in different locations
Adjust worldwide storage capacity
Change supplier of a component to another supplier
Set up equipment to produce components for the first time in a plant

1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567

1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567
1234567

Environmental Uncertainty
To what extent has your firm in the past year experienced a negative impact in supply chain management
due to… (Circle a number for each factor with 1 standing for “not at all” and 7 standing for “to a very
large extent”)
Not at all
Large extent
The suppliers consistently meet our requirements
1 2 3 4 5 6 7
The suppliers produce material with consistent quality
1 2 3 4 5 6 7
Our master production schedule has a high percentage of variation
1 2 3 4 5 6 7
Our demand fluctuates drastically from week to week
1 2 3 4 5 6 7
Our supply requirements vary drastically from week to week
1 2 3 4 5 6 7
Our industry is characterized by rapidly changing technology
1 2 3 4 5 6 7
If I don’t keep up with the changes in technology,
1 2 3 4 5 6 7
it will be difficult for us to remain competitive
The rate of process obsolescence is high in our industry
1 2 3 4 5 6 7
The production technology changes frequently and sufficiently
1 2 3 4 5 6 7
Supply Chain Risk
In this section, please rate the likelihood and severity that the various risk categories pose to your supply
chain for one of your most important products or product lines (Circle a number for each risk category
with 1 standing for “Low” and 7 standing for “High”)
Unanticipated or very volatile customer demand
Likelihood (probability of occurrence)
Severity (size of impact)
Sudden default of a supplier (e.g., due to bankruptcy)

116

Low
1 2
1 2

High
3 4
3 4

5 6
5 6

7
7

Likelihood (probability of occurrence)
1 2 3 4 5 6 7
Severity (size of impact)
1 2 3 4 5 6 7
Poor logistics performance of suppliers (delivery dependability, order fill capacity)
Likelihood (probability of occurrence)
1 2 3 4 5 6 7
Severity (size of impact)
1 2 3 4 5 6 7
Supplier quality problems
Likelihood (probability of occurrence)
1 2 3 4 5 6 7
Severity (size of impact)
1 2 3 4 5 6 7
Poor logistics performance of logistics service providers
Likelihood (probability of occurrence)
1 2 3 4 5 6 7
Severity (size of impact)
1 2 3 4 5 6 7
Insufficient or distorted information from your customers’ quantities about orders or demand
Likelihood (probability of occurrence)
1 2 3 4 5 6 7
Severity (size of impact)
1 2 3 4 5 6 7
Capacity fluctuation or shortages on the supply markets
Likelihood (probability of occurrence)
1 2 3 4 5 6 7
Severity (size of impact)
1 2 3 4 5 6 7
Political instability, war, civil unrest or other socio-political crises
Likelihood (probability of occurrence)
1 2 3 4 5 6 7
Severity (size of impact)
1 2 3 4 5 6 7
Disease or epidemics (e.g., SARS, Foot and Mouth Disease)
Likelihood (probability of occurrence)
1 2 3 4 5 6 7
Severity (size of impact)
1 2 3 4 5 6 7
Natural disasters (e.g., earthquake, flooding, extreme climate, tsunami)
Likelihood (probability of occurrence)
1 2 3 4 5 6 7
Severity (size of impact)
1 2 3 4 5 6 7
International terror attacks (e.g., 2005 London or 2004 Madrid terror attacks)
Likelihood (probability of occurrence)
1 2 3 4 5 6 7
Severity (size of impact)
1 2 3 4 5 6 7
Downtime or loss of own production capacity due to local disruptions (e.g., fire, industrial accidents, etc.)
Likelihood (probability of occurrence)
1 2 3 4 5 6 7
Severity (size of impact)
1 2 3 4 5 6 7
Perturbation or breakdown of internal IT infrastructure (e.g., caused by computer viruses, software bugs)
Likelihood (probability of occurrence)
1 2 3 4 5 6 7
Severity (size of impact)
1 2 3 4 5 6 7
Loss of own production capacity due to technical reasons (e.g., machine deterioration, poor quality)
Likelihood (probability of occurrence)
1 2 3 4 5 6 7
Severity (size of impact)
1 2 3 4 5 6 7
Perturbation or breakdown of external IT infrastructure
Likelihood (probability of occurrence)
1 2 3 4 5 6 7
Severity (size of impact)
1 2 3 4 5 6 7
Technology
Please rate how frequently you use the following Information Technologies (IT): (Circle a number for
each technology; 1=Never, 2=Once in a year, 3=once in a quarter, 4=once in a month, 5=once in a
week, 6=once in a day, 7=many times a day)
Never
Daily
Computer aided design manufacturing (CAD)
1 2 3 4 5 6 7
Robotics-automated systems
1 2 3 4 5 6 7
Computer aided manufacturing (CAM)
1 2 3 4 5 6 7
3D images (parts imaging for quality control)
1 2 3 4 5 6 7
Expert production monitoring system
1 2 3 4 5 6 7

117

Material resource planning systems (MRP I or MRP II) to ensure
1 2 3 4 5 6 7
accurate ordering of materials
Electronic data interchange (EDI) with suppliers
1 2 3 4 5 6 7
IT applications (such as desktop sharing) for simultaneously
1 2 3 4 5 6 7
working in real time with external suppliers
IT applications for collaborative planning and forecasting of demand
1 2 3 4 5 6 7
with external suppliers (CFAR).
IT applications to synchronize production plans across the external suppliers
1 2 3 4 5 6 7
IT applications to share production data (e.g. inventory levels, and
1 2 3 4 5 6 7
production schedules) with vendor managing replenishment systems
IT applications (such as analytics or data mining) for creating or
changing demand plans
IT applications (e.g. Intranet) to acquire marketing information from
customers relevant for creating or changing demand plans
Bar Coding technologies to acquire customer sales data
Radio Frequency Identification Tag (RFID) applications to
acquire customer sales data
Electronic data interchange (EDI) with clients

1 2 3 4 5 6 7
1 2 3 4 5 6 7
1 2 3 4 5 6 7
1 2 3 4 5 6 7
1 2 3 4 5 6 7

Financial Performance
For each of the items listed below, how does your company compare with its competitors (Circle a
number)
Far Worse
Far Better
Than Competitors
Than Competitors
Market share
1
2
3
4
5
6
7
Sales growth
1
2
3
4
5
6
7
Return on Sales (ROS)
1
2
3
4
5
6
7
Return on Investments (ROI)
1
2
3
4
5
6
7
Growth in ROI
1
2
3
4
5
6
7
1. Where is your firm headquartered (Country & City)? ___________________________________
2. What is the primary industry in which your firm operates? _______________________________
3. What year was your firm founded? _________________________________________________
4. How long has your firm used a systematic supply chain risk program? ___None or ______#years

118

REFERENCES

119

REFERENCES

Adams, C. P., Brantner, V. V., 2006. Estimating The Cost Of New Drug Development: Is It
Really $802 Million? Health Affairs. 25, 420-428
Ahuja, G. 2000. Collaboration networks, structural holes, and innovation: A longitudinal study.
Administrative Science Quarterly. 45, 425-455
Albert, R., BarabĂĄsi, A. L., 2002. Statistical mechanics of complex networks. Reviews of
Modern Physics. 74, 47
Álvarez, G., María J., Gonzålez de la Fe, P., 1999. Strategic alliances, organizational learning
and new product development: the cases of Rover and Seat. R&D Management. 29, 423426
Amason, A.C., 1996. Distinguishing the effects of functional and dysfunctional conflict on
strategic decision making: Resolving a paradox for top management teams. Academy of
Management Journal. 39, 123-148
Ambulkar, S., Blackhurst, J., Grawe, S., 2015. Firm's resilience to supply chain disruptions:
Scale development and empirical examination. Journal of Operations Management. 33,
111-122
Arrowsmith, J., 2011. Trial watch: phase II failures: 2008–2010. Nature Reviews Drug
Discovery 10, 328-329
Arrowsmith, J., Miller, P., 2013. Trial watch: phase II and phase III attrition rates 2011-2012.
Nature Reviews Drug Discovery. 12, 569-569
Ashkanasy, N., Gupta, V., Mayeld, M.S., Trevor-Roberts, E., 2004. Future Orientation. 282342
Avittathur, B., Swamidass, P., 2007. Matching plant flexibility and supplier flexibility: Lessons
from small suppliers of US manufacturing plants in India. Journal of Operations
Management. 25, 717-735
Banker, R.D., Charnes, A., Cooper, W.W., 1984. Some models for estimating technical and scale
inefficiencies in data envelopment analysis. Management Science. 30, 1078-1092
Bansal, P., 2003. From issues to actions: The importance of individual concerns and
organizational values in responding to natural environmental issues. Organization
Science. 14, 510-527
Barnes, J. H. Jr. 1984. Cognitive biases and their impact on strategic planning', Strategic
Management Journal. 5, 129-137

120

Baum, J. A., Calabrese, T., Silverman, B. S., 2000. Don't go it alone: Alliance network
composition and startups' performance in Canadian biotechnology. Strategic
Management Journal. 21, 267- 294
Baum, J. A., Li, Xiao, S., Usher, J. M., 2000. Making the next move: How experiential and
vicarious learning shape the locations of chains' acquisitions. Administrative Science
Quarterly. 45, 766-801
Benner, M. J., Tushman, M., 2002. Process management and technological innovation: A
longitudinal study of the photography and paint industries. Administrative Science
Quarterly. 47, 676-707
Beroggi, G.E.G., Wallace, W.A., 1995. Operational control of the transportation of hazardous
materials: An assessment of alternative decision models. Management Science. 41, 19621977
Betts, J.M., Johnston, R.B., 2005. Determining the optimal constrained multi-item (Q, r)
inventory policy by maximising risk-adjusted profit. IMA Journal of Management
Mathematics. 16
Bititci, U.S., Turner, T., Begemann, C., 1997, Integrated performance measurement systems: a
development guide. International Journal of Operations & Production Management. 17,
522-34
Bititci, U.S., Turner, T., Begemann, C., 2000. Dynamics of performance measurement systems.
International Journal of Operations & Production Management. 20, 692-704
Blackhurst, J., Dunn, K.S., Craighead, C.W., 2011. An Empirically Derived Framework of
Global Supply Resiliency. Journal of Business Logistics. 32, 374-391
Blome, C., Schoenherr, T., Rexhausen, D., 2013. Antecedents and enablers of supply chain
agility and its effect on performance: a dynamic capabilities perspective. International
Journal of Production Research. 51, 1295-1318
Bode, C., Wagner, S.M., 2015. Structural drivers of upstream supply chain complexity and the
frequency of supply chain disruptions. Journal of Operations Management. 36, 215-228
Bourland, K.E., Yano, C.A., 1994. The strategic use of capacity slack in the economic lot
scheduling problem with random demand. Management Science. 40, 1690-1704
Brandon-Jones, E., Squire, B., Autry, C. W., Petersen, K. J., 2014. A contingent resource-based
perspective of supply chain resilience and robustness. Journal of Supply Chain
Management. 50, 55-73
Braunscheidel, M. J., Suresh, N. C., 2009. The organizational antecedents of a firm’s supply
chain agility for risk mitigation and response. Journal of Operations Management. 27,
119-140
Buonansegna, E., Salomo, S., Maier, A.M., Li‐Ying, J., 2014. Pharmaceutical new product
development: why do clinical trials fail? R&D Management. 44, 189-202

121

Burton, T. M., 2004. Flop factor: By learning from failures, Lilly keeps drug pipeline full. Wall
Street Journal.
Buttle, F., Burton, J., 2002. Does service failure influence customer loyalty? Journal of
Consumer Behaviour. 1, 217-227
Byrne, B. M., Shavelson, R.J., Muthen, B., 1989. Testing for the Equivalence of Factor
Covariance and Mean Structures: The Issue of Partial Measurement Invariance.
Psychological Bulletin. 105, 456- 466
Ceryno, P.S., Scavarda, L.F., Klingebiel, K., 2015. Supply chain risk: empirical research in the
automotive industry. Journal of Risk Research. 18, 1145-1164
Charnes, A., Cooper, W.W., Rhodes, E., 1978. Measuring the efficiency of decision making
units. European Journal of Operational Research. 2, 429-444
Chen, H., Chen, J., Chen, Y., 2006. A Coordination Mechanism for a Supply Chain with
Demand Information Updating. International Journal of Production Economics. 103, 347
– 361
Chern, M.S., 1992. On the computational complexity of reliability redundancy allocation in a
series system. Operations Research Letters. 11, 309-315
Chopra, S., Sodhi, M.S., 2004. Managing risk to avoid supply-chain breakdown. Mit Sloan
Management Review. 46, 53-67
Churchill Jr, G.A., 1979. A paradigm for developing better measures of marketing constructs.
Journal of Marketing Research. 64-73
Cleves, M. A., Gould, W. W., Yulia V. M., 2016. An Introduction to Survival Analysis Using
Stata. College Station, TX: Stata
Cohen, M.A., Huchzermeier, A., 1999. Global supply chain management: a survey of research
and applications, Quantitative models for supply chain management. Springer, 669-702
Comrey, A. L., Lee, H. B., 1992. A first course in factor analysis. 2nd ed. Hillside, NJ: Erlbaum.
Cook, W.D., Zhu, J., 2006. Rank order data in DEA: A general framework. European Journal of
Operational Research. 174, 1021-1038
Cooper, W.W., Seiford, L.M., Tone, K., 2007. Data Envelopment Analysis: A Comprehensive
Text with Models, Applications, References and DEA-SolverSoftware, 2nd ed., Springer
Das, T. K., & Teng, B. S., 1998. Resource and risk management in the strategic alliance making
process. Journal of Management. 24, 21-42
Datta, S., Granger, C. W. J., Barari, M., Gibbs, T., 2007. Management of Supply Chain: An
Alternative Modelling Technique for Forecasting. Journal of the Operational Research
Society. 58, 1459 – 1469

122

Dell'Era, C., Verganti, R., 2010. Collaborative strategies in design-intensive industries:
knowledge diversity and innovation. Long Range Planning. 43, 123-141
Desai, V.M., 2008. Constrained growth: How experience, legitimacy, and age influence risk
taking in organizations. Organization Science. 19, 594-608
Dilts, D. M., 2013. A “Three-Plus-One” Evaluation Model for Clinical Research Management.
Evaluation & the Health Professions. 36, 464-477
Dilts, D., Cheng, S., Crites, J., Sandler, A., Doroshow, J., 2010. Phase III clinical trial
development: a process of chutes and ladders. Clinical Cancer Research. 16, 5381-5389
DiMasi, J. A., 2001. Risks in new drug development: approval success rates for investigational
drugs. Clinical Pharmacology & Therapeutics. 69, 297-307
DiMasi, J. A., 2013. Innovating by Developing New Uses of Already-Approved Drugs: Trends
in the Marketing Approval of Supplemental Indications. Clinical Therapeutics. 35, 808818
DiMasi, J. A., 2014. Pharmaceutical R&D Performance by Firm Size: Approval Success Rates
and Economic Returns. American Journal of Therapeutics. 21, 26-34
DiMasi, J. A., Faden, L.B., 2011. Competitiveness in follow-on drug R&D: a race or imitation?
Nature Reviews Drug Discovery. 10, 23-27
DiMasi, J. A., Feldman, L., Seckler, A., Wilson, A., 2010. Trends in Risks Associated With New
Drug Development: Success Rates for Investigational Drugs. Clinical Pharmacology &
Therapeutics. 87, 272-277
DiMasi, J. A., Kim, J., Getz, K.A., 2014. The Impact of Collaborative and Risk-Sharing
Innovation Approaches on Clinical and Regulatory Cycle Times. Therapeutic Innovation
& Regulatory Science. 48, 482-487
Dimitrov, D.M., 2006. Comparing groups on latent variables: A structural equation modeling
approach. Work. 26, 429-436
DiStefano, C., Zhu, M., Mindrila, D., 2009. Understanding and using factor scores:
Considerations for the applied researcher. Practical Assessment, Research & Evaluation.
14, 1-11
Doctor, R., Newton, D.P., Pearson, A., 2001. Managing uncertainty in research and
development. Technovation. 21, 79-90
Doty, D.H., Glick, W.H., 1998. Common methods bias: does common methods variance really
bias results? Organizational Research Methods. 1, 374-406
Dutta, S., & Weiss, A. M., 1997. The relationship between a firm's level of technological
innovativeness and its pattern of partnership agreements. Management Science. 43, 343356

123

Ellis, S.C., Henry, R.M., Shockley, J., 2010. Buyer perceptions of supply disruption risk: a
behavioral view and empirical assessment. Journal of Operations Management. 28, 34-46
Eppen, G.D., Iyer, A.V., 1997. Backup agreements in fashion buying—the value of upstream
flexibility. Management Science. 43, 1469-1484
Estampe, D., Lamouri, S., Paris, J.-L., Brahim-Djelloul, S., 2013. A framework for analysing
supply chain performance evaluation models. International Journal of Production
Economics. 142, 247-258
Faems, D., Van Looy, B., Debackere, K., 2005. Interorganizational collaboration and innovation:
toward a portfolio approach. Journal of Product Innovation Management. 22. 238-250
Fang, E., 2011. The effect of strategic alliance knowledge complementarity on new product
innovativeness in China. Organization Science. 22, 158
Fortuin, L., 1988. Performance indicators—why, where and how? European Journal of
Operational Research 34, 1-9
George, G., 2005. Slack resources and the performance of privately held firms. Academy of
Management Journal. 48, 661-676
Ghare, P. M., Taylor, R. E., 1969 Optimal redundancy for reliability in series systems.
Operations Research. 17, 838-847
Giaglis, G.M., Minis, I., Tatarakis, A., Zeimpekis, V., 2004. Minimizing logistics risk through
real-time vehicle routing and mobile technologies: Research to date and future trends.
International Journal of Physical Distribution & Logistics Management. 34, 749-764
Gino, F., Pisano, G., 2008. Toward a theory of behavioral operations. Manufacturing & Service
Operations Management. 10, 676-691
Gregorich, S. E., 2006. Do self-report instruments allow meaningful comparisons across diverse
population groups? Testing measurement invariance using the confirmatory factor
analysis framework. Medical care. 44, S78
Gresov, C., Drazin, R., 1997. Equifinality: Functional equivalence in organization design.
Academy of Management Review. 22, 403-428
Grotsch, V.M., Blome, C., Schleper, M.C., 2013. Antecedents of proactive supply chain risk
management a contingency theory perspective. International Journal of Production
Research. 51, 2842-2867
Gunasekaran, A., Patel, C., Tirtiroglu, E., 2001. Performance measures and metrics in a supply
chain environment. International Journal of Operations & Production Management. 21,
71-87
Gunasekaran, A., Subramanian, N., Rahman, S., 2015. Supply chain resilience: role of
complexities and strategies. International Journal of Production Research. 53, 6809-6819

124

Guo, Z., Fang, F., Whinston, A. B., 2006. Supply Chain Information Sharing in a Macro
Prediction Market. Decision Support Systems. 42, 1944 – 1958
Gupta, A.K., Smith, K.G., Shalley, C.E., 2006. The interplay between exploration and
exploitation. Academy of Management Journal. 49, 693-706
Ha, A.Y., 1997. Inventory rationing in a make-to-stock production system with several demand
classes and lost sales. Management Science. 43, 1093-1103
Haeussler, C., Patzelt, H., Zahra, S.A., 2012. Strategic alliances and product development in high
technology new firms: The moderating effect of technological capabilities. Journal of
Business Venturing. 27, 217-233
Haimes, Y. Y., 1998. Risk Modeling, Assessment, and Management. John Wiley & Sons, 2nd
ed. Print.
Hanagal, D.D., 2006. Bivariate Weibull regression model based on censored samples. Statistical
Papers. 47, 137-147
Heckman, J. J., 1979. Statistical models for discrete panel data. Department of Economics and
Graduate School of Business, University of Chicago.
Hendricks, K.B., 1992. The output processes of serial production lines of exponential machines
with finite buffers. Operations Research. 40, 1139-1147
Hendricks, K.B., Singhal, V.R., Zhang, R.R., 2009. The effect of operational slack,
diversification, and vertical relatedness on the stock market reaction to supply chain
disruptions. Journal of Operations Management. 27, 233-246
Henri, J.-F., 2006. Organizational culture and performance measurement systems. Accounting,
organizations and society. 31, 77-103
Ho, W., Zheng, T., Yildiz, H., Talluri, S., 2015. Supply chain risk management: a literature
review. International Journal of Production Research. 53, 5031-5069
Hoang, H., Rothaermel, F.T., 2010. Leveraging internal and external experience: exploration,
exploitation, and R&D project performance. Strategic Management Journal. 31, 734-758
Hoffmann, P., Schiele, H., Krabbendam, K., 2013. Uncertainty, supply risk management and
their impact on performance. Journal of Purchasing and Supply Management. 19, 199211
Hofstede, G., 1980a. Culture's consequences: International differences in work-related values.
Sage Publications, Inc
Hofstede, G., 1980b. Culture and organizations. International Studies of Management &
Organization, 15-41
Hofstede, G., 1984. The cultural relativity of the quality of life concept. Academy of
Management Review. 9, 389-398

125

Hofstede, G., 2001. Culture’s Consequences: Comparing Values, Behaviors, Institutions, and
Organizations Across Nations. 2nd Ed., Sage Publications, Inc.
Horn, J. L., McArdle, J., Mason, R., 1983. When Is Invariance Not Invariant: A Practical
Scientist’s Look at the Ethereal Concept of Factor Invariance. The Southern
Psychologist. 1, 179–188.
Huang, Y.-F., Chen, C.-J., 2010. The impact of technological diversity and organizational slack
on innovation. Technovation. 30, 420-428
Huchzermeier, A., Cohen, M.A., 1996. Valuing operational flexibility under exchange rate risk.
Operations Research. 44, 100-113
Inman, R.A., Mehra, S., 1993. Financial justification of JIT implementation. International
Journal of Operations & Production Management. 13, 32-39
Ittner, C.D., Larcker, D.F., 2003. Coming up short on nonfinancial performance measurement.
Harvard Business Review. 81, 88-99
Jacobs, B.W., Kraude, R., Narayanan, S., 2016. Operational productivity, corporate social
performance, financial performance, and risk in manufacturing firms. Production and
Operations Management. 25, 2065-2085
Johnson, N., Elliott, D., Drake, P., 2013. Exploring the role of social capital in facilitating supply
chain resilience. Supply Chain Management: An International Journal. 18, 324-336
Kalaignanam, K., Shankar, V., Varadarajan, R., 2007. Asymmetric new product development
alliances: win-win or win-lose partnerships? Management Science. 53, 357-374
Kandemir, D., Yaprak, A., Cavusgil, S.T., 2006. Alliance orientation: conceptualization,
measurement, and impact on market performance. Journal of the Academy of Marketing
Science. 34, 324-340
Kaplan, R.S., Mikes, A., 2012. Managing Risks: A New Framework. Harvard Business Review.
90, 48-60
Kaplan, R.S., Norton, D.P., 1996. The balanced scorecard: translating strategy into action.
Harvard Business Press
Katila, R., Ahuja, G., 2002. Something old, something new: A longitudinal study of search
behavior and new product introduction. Academy of Management Journal. 45, 1183-1194
Kennedy, T., 1997. Managing the drug discovery/development interface. Drug Discovery Today,
2, 436-444
Kern, D., Moser, R., Hartmann, E., Moder, M., 2012. Supply risk management: model
development and empirical analysis. International Journal of Physical Distribution &
Logistics Management. 42, 60-82
Khanna, I., 2012. Drug discovery in pharmaceutical industry: productivity challenges and trends.
Drug Discovery Today. 17, 1088-1102
126

Kim, W. S., 2013. A Supply Chain Contract with Flexibility as a Risk-sharing Mechanism for
Demand Forecasting. International Journal of Systems Science, 44, 1134 – 1149
Kleczyk, E., 2008. Risk management in the development of new products in the pharmaceutical
industry. African Journal of Business Management. 2, 186
Kleindorfer, P.R., Saad, G.H., 2005. Managing disruption risks in supply chains. Production and
Operations Management. 14, 53-68
Koste, L.L., Malhotra, M.K., Sharma, S., 2004. Measuring dimensions of manufacturing
flexibility. Journal of Operations Management. 22, 171-196
Kotabe, M., Scott Swan, K., 1995. The role of strategic alliances in high‐technology new product
development. Strategic Management Journal. 16, 621-636
Koufteros, X., Verghese, A., Lucianetti, L., 2014. The effect of performance measurement
systems on firm performance: A cross-sectional and a longitudinal study. Journal of
Operations Management. 32, 313-336
Koza, M.P., Lewin, A.Y., 1998. The co-evolution of strategic alliances. Organization science. 9,
255- 264
Krishnan, V., Ulrich, K. T., 2001. Product development decisions: A review of the literature.
Management science. 47, 1-21
Kull, T.J., Wacker, J.G., 2010. Quality management effectiveness in Asia: The influence of
culture. Journal of Operations Management. 28, 223-239
Lado, A, A., Boyd, N. G., Hanlon, S. C., 1997. Competition, cooperation, and the search for
economic rents: a syncretic model. Academy of Management Review. 22, 110-141
Lambert, D. M., Pohlen, T. L., 2001, Supply Chain Metrics. The International Journal of
Logistics Management. 12, 1-19
Lane, P.J., Lubatkin, M., 1998. Relative absorptive capacity and interorganizational learning.
Strategic Management Journal. 19, 461
Lavie, D., Rosenkopf, L., 2006. Balancing exploration and exploitation in alliance formation.
Academy of Management Journal. 49, 797-818
Lee, C.L., Wu, H.C., 2015. How do slack resources affect the relationship between R&D
expenditures and firm performance? R&D Management.
Lee, H.L., Padmanabhan, V., Whang, S., 2004. Information distortion in a supply chain: the
bullwhip effect. Management Science. 50, 1875-1886
Lee, S.H., Makhija, M., 2009. Flexibility in internationalization: is it valuable during an
economic crisis? Strategic Management Journal. 30, 537-555
Levinthal, D.A., 1997. Adaptation on rugged landscapes. Management Science. 43, 934-950

127

Levitt, B., March, J. G., 1988. Organizational learning. Annual Review of Sociology. 319- 340
Lorenzoni, G., Lipparini, A., 1999. The leveraging of interfirm relationships as a distinctive
organizational capability: a longitudinal study. Strategic Management Journal. 20, 317338
Lytle, A., Brett, J., Barsness, Z., Tinsley, C., Janssens, M., 1995. A paradigm for confirmatory
cross-cultural research in organizational-behavior. Research in organizational behavior:
an annual series of analytical essays and critical reviews. 17, 167-214
March, J. G., 1991. Exploration and exploitation in organizational learning. Organization
Science. 2, 71-87
Melnyk, S.A., Stewart, D.M., Swink, M., 2004. Metrics and performance measurement in
operations management: dealing with the metrics maze. Journal of Operations
Management. 22, 209-217
Meredith, J., 1993. Theory building through conceptual methods. International Journal of
Operations & Production Management. 13, 3–11
Meredith, W., 1993. Measurement Invariance, Factor Analysis, and Factorial Invariance.
Psychometrika. 58, 525–543
Miller, K. D, Zhao, M., Calantone, R. J., 2006. Adding interpersonal learning and tacit
knowledge to March's exploration-exploitation model. Academy of Management Journal.
49, 709-722
Mishra, S., Modi, S.B., Animesh, A., 2013. The relationship between information technology
capability, inventory efficiency, and shareholder wealth: A firm-level empirical analysis.
Journal of Operations Management. 31, 298-312
Nagaoka, S., Motohashi, K., Goto, A., 2010. Patent statistics as an innovation indicator.
Handbook of the Economics of Innovation. 2, 1083-1127
Naor, M., Linderman, K., Schroeder, R., 2010. The globalization of operations in Eastern and
Western countries: Unpacking the relationship between national and organizational
culture and its impact on manufacturing performance. Journal of Operations
Management. 28, 194-205
Narasimhan, R., Talluri, S., 2009. Perspectives on risk management in supply chains. Journal of
Operations Management. 27, 114-118
Neely, A., 2005. The evolution of performance measurement research - Developments in the last
decade and a research agenda for the next. International Journal of Operations &
Production Management. 25, 1264-1277
Newman, K. L. Nollen, S. D., 1996. Culture and Congruence: The Fit Between Management
Practices and National Culture. Journal of International Business Studies. 27, 753-779

128

Nielsen, B.B., & Gudergan, S., 2012. Exploration and exploitation fit and performance in
international strategic alliances. International Business Review. 21, 558-574
Nohria, N., Gulati, R., 1996. Is slack good or bad for innovation? Academy of Management
Journal. 39, 1245-1264
Novaes, A., 2000. Global Supply Chain Flexibility Under Risk: the Applichem Case. Building
Competencies for International Manufacturing. 209-217
O'Reilly, J., 2014. Risk mitigation: supply chain safety net. Inbound Logistics. 34
Oxley, J. E, Sampson, R. C., 2004. The scope and governance of international R&D alliances.
Strategic Management Journal. 25, 723-749
Park, S.H., Chen, R.R., Gallagher, S., 2002. Firm resources as moderators of the relationship
between market growth and strategic alliances in semiconductor start-ups. Academy of
Management Journal. 45, 527-545
Patel, P.C., Jayaram, J., 2014. The antecedents and consequences of product variety in new
ventures: An empirical study. Journal of Operations Management. 32, 34-50
Patel, P.C., Terjesen, S., Li, D., 2012. Enhancing effects of manufacturing flexibility through
operational absorptive capacity and operational ambidexterity. Journal of Operations
Management. 30, 201-220
Petersen, K. J, Handfield, R. B, Ragatz, G. L., 2005. Supplier integration into new product
development: coordinating product, process and supply chain design. Journal of
Operations Management. 23, 371-388
Pettit, T.J., Fiksel, J., Croxton, K.L., 2010. Ensuring supply chain resilience: development of a
conceptual framework. Journal of Business Logistics. 31, 1-21
Phelps, C. C., 2010. A longitudinal study of the influence of alliance network structure and
composition on firm exploratory innovation. Academy of Management Journal. 53, 890913
Pisano, G. P., 1996. Learning-before-doing in the development of new process technology.
Research Policy. 25, 1097-1119
Postrel, S., 2002. Islands of shared knowledge: Specialization and mutual understanding in
problem- solving teams. Organization Science. 13, 303-320
Power, D., Schoenherr, T., Samson, D., 2010. The cultural characteristic of
individualism/collectivism: A comparative study of implications for investment in
operations between emerging Asian and industrialized Western countries. Journal of
Operations Management. 28, 206-222
Prater, E., Ghosh, S., 2006. A comparative model of firm size and the global operational
dynamics of US firms in Europe. Journal of Operations Management. 24, 511-529

129

Ramamoorthy, N., Kulkarni, S.P., Gupta, A., Flood, P.C., 2007. Individualism–collectivism
orientation and employee attitudes: A comparison of employees from the hightechnology sector in India and Ireland. Journal of International Management. 13, 187-203
Realo, A., Allik, J., Greenfield, B., 2008. Radius of trust: Social capital in relation to familism
and institutional collectivism. Journal of Cross-Cultural Psychology. 39, 447-462
Rettig, R. A., 2000. The industrialization of clinical research. Health Affairs.19, 129-146
Richardson, H.A., Simmering, M.J., Sturman, M.C., 2009. A tale of three perspectives:
Examining post hoc statistical techniques for detection and correction of common method
variance. Organizational Research Methods.
Ritchie, B., Brindley, C., 2007. An emergent framework for supply chain risk management and
performance measurement. Journal of the Operational Research Society. 58, 1398-1411
Rothaermel, F. T., Deeds, D. L., 2004. Exploration and exploitation alliances in biotechnology:
A system of new product development. Strategic Management Journal. 25, 201-221
Rothaermel, F., Hess, A., 2010. Innovation strategies combined. MIT Sloan Management
Review. 51, 13
Sampson, R. C. 2007. R&D alliances and firm performance: The impact of technological
diversity and alliance organization on innovation. Academy of Management Journal. 50,
364-386
Schachter, A. D., Ramoni, M. F., 2007. Clinical forecasting in drug development. Nature
Reviews Drug Discovery. 6, 107-108
Schein, E. H., 2004.Organizational Culture and Leadership. 3rd ed. San Francisco: Jossey-Bass.
Scholten, K., Sharkey Scott, P., Fynes, B., 2014. Mitigation processes–antecedents for building
supply chain resilience. Supply Chain Management: An International Journal. 19, 211228
Shan, W., Walker, G., Kogut, B., 1994. Interfirm cooperation and startup innovation in the
biotechnology industry. Strategic Management Journal. 15, 387-394
Sharfman, M.P., Wolf, G., Chase, R.B., Tansik, D.A., 1988. Antecedents of organizational slack.
Academy of Management Review. 13, 601-614
Sheffi, Y., 2001. Supply chain management under the threat of international terrorism. The
International Journal of Logistics Management. 12, 1-11
Sheffi, Y., Rice Jr, J.B., 2005. A supply Chain View of the resilient Entreprise. Mit Sloan
Management Review. 47
Sirmon, D.G., Hitt, M.A., Ireland, R.D., Gilbert, B.A., 2011. Resource Orchestration to Create
Competitive Advantage. Journal of Management. 37, 1390-1412

130

Slack, N., 1983. Flexibility as a manufacturing objective. International Journal of Operations &
Production Management. 3, 4-13
Sodhi, M. S., Tang, C. S., 2012. Managing supply chain risk, Springer Science & Business
Media. 172
Srivastava, M. K, Gnyawali, D. R., 2011. When do relational resources matter? Leveraging
portfolio technological resources for breakthrough innovation. Academy of Management
Journal. 54, 797-810
Stevenson, M., Spring, M., 2007. Flexibility from a supply chain perspective: definition and
review. International Journal of Operations & Production Management. 27, 685-713
Stuart, T. E., 2000. Interorganizational alliances and the performance of firms: a study of growth
and innovation rates in a high-technology industry. Strategic Management Journal. 21,
791-811
Sun, H., Wu, J., 2005. Scale-free characteristics of supply chain distribution networks. Modern
Physics Letters B. 19, 841-848
Swink, M., Talluri, S., Pandejpong, T., 2006. Faster, better, cheaper: A study of NPD project
efficiency and performance tradeoffs. Journal of Operations Management. 24, 542-562
Talluri, S.S., Kull, T.J., Yildiz, H., Yoon, J., 2013. Assessing the efficiency of risk mitigation
strategies in supply chains. Journal of Business Logistics. 34, 253-269
Tan, J., Peng, M.W., 2003. Organizational slack and firm performance during economic
transitions: Two studies from an emerging economy. Strategic Management Journal. 24,
1249-1263
Tang, C., Tomlin, B., 2008. The power of flexibility for mitigating supply chain risks.
International Journal of Production Economics. 116, 12-27
Tang, C.S., 2006. Perspectives in supply chain risk management. International Journal of
Production Economics. 103, 451-488
Thaler, R. S., Tversky, A., Kahneman, D., & Schwartz, A., 1997. The effect of myopia and loss
aversion on risk taking: an experimental test. Quarterly Journal of Economics. 112, 647+
Thun, J.-H., DrĂźke, M., Hoenig, D., 2011. Managing uncertainty - an empirical analysis of
supply chain risk management in small and medium-sized enterprises. International
Journal of Production Research. 49, 5511
Upton, D., 1995. What really makes factories flexible? Harvard Business Review. 73, 74-84
Urbig, D., BĂźrger, R., Patzelt, H., Schweizer, L., 2013. Investor Reactions to New Product
Development Failures The Moderating Role of Product Development Stage. Journal of
Management. 39, 985-1015

131

Vandenberg, R.J., Lance, C.E., 2000. A review and synthesis of the measurement invariance
literature: Suggestions, practices, and recommendations for organizational research.
Organizational Research Methods. 3, 4-70
Vanderbyl, S., Kobelak, S., 2008. Risk management for the biotechnology industry: A Canadian
perspective. Journal of Commercial Biotechnology. 14, 128-140
Vasudeva, G., Anand, J., 2011. Unpacking absorptive capacity: A study of knowledge utilization
from alliance portfolios. Academy of Management Journal. 54, 611-623
Vermeulen, F., Barkema, H., 2001. Learning through acquisitions. Academy of Management
Journal. 44, 457-476
Vokurka, R.J., O'Leary-Kelly, S.W., 2000. A review of empirical research on manufacturing
flexibility. Journal of Operations Management. 18, 485-501
Wagner, S.M., Bode, C., 2006. An empirical investigation into supply chain vulnerability.
Journal of Purchasing and Supply Management. 12, 301-312
Wagner, S.M., Bode, C., 2008. An empirical examination of supply chain performance along
several dimensions of risk. Journal of Business Logistics. 29, 307-325
Wagner, S.M., Neshat, N., 2012. A comparison of supply chain vulnerability indices for different
categories of firms. International Journal of Production Research. 50, 2877-2891
Weber, E.U., Hsee, C., 1998. Cross-cultural differences in risk perception, but cross-cultural
similarities in attitudes towards perceived risk. Management Science. 44, 1205-1217
Weiss, D., Maher, M.W., 2009. Operational hedging against adverse circumstances. Journal of
Operations Management. 27, 362-373
Wu, Q., Luo, X., Slotegraaf, R. J, Aspara, J., 2015. Sleeping with competitors: the impact of
NPD phases on stock market reactions to horizontal collaboration. Journal of the
Academy of Marketing Science. 43, 490-511
Xiao, T., Yang, D., 2008. Price and Service Competition of Supply Chains with Risk-averse
Retailers under Demand Uncertainty. International Journal of Production Economics.
114, 187 – 200
Zhang, M., Batta, R., Nagi, R., 2009. Modeling of Workflow Congestion and Optimization of
Flow Routing in a Manufacturing/Warehouse Facility. Management Science. 55, 267-280
Zollo, M., Winter, S. G., 2002. Deliberate learning and the evolution of dynamic capabilities.
Organization science. 13, 339-351

132