BUSINESS RESPONSES TO SUPPLY CHAIN DYNAMICS: IMPLICATIONS FOR
SOURCING AND PUBLIC POLICY
By
Zhenzhen Yan
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
2022
ABSTRACT
ESSAYS ON BUSINESS RESPONSES TO SUPPLY CHAIN DYNAMICS: IMPLICATIONS
FOR SOURCING AND PUBLIC POLICY
By
Zhenzhen Yan
This dissertation studies drivers and consequences of firms' policy decisions from a supply chain
perspective. Policy decisions can be both internal and external to the firms and carry substantial
implications for firms’ performance outcomes. Both internal and external aspects of policy
engagement are critical since they help further disciplinary understanding of not only how policy
decisions influence the performance of products firms manufacture, but also the fact that firms can
reshape their environmental conditions through adopting policy decisions to engage with
regulators and impact their own business practices. Specifically, I investigate firms' sourcing
policies and their product performance implications – an internal policy set by firms. Within the
context of external policy engagement, my research investigates firms' policy responses to external
conditions, such as climate change public policies.
The first essay focuses on manufacturing localization, an internal policy set by automakers
to relocate manufacturing activities closer to their target market. I apply a causal estimation method
to 23-year panel data of the automotive industry complied from diverse data sources including
Wards, National Highway and Traffic Safety Authority (NHTSA) recall data to investigate the
quality implications of manufacturing localization. The results show an immediate quality decline
for localized vehicles, indicated by an increase of 68.203% in the number of recalls (equivalent to
0.552 more recall campaigns) and an increase of customer complaints by 56.831% (equivalent to
9.997 more complaints) in the three years after the manufacturing localization. The increased
number of recalls can lead to an extra expense of $2.76 million for a localized vehicle based on
conservative estimations. A cautionary note is thus issued for automakers to adjust their planned
budgets to account for warranty claims before the localization. This study provides guidance for
firms considering the relocation of their manufacturing activities and for regulators that seek to
reduce vehicle recalls.
The second and third essays of my dissertation investigate firms’ engagement in
influencing climate change policies (EICCP), which refers to firms’ strategic actions to influence
climate change policymaking processes, aiming at reshaping policies or promoting policy changes
in favor of their interests. In the second essay, I conceptualize EICCP and propose a taxonomy for
EICCP strategies considering firms’ perceptions of external conditions and internal resources. I
also create measures for EICCP to empirically validate this taxonomy by performing text analytics
with machine-learning techniques on firms’ self-disclosure in CDP Climate Change data. Building
on the second essay, I undertake a large-scale empirical analysis to investigate antecedents of
EICCP in the third essay. Specifically, I examine the regulatory risks associated with climate
change and firms’ supply network complexity as critical and interrelated factors for firms’ EICCP.
The latter two essays contribute to the growing literature on climate change and firm responses.
Overall, my thesis responds to the call for policy-related studies in the supply chain field
and provides insights for policy decision-makers.
ACKNOWLEDGEMENTS
I would like to express my deepest appreciation to Professor Sriram Narayanan, and Professor
Tobias Schoenherr, the co-chairs of my dissertation committee, for their constructive suggestions
and warm support. This dissertation would not be possible without your continuous guidance. I
would also like to extend my sincere gratitude to Professor Adrian Choo, and Professor Shawnee
K. Vickery, for their commitment to my dissertation committee and their insightful comments.
iv
TABLE OF CONTENTS
LIST OF TABLES ........................................................................................................................ vii
LIST OF FIGURES ..................................................................................................................... viii
CHAPTER 1- Introduction ............................................................................................................. 1
1.1. Introduction ........................................................................................................................ 1
CHAPTER 2 – Manufacturing Localization and its Performance Implications: An Empirical
Study in the Automotive Industry ................................................................................................... 4
2.1. Introduction ........................................................................................................................ 4
2.2. Literature Review............................................................................................................... 8
2.3. Methodology .................................................................................................................... 12
2.3.1 Data and Sample Construction................................................................................. 12
2.3.2. Model and Estimation ............................................................................................. 16
2.3.3. Validity of DID Approach ...................................................................................... 19
2.4. Results .............................................................................................................................. 21
2.4.1. Overall Effects of Localization on Vehicle Quality ............................................... 21
2.4.2 Post-Hoc Analysis.................................................................................................... 23
2.5. Conclusion and Implications............................................................................................ 24
APPENDIX ................................................................................................................................... 29
REFERENCES ............................................................................................................................. 33
CHAPTER 3 – Firms' Policy Engagement on Climate Change: Taxonomy Development and
Validation...................................................................................................................................... 40
3.1. Introduction ...................................................................................................................... 40
3.2. Literature Review and Theory Development for EICCP ................................................. 44
3.2.1. Literature Review.................................................................................................... 44
3.2.2. Typologies of EICCP strategies .............................................................................. 46
3.3. Measuring EICCP strategies ............................................................................................ 56
3.3.1. Data ......................................................................................................................... 57
3.3.2 Text Analytics Using Natural Language Processing (NLP) with Random Forest
Classification..................................................................................................................... 57
3.3.3 Manual Labelling ..................................................................................................... 60
3.3.4 Feature Engineering ................................................................................................. 65
3.3.5 Model Training ........................................................................................................ 68
3.4. Evaluation and Results of EICCP Measures .................................................................... 69
3.5. Discussion ........................................................................................................................ 76
3.5.1. Overview of EICCP ................................................................................................ 76
3.5.2. Contributions .......................................................................................................... 79
REFERENCES ............................................................................................................................. 81
CHAPTER 4 – Supply Network Complexity, Regulatory Risks, and Firms’ Engagement in
Influencing Climate Change Policies............................................................................................ 88
v
4.1. Introduction ...................................................................................................................... 88
4.2. Literature Review............................................................................................................. 91
4.2.1 Theoretical Underpinnings for EICCP Studies ........................................................ 91
4.2.2. Supply Chain Complexity Research ....................................................................... 92
4.3. Hypothesis Development ................................................................................................. 96
4.3.1 Regulatory Risks and EICCP ................................................................................... 96
4.3.2. Moderating Effects of Supply Network Complexity on the Regulatory Risks-
EICCP Link....................................................................................................................... 97
4.4. Methodology .................................................................................................................... 99
4.4.1. Data ......................................................................................................................... 99
4.4.2. Variables ............................................................................................................... 101
4.4.3. Econometric Models ............................................................................................. 103
4.5. Results ............................................................................................................................ 106
4.6. Discussion and Conclusion ............................................................................................ 108
APPENDICES ............................................................................................................................ 111
REFERENCES ........................................................................................................................... 122
vi
LIST OF TABLES
Table 1.1 Descriptions of treated vehicles .................................................................................... 15
Table 1.2 Summary statistics ........................................................................................................ 20
Table 1.3 Testing for parallel trends assumption .......................................................................... 21
Table 1.4 Estimates of the treatment effects on vehicle quality ................................................... 22
Table 1.5 Estimates of the treatment effects of localization on vehicle quality, differentiating the
plant history (Panel A) .................................................................................................................. 25
Table 1.6 Estimation of cost per recall campaign ......................................................................... 28
Table 2.1 Illustrations of manual labeling standards and text examples for each strategy ........... 61
Table 2.2 Word features with the highest biserial correlation coefficients by manual label ........ 68
Table 2.3 Results of the random forest models by manual labeling ............................................. 72
Table 2.4 Top-ranked word features with their importance value in the random forest model by
manual labeling ............................................................................................................................. 72
Table 3.1 Summary statistics and data sources ........................................................................... 105
Table 3.2 Results of the Poisson regression................................................................................ 107
Table 3.3 Summary of Existing Research on Supply Chain Complexity ................................... 112
Table 3.4 Sensitivity Tests .......................................................................................................... 121
vii
LIST OF FIGURES
Figure 2.1 Taxonomy of EICCP ................................................................................................... 50
Figure 2.2 Presumptive results of decision tree model and random forest models ...................... 59
Figure 2.3 The interface of Atlas.ti ............................................................................................... 60
Figure 2.4 Data pre-processing for feature engineering ............................................................... 65
Figure 2.5 The relationship between the accuracy and the number of randomly selected word
features for the model training for the promotion strategy ........................................................... 76
Figure 2.6 Implementation trends for different EICCP strategies ................................................ 78
Figure 2.7 Longitudinal comparison of four EICCP strategies classified by value perspectives
and engagement levels .................................................................................................................. 78
Figure 3.1 Theoretical model ........................................................................................................ 99
Figure 3.2 Effects of regulatory risks on EICCP at different values of supply network complexity
.....................................................................................................................................................108
viii
CHAPTER 1- Introduction
1.1. Introduction
Responding to environmental changes is essential for business success. Firms can respond to such
dynamics in two ways: First, firms can revamp their internal processes or operations to adapt to
the new environmental conditions. And second, firms can reshape their external environment in
favor of their interests. This dissertation focuses on two understudied strategic decisions that
belong to these two ways to respond to the ever-present environmental dynamics. The first one is
the manufacturing localization decision, which reflects firms' efforts to adapt to global competitive
conditions by redefining their manufacturing and sourcing policy. The second one is firms’
engagement in climate change policymaking processes, which is indicative of their endeavor to
enhance organizational legitimacy and maintain competitive advantages by influencing climate
change policies.
The first essay focuses on manufacturing localization, which refers to firms’ decision to
relocate manufacturing activities closer to the target market. Manufacturing localization can bring
substantial changes to production and lead to uncertain quality outcomes. However, investigations
into the consequences of manufacturing localization are largely absent, primarily due to the
difficulty of tracking the manufacturing performance before and after the relocation of production
activities. I overcome these challenges and, using the automotive industry as the empirical context,
investigate the overall quality implications of localizing vehicle production closer to the U.S.
market. I apply causal estimation to a unique dataset across vehicle profiles, automakers’
relocation decisions, recalls, customer complaints and other industry-specific proprietary data
spanning more than 23 years. The results show an immediate quality decline for localized vehicles,
indicated by an increase of 68.203% in the number of recalls (equivalent to 0.552 more recall
1
campaigns) and an increase of customer complaints by 56.831% (equivalent to 9.997 more
complaints) in the three years after the manufacturing localization. The increased number of recalls
can lead to an extra expense of $2.76 million for a localized vehicle based on conservative
estimations. A cautionary note is thus issued for automakers to adjust their planned budgets to
account for warranty claims before the localization. My post-hoc exploration suggests that the
quality decline might result from both the learning effects and the difficulties in technological
transfer. Overall, by investigating the quality implications of manufacturing localization decisions
in the automotive industry, my results provide guidance for firms considering the relocation of
their manufacturing activities and for regulators that seek to reduce vehicle recalls.
The second and third essays of my dissertation explore external policy engagement aspects
of a firm in the climate change context. Extant studies note policies as irresistible contexts and
concentrate on how firms survive the climate change scrutiny by adopting sustainable practices.
The attention to firms’ endeavor to reversely influence policymaking to redirect or evade
regulatory scrutiny is scarce. Yet, the investigation of how firms reshape the regulatory context to
solve the problem of practice-policy decoupling and obtain legitimacy under scrutiny over climate
impact is important to further the understanding of firms' overall sustainability strategy. In the
second essay, I introduce the concept of engagement in influencing climate change policies
(EICCP). EICCP refers to firms’ strategic actions to influence climate change policymaking
processes, aiming at reshaping policies or promoting policy changes in favor of their interests.
Building on the literature on corporate political actions and environmental politics, I propose a
taxonomy for EICCP strategies considering firms’ perceptions of external conditions and internal
resources. Then, I validate this taxonomy by examining strategies of EICCP adopted in the real
world. Specifically, I perform text analytics with machine-learning techniques on firms’ self-
2
disclosure in CDP Climate Change data to identify different strategies and generate automated
approaches to coding the strategies pursued by these companies with the aim of analyzing them.
The measures I created for EICCP can be applied to future research. In the third essay, I incorporate
the data generated from the machine-learning-based strategy identification to undertake a large-
scale empirical analysis to investigate antecedents of EICCP. Specifically, I examine the
regulatory risks associated with climate change and firms’ supply network complexity as critical
and interrelated factors for firms’ EICCP. The latter two essays contribute to the growing literature
on climate change and firm responses.
Overall, my thesis responds to the call for policy-related studies in the supply chain field
and provides insights for policy decision-makers.
3
CHAPTER 2 – Manufacturing Localization and its Performance Implications:
An Empirical Study in the Automotive Industry
2.1. Introduction
Uncertainties associated with the management of global supply chains have intensified
significantly in recent years, with the Covid-19 pandemic and the far-reaching geopolitical issues
such as the U.S.– China trade tensions serving as illustrative examples. Considering the increasing
challenges to manage cross-border movements of goods, coupled with the worldwide shortage of
labor and supply, manufacturers have been keen to regain control of their supply chains (Cherney,
2020; Gryta and Cutter, 2021). One way to accomplish this is to relocate manufacturing activities
closer to the target market, which has been a prevalent option as indicated by several surveys in
the recent past (BDO, 2021; Ma et al., 2021; NielsonIQ, 2020). I refer to this phenomenon as
manufacturing localization (hereinafter, localization). Localization carries the promise to
strengthen a supply chain’s resilience (Schwartz, 2022), which is imperative in our current
environment in which volatility is here to stay (Rosenbaum, 2021). Furthermore, emerging policies
increasingly favor localization (Raza et al., 2021; The White House, 2021), boding well for it to
continue to be a priority for many firms going forward (Haex and Buck, 2022; Wellener et al.,
2022). Yet, getting closer to the market is easier said than done, often being associated with
significant costs—the relocation of production capacity may require substantial investments and
reconstruction of a firm’s supply and manufacturing networks (Curran, 2021). Moreover, supply
chain scholars have indicated that relocation per se does not imply resilience—firms may still be
challenged by industrial, technological, and operational constraints (Simchi-Levi and Simchi-Levi,
2020).
4
With such a plethora of challenges and roadblocks on the way toward localization, firms
need to have a clear picture of their localization decision’s implications. However, while literature
is rich in the overall economic potential of localization at the country or regional level (Ma et al.,
2021; Raza et al., 2021), a significant void exists for insight into the consequences of localization
at the firm level. Exceptions include broad-based managerial research on reshoring, which can be
considered a special type of localization—in this vein, survey findings suggest that reshoring
influences customer-perceived product quality (Cassia, 2020) and employees’ citizenship
behaviors (Grappi et al., 2020). Where empirical evidence is however missing is the impact of
localization on supply chain and operations management derived through secondary data analysis
(rather than perceptual data). What may have prevented this research in the past is that collecting
rich, secondary data from firms across the world that engage in localization is challenging, if not
impossible (Gray et al., 2011, p. 737; 2013, p. 31).
I overcome this challenge in the present study by choosing the automotive industry as my
empirical setting and generating a unique panel dataset at the vehicle level compiled from various
sources. The automotive industry was chosen since the practice of localization has been prevalent
in this industry for decades, making it a fertile ground to study the consequences of localization.
For example, Hyundai established a manufacturing plant in Alabama to better serve the U.S.
market in 2005 (HMMA, 2006), Volvo opened its first U.S. car assembly plant to assemble sedan
and sport utility vehicles in 2015 (CNN, 2015), and Volkswagen established an assembly plant in
Kenya to target the East African market in 2016 (Reuters, 2016). Recently, Tesla broke ground
for the company’s largest battery plant in Berlin-Brandenburg, Germany (Tesla, 2020), to tailor
products for the European market (Tech Explore, 2020).
5
With an average of 30,000 parts that comprise a vehicle and the inherent coordination and
manufacturing complexity (Foldy, 2020), the localization decision can bring substantial changes
to a vehicle’s production operations with the potential for significant quality implications. For
example, moving an existing production line and re-establishing it elsewhere can disrupt
production planning of the relocated product, while at the same time disrupting the destination
plant’s operations through the potential redesign of workflows, the reallocation of the workforce
and other resources, and the additional coordination now necessary among the product lines (Gopal
et al., 2013). These operational changes may yield an increase in production errors due to learning
processes and the adaptation to the new manufacturing conditions (Badiru, 1998; Clark and
Fujimoto, 1992). It is thus reasonable to expect that localization will negatively impact product
quality, at least in the short term. However, getting closer to the target market also allows
automakers to reduce manufacturing complexity through the postponement of product
customization and the consolidation of customer requirements in a region (Brun and Zorzini, 2009).
It also enables automakers to better capture consumer tastes and market trends, offering the
opportunity to enhance and tailor product designs accordingly (Ellram et al., 2013). As such,
localization may also lead to better vehicle quality through the reduction of manufacturing
complexity and superior response to market demands. These differing perspectives establish
localization as an intriguing context, motivating my investigation of how localization influences
vehicle quality.
Given that automakers have been actively localizing manufacturing operations (HMMA,
2006; CNN, 2015; Reuters, 2016; Tesla, 2020; Tech Explore, 2020), research that uncovers the
quality outcomes of such decisions is timely, with the results destined to be impactful given the
different trajectories leading to either better or worse quality as outlined above. Using the U.S.
6
automotive industry as my empirical setting, I seek in this study to uncover the quality implications
of relocating vehicle production from foreign plants to domestic plants. My first research question
is thus stated as follows: Does localization result in product quality changes for localized products?
(RQ1)
The impact of a location change on quality is likely dependent on contextual factors at the
destination plant, which I thus consider in my analysis. For example, localization to a new plant
may negatively influence vehicle quality due to the learning effect, while localization to an existing
plant may cause quality problems due to difficulties in technology transfer. With this framing, an
intriguing question that remains unanswered is what the mechanisms are through which quality
declines after localization. I will aim to provide an answer to this question in my post-hoc analysis.
My second research question aims to assess the likely disruptions to overall plant
operations caused by the introduction of the newly localized vehicles in that plant. These
disruptions can be reflected in the quality of other vehicles that had already been produced at that
location. Considering these potential effects is important since prior literature suggests that the
introduction of a new model can lead to a substantial decline in plant-level productivity (Gopal et
al. 2013). This lower productivity may then translate into quality implications for other existing
products manufactured in the plant. I, therefore, investigate the potential impact on product quality
for the existing products in that plant, i.e., those vehicles that are not the subject of the localization.
My second research question is thus as follows: Does localization result in changes to product
quality for other vehicles produced in the destination plant? (RQ2)
To address the aforementioned research questions, I compile a secondary dataset that spans
from 1996 to 2019 for automobile sales in the U.S. and apply a general difference-in-difference
(DID) approach to examine the causal effect of localization. The U.S. provides a unique context
7
to investigate these dynamics since it is the second-largest automobile market in the world, having
attracted foreign automakers to invest more than $75 billion for local and nearby production from
1982 to 2018 (ITA, 2018).
To foreshadow my results, I find that localization of vehicle production negatively
influences the quality of localized vehicles. Utilizing the number of recalls and complaints as two
distinct quality indicators, I observe an increase in both measures after the localization, with these
effects being persistent even several years after the localization event. Specifically, I find that the
number of recalls increases by 68.203%, and the number of complaints increase by 56.831% for
localized vehicles. The increased number of recalls alone (0.552 more recall campaigns for
localized vehicles after localization than before localization) would cause $2.76 million in extra
expenses for a localized vehicle, which is a conservative estimate based on financial damage
statistics of historical recalls (Held et al., 2018; Isidore, 2015). The increase in the number of
recalls is agnostic to different localization decisions (localization to a new plant vs. localization to
an existing plant), but the increase in the number of complaints only happens with localization to
a new plant. Further, I find that the introduction of localized vehicles does not trigger more recalls
or complaints about other vehicles produced in the same plants.
2.2. Literature Review
Research on manufacturing location decisions is rich due to the complexities associated with
location-specific dynamics such as geopolitical and cultural dimensions, as well as the ensuing
impact on optimal manufacturing network configurations (Brennan et al., 2015; Cheng et al., 2015).
As such, research at the firm level has a long history of investigating firms’ decision patterns in
moving manufacturing facilities from their domestic locations to emerging or developing
economies based on a low-cost rationale (e.g., Bock, 2008; Gray et al., 2011; Srai and Ané, 2016).
8
Another stream of literature on firms’ internationalization also investigates offshoring as an
indispensable process through which firms improve cost efficiency (e.g., Dunning, 1970; Vahlne
and Johanson, 2017; Vernon, 1966). More recently, triggered by discussions on the best global
manufacturing locations that do not conform to the low-cost imperative (Ellram et al., 2013),
scholars have started to investigate less traditional strategies such as reshoring and manufacturing
in high-cost countries (J. V. Gray et al., 2013; Ketokivi et al., 2017). For example, studies noted
that manufacturing lead times and local market responsiveness are critical decision factors in
relocating operations to high-cost countries (de Treville et al., 2014), as is the consideration of
cultural distance in driving quality risk in offshore locations (Gray et al., 2013). Yet others justify
the decision to manufacture in high-cost countries based on the interdependencies between
production and other value chain activities, such as research and development (Ketokivi et al.,
2017). The geographical proximity of upstream and downstream production can also have benefits
for the assurance of product quality due to enhanced communication among supply chain
stakeholders (Bray et al., 2019). These observations favor manufacturers bringing production
activities closer to the respective target markets, instead of relying on import-oriented operations
from other countries.
Studies on localization decisions, including reshoring and offshoring literature, provide
various rationales for firms’ manufacturing location decisions (Foerstl et al., 2016; Fratocchi et al.,
2016; Wiesmann et al., 2017). Most of these studies are qualitative and highlight product quality
as a key driver for manufacturers to move to higher-cost countries. However, empirical evidence
on the actual quality implications of manufacturing location/relocation decisions is limited. I aim
to fill this void with the present study, and in doing so, I rely on two streams of literature.
9
The first stream explores the impact of manufacturing location strategies on product quality.
By and large, studies in this realm suggest that offshoring manufacturing to other countries is
negatively associated with product quality, due to the difficulty of monitoring quality and the lack
of skilled labor (da Silveira, 2014; Dana et al., 2007). In this vein, Gray et al. (2011) provide
empirical evidence of the quality risks that U.S. pharmaceutical firms incur when adopting
offshore manufacturing in Puerto Rico. However, Stentoft et al. (2018) report no significant
differences in product quality under domestic production, offshoring, and reshoring scenarios; a
caveat here however is that the study focused on perceived quality captured via survey research.
The relevance of these studies to the localization–quality link within my context goes back to the
possibility that products made in offshored facilities are sold to customers residing in these
offshore locations. In this case, offshore manufacturing is equivalent to localization in a lower-
cost country. None of these studies specify the target market of the offshore-manufactured
products.
The second stream of literature couples the manufacturing location decision with marketing
and identifies customer proximity as a key motivation for localization. The distance between the
point of production and the point of consumption is a continuing concern for firms due to high
shipping costs, long lead times, lack of market responsiveness, and inventory management
challenges (The Economist, 2013). However, studies on manufacturing location decisions have
identified that moving manufacturing activities closer to the target market is one of the most
pursued solutions (e.g., Johansson and Olhager, 2018). From a marketing standpoint, localization
makes it easier for firms to understand customer tastes and market trends, improving product
design and product performance (Ellram et al., 2013). This is also consistent with operations
literature noting that localization allows the postponement of product customization and
10
consolidates a region’s customer requirements, reducing overall manufacturing complexity and
quality risk (Bailey and De Propris, 2014; Ben-Ner and Siemsen, 2017; Moradlou et al., 2017).
These studies, however, do not offer direct evidence of quality performance post localization.
Furthermore, leaving the impact on the focal product aside, the localization decision can
also have significant ramifications on the existing products produced in that plant, which is a
dynamic that has not been examined. The logic is that the newly introduced vehicles can cause
disruptions in the plant’s regular operations, for instance in terms of production planning and
coordination, resource allocations, and product flows. In this realm, Gopal et al. (2013)
demonstrate that a new product introduction reduces a plant’s productivity in the automotive
industry due to “the engineering changes, retooling, and reprogramming of equipment,
resequencing of processes, possible retraining of workers and correction of ongoing errors” (p.
2218). Thus, it is likely that quality challenges in localizing a vehicle may spill over to existing
vehicles’ production lines, offering further motivation for my investigation. After all, quality
challenges happen at the product level, rather than at the plant level.
Finally, my research is also related to studies that have examined the quality implications
of manufacturing plant strategies within the automotive industry. Specifically, Shah et al. (2017)
examine the impact of utilization on product recalls tied to a specific plant. In contrast to their
study, I examine the quality implications of localization decisions. Further, Lacetera and Sydnor
(2015) compare the quality of vehicles made by the same manufacturers but assembled in different
locations, including Japan and the U.S., using data from U.S. wholesale used-car auctions. While
the authors find that there are no significant quality differences between products manufactured in
these two countries, they do not provide direct insight into the effect of localization, as the
relocation event from Japan to the U.S. is not explicitly considered.
11
2.3. Methodology
2.3.1 Data and Sample Construction
To examine the quality implications of localization decisions, I use the U.S. automotive market as
my empirical setting and study the population of 704 vehicles models that were sold in the U.S.
market from 1996 to 2019. I obtain vehicle-level data from two data sources, the Wards
Intelligence data set (Wards hereinafter) and the NHTSA/ODI (National Highway Traffic
Administration/Office of Defects Investigation) database (NHTSA hereinafter). Each vehicle’s
profile (including the manufacturer and brand information, classification, and segments), annual
sales volume, and location of production are derived from Wards. The count of recalls and
complaints about a vehicle each year are provided by NHTSA. Considering that the recalls or
complaints might not immediately occur upon a vehicle entering the market, I track the recall and
complaint announcements for three years after a vehicle’s official launch. Therefore, the data from
NHTSA spans from 1996 to 2021. The complied panel contains 6,826 vehicle-year observations.
Spanning the years 1996 to 2019, my panel dataset records each vehicle’s source of
production for every year. As defined by Wards, domestically produced vehicles are assembled in
domestic plants located in the U.S., Canada, and Mexico, and imported vehicles are assembled in
foreign plants located in other countries. I am thus able to identify the event of localization as the
point in time when automakers relocate the production of a vehicle from a foreign plant to a
domestic plant. This setting is consistent with my definition of localization, which is the relocation
of manufacturing activities closer to the U.S. market. For example, Honda moved the production
of its 2007 Honda CR-V from its Japanese plant to its U.S. plant in East Liberty, Ohio (Honda,
2006).
12
I manually verified the shift of production location using the automakers’ official news
releases and other reliable sources such as Automotive News and Autoblog. To investigate the
quality change resulting from localization, I restrict my sample to vehicles that were originally
produced in foreign plants, ensuring a consistent production condition for all vehicles in the sample.
This sample, which I refer to as Panel A hereinafter, contains 365 vehicles. A total of 35 vehicles
in Panel A had their production relocated from foreign plants to domestic plants between 1997 and
2019 and maintained such production location until the end of 2019 or until the vehicles were
discontinued—this constitutes the treatment group for localization. One vehicle was once localized
but then relocated to a foreign plant again. I removed this vehicle from my sample to avoid the
confounding effect of multiple relocation decisions. The other 329 vehicles were produced in
foreign plants during the time span of my observation, which constitute my control group. A list
of the 35 treated vehicles is presented in Panel A of Table 1.1, including their year of localization
and plant information.
Among the 35 localized vehicles, 14 vehicles had their production moved to a domestic
plant that was opened within three years prior to the relocation. I consider this type of relocation
decision as localization to a new plant. I choose three years as a cut-off point, when identifying
new plants, instead of only the year when the plant was opened to account for manufacturing
changes that potentially go on in the first few years of plant opening. In contrast, the production
of the other 21 vehicles was relocated to plants that the automakers had owned for more than three
years at the time of relocation, which I refer to as localization to an existing plant. To examine
whether introducing the localized vehicles to an existing plant interrupts the plant operation and
provokes quality problems for the vehicles produced in this plant, I construct another sample,
which I refer to as Panel B. Panel B contains 314 vehicles that were produced in domestic plants
13
during the time span of my observation. Among them, 15 vehicles experienced the introduction of
localized vehicles in their plants in only one year in any five-year window, and nine vehicles
experienced such disruption in at least two years within a five-year window. I eliminated the nine
vehicles that experienced the multiple introductions of localized vehicles to remove confounding
effects of disruptions for this vehicle in different years. I choose to use the five-year window since
automakers rely on mid-term planning and five years should be enough to ease any changes
provoked by the prior introduction of any vehicle. Therefore, the treatment group for the
introduction of localized vehicles consists of 15 vehicles, with the control group consisting of the
other 290 vehicles that never experienced the introduction of localized vehicles. I list the 15 treated
vehicles, the year of intervention, their plant information, and the introduced localized vehicles in
Panel B of Table 1.1.
I further removed vehicles that have less than five years of records to ensure the data of
each vehicle in the sample is longitudinal. The final sample of Panel A includes 3,241 vehicle-year
observations for 274 vehicles, with 34 vehicles in the treatment group and 240 in the control group.
The final sample of Panel B contains 2,555 vehicle-year observations for 218 vehicles, with 13
vehicles in the treatment group and 205 vehicles in the control group. Using these final samples
versus the full samples yields similar results with identical findings (I report the results for the
final samples in the result section).
14
Table 1.1 Descriptions of treated vehicles
Panel A. List of vehicles in the treatment group of Localization
Vehicle Year of Localization Plant New plant
VOLKSWAGEN CABRIO 1997 Puebla No
ISUZU AMIGO 1998 Lafayette No
ACURA TL 1999 Marysville No
HONDA ODYSSEY 1999 Alliston 2 Yes
NISSAN MAXIMA 2003 Smyrna No
LEXUS RX330 2004 Cambridge No
HYUNDAI SONATA 2005 Montgomery Yes
NISSAN PATHFINDER 2005 Smyrna No
HYUNDAI SANTA FE 2006 Montgomery Yes
HONDA CR-V 2007 East Liberty No
TOYOTA RAV4 2009 Woodstock Yes
KIA SORENTO 2010 West Point Yes
TOYOTA HIGHLANDER 2010 Princeton No
BMW X3 2011 Spartanburg No
HYUNDAI ELANTRA 2011 Montgomery No
VOLKSWAGEN PASSAT 2011 Chattanooga Yes
KIA OPTIMA 2012 West Point No
NISSAN LEAF 2013 Smyrna No
HONDA FIT 2014 Celaya Yes
LEXUS RX450 2014 Cambridge No
MAZDA MAZDA3 2014 Salamanca Yes
NISSAN ROGUE 2014 Smyrna No
VOLKSWAGEN GTI 2014 Puebla No
MERCEDES-BENZ C CLASS 2015 Vance No
NISSAN MURANO 2015 Canton No
LEXUS ES350 2016 Georgetown No
AUDI Q5 2017 San Jose Chiapa Yes
KIA FORTE 2017 Monterrey Yes
SUBARU IMPREZA 2017 Lafayette No
VOLKSWAGEN TIGUAN 2017 Puebla No
HYUNDAI ACCENT 2018 Monterrey Yes
INFINITI QX50 2018 Aguascalientes No
KIA RIO 2018 Monterrey Yes
MERCEDES-BENZ SPRINTER VAN 2019 Ladson Yes
VOLVO S60 2019 Ridgeville Yes
Note. LEXUS RX330 was dropped from the analysis since it has less than five years of
observations.
15
Table 1.1 (cont’d)
Panel B. List of vehicles in the treatment group of Introduction
Vehicle Year of Plant Localized Vehicles
Introducing
Localized
Vehicles
ISUZU RODEO 1998 Lafayette ISUZU AMIGO
NISSAN ALTIMA 2003 Smyrna NISSAN MAXIMA
HONDA ELEMENT 2007 East Liberty HONDA CR-V
TOYOTA SEQUOIA 2010 Princeton TOYOTA HIGHLANDER
TOYOTA SIENNA 2010 Princeton TOYOTA HIGHLANDER
BMW X5 2011 Spartanburg BMW X3
BMW X6 2011 Spartanburg BMW X3
INFINITI JX 2013 Smyrna NISSAN LEAF
SUZUKI EQUATOR 2013 Smyrna NISSAN LEAF
MERCEDES-BENZ C
MERCEDES-BENZ GL 2015 Vance CLASS
MERCEDES-BENZ C
MERCEDES-BENZ M CLASS 2015 Vance CLASS
NISSAN NV 2015 Canton NISSAN MURANO
NISSAN TITAN 2015 Canton NISSAN MURANO
TOYOTA AVALON 2016 Georgetown LEXUS ES350
NISSAN KICKS 2018 Aguascalientes INFINITI QX50
Note. INFINITI JX was dropped from the main analysis since it has less than five years of
observations.
In my main analysis, I use three years of data in the pre-localization and pre-introduction
phases and three years in the post-localization and post-introduction phases. I select this time span
based on the expectation that manufacturers will work to continuously improve product quality
such that the impact of manufacturing localization on car quality lasts only for a short term
(Bandyopadhyay and Jenicke, 2007; Staeblein and Aoki, 2015). Further, a period of three years is
reasonably long that any impact on quality can be captured clearly.
2.3.2. Model and Estimation
I use a difference-in-differences (DID) design to estimate how localization influences the quality
16
of the localized vehicles and the other vehicles in localized vehicles’ destination plants within a
23-year window across 14 events of localization to a new plant and 21 events of localization to an
existing plant. DID was deemed as the most appropriate design for investigating the treatment
effects of an intervention (i.e., the localization and introduction of localized car models in a
domestic plant). I noted that in this study, the events of manufacturing localization, and
consequently the events of the introduction of localized vehicles, did not happen in the same year
for all vehicles. For example, BMW localized the production of the vehicle BMW X3 in the U.S.
in 2011, while Nissan made the localization decision for Nissan Rogue in 2014 (see Table 1.1). To
account for this varying treatment timing, I utilize the staggered DID analysis with the two-way
fixed effect (TWFE) regression model, which has become prevalent in staggered DID designs over
the past two decades (Baker et al., 2022). Following Baker et al.'s (2022) notation, the model
specification is as follows:
!!" = $! + &" + ' ## (!" + )!" ,
where $! and &" specify the unit and time fixed effects, respectively. (!" is the indicator for the
treated group in the post-treatment periods, with ' ## being the estimate of an average treatment
effect across all treatment years. However, recent econometric studies in DID applications posit
that the average treatment effect (ATE) or the average treatment effect on the treated (ATT)
estimated by the TWFE regression model may not be valid and interpretable (e.g., Callaway and
Sant’Anna, 2021; Imai and Kim, 2021; Sun and Abraham, 2021). Specifically, Goodman-Bacon
(2021) indicates that the TWFE treatment-effect estimate is a weighted average of all possible
constituent 2x2 DID estimates. To overcome the challenge that the late treatment events confound
the early treatment events in the estimation, I conduct a stacked regression following Cengiz et al.
(2019). This approach considers the time-varying treatments and treatment heterogeneity in a
17
generalized DID setting via two steps. The first step consists in creating a separate dataset for each
year of treatment that contains observations of vehicles that were treated that year, with all others
that are not treated at any time serving as “clean controls”. For each dataset, I use the year of
treatment as the demarcation point to identify the three pre-treatment and post-treatment periods
for both treated vehicles and control vehicles, respectively. I then stack the treatment-time-specific
datasets together to generate the complete data for the main analysis. The second step is to perform
the DID analysis on the stacked data using the following specifications.
To estimate the treatment effect of localization, I use Panel A and model vehicle quality as
described below:
*!"$ = +,$!$ + &"$ + ' ## -./0123042.5!" + 6!" 7!" + )!"$ 8, (1)
where *!"$ is a vector of outcome variables including the number of recalls and the number of
customer complaints; i indexes the vehicle; t indexes the model year; and g indexes the treatment
year. The link function, +, denotes a Poisson regression considering that both outcome variables
are count variables. $!$ and &"$ are the vehicle and time fixed effects for treatment year g,
respectively. -./0123042.5!" is a dichotomous variable that is equal to 1 if the vehicle has had its
production localized from a foreign to a domestic plant, and is 0 otherwise. ' ## is the estimate of
an average treatment effect across all treatment years. 7!" is a vector of time-varying covariates,
including the annual sales volume and cumulative sales volume since a vehicle’s first generation,
which controls for economies of scale (Ball et al., 2018; Shah et al., 2017) and automakers’
learning experiences in addressing quality concerns (Haunschild and Rhee, 2004).
I also investigate how the introduction of localized vehicles influences the quality of other
vehicles produced in that plant by analyzing Panel B, estimating the following model:
*!"$ = +($!$ + &"$ + ' ## ;54<.=>/42.5!" + 6!" 7!" + )!"$ ). (2)
18
;54<.=>/42.5!" is a dichotomous variable that is equal to 1 if a localized vehicle had been
introduced to the plant of vehicle t in year i, and is 0 otherwise. The rest of the variables are the
same as those in equation (1). Details on the construction of the variables and measures are
included in the Appendix, and summary statistics are included in Table 1.2. I perform fixed effects
estimation for each model and cluster the standard errors at the vehicle level.
2.3.3. Validity of DID Approach
The prerequisite of conducting DID analysis is that treatment and control groups have parallel
underlying trends in the dependent variables. It is therefore important to examine pre-treatment
trends to ensure that vehicles in the treatment group and those in the control group have no
significant statistical differences in the number of recalls and the number of customer complaints
conditional on the controls. To do so, I examine the dynamic treatment effects by including the
leads and lags of the treatment variable instead of using the binary treatment indicator as in
equations (1) and (2). The models for the two treatments of interest take the following forms:
*!"$ = +($!$ + &"$ + ∑&%'() '% ## -./0123042.5!" %
+ 6!" 7!" + )!"$ ), (3)
*!"$ = +($!$ + &"$ + ∑&%'() '% ## ;54<.=>/42.5!" %
+ 6!" 7!" + )!"$ ), (4)
%
where -./0123042.5!" is the treatment indicator variable that is equal to 1 if the vehicle i had its
production localized from a foreign to a domestic plant τ years from year t, and is 0 otherwise,
with A = 0 representing the first year following the localization, and A = −1 denoting the first
%
year before treatment. Similarly, ;54<.=>/42.5!" is the treatment indicator variable that is equal
to 1 if the vehicle i experienced the introduction of localized vehicles in its plant τ years from year
t, and is 0 otherwise. The remainder of the variables are the same as those in equation (1). If my
models are properly identified, the '% ## should remain insignificant for A < 0 (i.e., the pre-
treatment periods).
19
Table 1.2 Summary statistics
Variables Descriptions Obs. Mean SD Min. Max.
Number of recalls within three years from the
Recalls 6,826 1.258 2.011 0 26
launch of a vehicle
Number of complaints within three years from
Complaints 6,826 20.572 52.865 0 1080
the launch of a vehicle
Indicator for the manufacturing localization of
Localization vehicle assembly from foreign to domestic 6,826 0.039 0.194 0 1
plants
Indicator for the introduction of a localized
Introduction 6,826 0.034 0.181 0 1
model in a domestic plant
Log transformation of the number of cars sold
Log (Sales) 6,826 9.513 2.469 0 13.727
for a vehicle in a year
Log (Cumulative Log transformation of the overall number of
6,826 11.650 2.162 0 16.678
Sales) cars sold for a vehicle since its launch
20
Table 1.3 presents the test results for the parallel trends assumption. The test suggests that
the parallel trends assumption is satisfied for my data prior to both treatments.
Table 1.3 Testing for parallel trends assumption
Panel A Panel B
Recalls Complaints Recalls Complaints
(1) (2) (3) (4)
-3 -3
Localization -0.296 0.127 Introduction 0.066 -0.134
(0.305) (0.167) (0.292) (0.332)
-2 -2
Localization 0.152 0.083 Introduction 0.160 -0.160
(0.163) (0.122) (0.244) (0.244)
Observations 9,304 9,980 Observations 4,977 4,635
Log-likelihood -10,885.485 -41,179.264 Log-likelihood -7,397.410 -44,517.739
Pseudo R2 0.311 0.73 Pseudo R2 0.298 0.729
Note. *** p< 0.001, ** p< 0.01, * p< 0.05, + p< 0.1. Table 1.3 reports the estimated coefficients
and their robust standard errors (in parenthesis) for the key terms in equations (3) and (4) regarding
the number of recalls and the number of customer complaints. As all coefficients are insignificant,
I conclude that the assumption of parallel trends is satisfied. The superscripts give the value of !
such that Localization-3 means three years before localization.
2.4. Results
2.4.1. Overall Effects of Localization on Vehicle Quality
The results for the treatment effect of localization on vehicle quality based on equation (1) are
captured in Table 1.4, Panel A. From columns (1) and (3) I observe that the number of recalls
increases by 68.203% after a localization, which equates to 0.552 more recall campaigns than the
average number of recall campaigns before the treatment (0.810 recall campaigns). Also, the
number of complaints increases by 56.831% after localization, which represents 9.977 more
complaints than the average number of complaints in the pre-treatment scenario (17.556
complaints). Overall, the results thus show that localization leads to a significant quality decline
for localized vehicles.
21
To assess the dynamic treatment effects of localization, I replace the binary treatment
variable (Localization) with a series of dummy variables for each year of localization (Localization
Time 1 to Time 3). These three dummy variables take on the value of 1 if the localization has been
occurring for one, two, or three years, respectively, and 0 otherwise. The results in columns (2)
and (4) of Table 1.4 show that the quality decline of localized vehicles diminishes with time and
is no longer significant after two years, indicating a temporary impact.
Table 1.4 Estimates of the treatment effects on vehicle quality
Panel A. Manufacturing localization and the quality of localized vehicles
Recalls Complaints
(1) (2) (3) (4)
Localization 0.520* 0.450*
(0.217) (0.221)
1
Localization 0.545* 0.362+
(0.229) (0.204)
Localization2 0.696** 0.590*
(0.218) (0.296)
3
Localization 0.253 0.355
(0.281) (0.275)
Post 0.007 -0.007 -0.043 -0.048
(0.021) (0.015) (0.058) (0.056)
Sales 0.633*** 0.635*** 1.013*** 1.011***
(0.068) (0.068) (0.179) (0.180)
Cumulative Sales -0.122** -0.123** -0.159** -0.158**
(0.043) (0.043) (0.050) (0.050)
Constant -3.350*** -3.331*** -5.935** -5.908**
(0.926) (0.927) (1.856) (1.878)
Observations 9,301 9,301 9,974 9,974
Log-pseudolikelihood -10,886.376 -10,883.138 -41,108.101 -41,088.269
Pseudo R2 0.310 0.311 0.730 0.730
Note. Robust standard errors in parentheses. *** p< 0.001, ** p< 0.01, * p< 0.05, + p< 0.1
22
Table 1.4 (cont’d)
Panel B. Introduction of localized vehicles in a plant and the quality of other vehicles in the plant
Recalls Complaints
(1) (2) (3) (4)
Introduction -0.050 -0.062
(0.170) (0.229)
1
Introduction 0.016 -0.216
(0.259) (0.155)
Introduction2 0.298 0.204
(0.222) (0.463)
3
Introduction -0.579+ -0.204
(0.340) (0.251)
Post 0.165 0.171 0.498** 0.498**
(0.197) (0.198) (0.155) (0.155)
Sales 0.517*** 0.519*** 1.523*** 1.524***
(0.141) (0.142) (0.152) (0.152)
Cumulative Sales -0.080** -0.080** -0.160*** -0.160***
(0.030) (0.030) (0.041) (0.041)
Constant -3.681* -3.687* -11.310*** -11.319***
(1.546) (1.549) (1.983) (1.984)
Observations 4,223 4,223 3,957 3,957
Log-pseudolikelihood -6,281.424 -6,277.903 -36,899.029 -36,887.849
Pseudo R2 0.305 0.305 0.742 0.742
Note. Robust standard errors in parentheses. *** p< 0.001, ** p< 0.01, * p< 0.05, + p< 0.1
The results for the treatment effect of introducing localized vehicles to a plant on the quality
of other vehicles in those plants are captured in Table 1.4, Panel B. As can be seen in columns (1)
and (3), there is no evidence of statistical differences across treated and untreated vehicles in the
number of recalls or the number of complaints. The time-varying treatment effects reported in
columns (2) and (4) show the same results. I, thus, conclude that introducing localized vehicles to
a plant does not influence the quality of the existing vehicles in the plant.
2.4.2 Post-Hoc Analysis
To investigate the mechanism through which the localization leads to the quality decline for
localized vehicles, I further differentiate the treatment effects of two types of localization decisions.
23
The first decision type is the localization to a new plant, which is when automakers localized a
vehicle to a plant that was opened within three years of the time of localization. Localization to a
new plant might negatively influence the vehicle quality due to the learning effect, with however
the quality expected to improve over time (G. Li and Rajagopalan, 1998). The second decision
type is localization to an existing plant, which refers to automakers’ decision to localize a vehicle
to a plant that has been operating for more than three years. Vehicles having a lower level of quality
after localization to an existing plant might suggest that the plants experienced difficulties in
technology transfer. This is especially relevant when the destination plants were built many years
ago.
The results in Table 1.5 reaffirm the quality decline triggered by both types of localization
decisions. Specifically, Localization to an Existing Plant significantly increases the number of
recalls of the localized vehicles, while the effect of Localization to a New Plant is marginal.
However, only Localization to a New Plant leads to a greater number of customer complaints.
Overall, both mechanisms considered contribute to the quality decline of localized vehicles.
2.5. Conclusion and Implications
I investigated whether the relocation of vehicle production to the target market can influence the
quality of localized vehicles and the quality of existing vehicles in those plants in which the
localized vehicles were introduced. Causal data analysis of vehicles sold in the U.S. market
indicates that localization increases the number of recalls and customer complaints for localized
car models, indicating a decline in quality. Specifically, a localized vehicle has 68.203% more
recalls and 56.831% more complaints in three years following the launch year of this vehicle. The
pattern of time-varying treatment effects shows that the increase in quality concerns is temporary
and diminishes over time. The increase of recalls is robust to the conditions of the destination
24
plants—both localization to a new plant and localization to an existing plant leads to more recalls.
Since the introduction of localized vehicles in a plant can interrupt existing plant operations,
existing vehicles in the plant, in theory, may also experience quality losses. However, I did not
find evidence that the number of recalls or complaints for existing vehicles increased after the
introduction of localized vehicles.
Table 1.5 Estimates of the treatment effects of localization on vehicle quality, differentiating the
plant history (Panel A)
Recalls Complaints
(1) (2)
Localization to an Existing Plant 0.678* 0.108
(0.340) (0.334)
Localization to a New Plant 0.305+ 0.731**
(0.182) (0.244)
Post 0.011 -0.044
(0.020) (0.058)
Sales 0.633*** 1.012***
(0.068) (0.179)
Cumulative Sales -0.122** -0.161**
(0.043) (0.049)
Constant -3.352*** -5.816**
(0.927) (1.839)
Observations 9,301 9,974
Log-pseudolikelihood -10,885.343 -41,054.386
Pseudo R2 0.311 0.730
Note. Robust standard errors in parentheses. *** p< 0.001, ** p< 0.01, * p<
0.05, + p< 0.1
While automakers do not usually disclose the cost of recalls, I aim to estimate the cost of
recall campaigns to evaluate part of the economic value lost due to an increased number of recall
campaigns for localized vehicles based on aggregated numbers. The 2016 data of the NHTSA
Recall Annual Report, which records the number of recall campaigns and the number of vehicles
affected per year, shows that there was a total of 919 recall campaigns that affected over 50.14
million vehicles. With more than $10.3 billion spent on warranty and recall accruals in 2016 (Held
25
et al., 2018), each recall campaign is estimated to cost $11.21 million and to affect an average of
54,559 vehicles, yielding a recall cost of $205.28 per vehicle. More specific cases, such as General
Motor’s 2014 recalls, suggest that the company spent $4.10 billion to recall 30.4 million cars
(Isidore, 2015), with $134.87 specified as the recall cost per vehicle. Using these estimations of
the recall cost per vehicle as a foundation, I estimate the cost per recall campaign to be between
$4.99 and $7.60 million in the 20 years between 2000 to 2019 (see Table 1.6 for further detail). A
conservative estimate that I can thus develop is that with 0.552 more recall campaign for a
localized vehicle, the associated cost can be $2.76 million; this number may however be much
larger based on the fines NHTSA can impose on automakers, the repairs needed, or products to be
replaced to correct the quality defect. It can become particularly costly when an entire battery for
an electric vehicle needs to be replaced (Isidore and Valdes-Dapena, 2021). Due to recalls
following localization jeopardizing substantial financial losses, a cautionary note is issued to
automakers to budget warranty spending and accrual with the potential quality decline after
localization. Based on the data I collected from a North American automotive manufacturer, the
warranty per car unit has an average cost of about $88. With the estimated recall cost per car unit
that I specified above, I suggest that automakers increase the budget on warranty by at least 53.261%
to prevent further operational disruptions triggered by this financial distress.
Several intriguing research avenues exist that can build on my work. For example, it is
possible that the localization of an entire (or large parts of a) supply chain may reduce recalls. Prior
literature has indicated that quality defects of vehicles increase as the distance between upstream
component suppliers and downstream assembly plants increases (Bray et al., 2019), which is likely
to happen when the manufacturing localization does not come with an extensive supply
localization. This is especially true for high-end vehicles as well as vehicles with numerous
26
subcomponents and large production volumes. In those cases, process-level changes, including
standardization, modularity, and small batch production, are helpful to mitigate the negative
impact of localization on vehicle quality.
From the perspective of regulators, closer attention to the shifts in manufacturing locations
can help reduce automotive recalls. Protocols can be developed to specify how automakers should
manage the localized vehicles and the required analysis and procedures of plant relocation—this
may mitigate the problem of quality decline identified in my research.
27
Table 1.6 Estimation of cost per recall campaign
Recall Cost per Unit Recall Cost per Unit
of Car Estimation 1 of Car Estimation 2
$134.87 $205.28
Year Number Number of Cost per Recall Cost per Recall
of Recalls Vehicles Affected Estimation 1 Estimation 2
2000 541 24,636,743 $6.14 $9.35
2001 453 13,626,263 $4.06 $6.17
2002 434 18,435,673 $5.73 $8.72
2003 527 19,062,913 $4.88 $7.43
2004 600 30,806,580 $6.92 $10.54
2005 562 18,962,510 $4.55 $6.93
2006 490 11,203,534 $3.08 $4.69
2007 587 14,816,417 $3.40 $5.18
2008 683 10,207,696 $2.02 $3.07
2009 491 16,125,894 $4.43 $6.74
2010 647 19,691,419 $4.10 $6.25
2011 597 13,612,039 $3.08 $4.68
2012 582 16,486,229 $3.82 $5.81
2013 628 20,260,042 $4.35 $6.62
2014 771 50,032,376 $8.75 $13.32
2015 862 49,863,794 $7.80 $11.87
2016 919 50,138,221 $7.36 $11.20
2017 809 30,689,022 $5.12 $7.79
2018 912 29,455,396 $4.36 $6.63
2019 881 38,583,951 $5.91 $8.99
Average $4.99 $7.60
28
APPENDIX
29
Construction of Variables and Measures
1. Dependent Variable
Our dependent variable is car quality. We consider auto recalls and complaints as two indicators
of quality issues, which we obtained from the NHTSA/DOI database.
Recalls. We measure Recalls using the number of recalls a vehicle of a specific year had within
three years since its launch. Considering the mismatch between the year an OEM officially assigns
to a vehicle-year and the time the vehicle-year enters the market, we consider the sales of all
vehicle-years to always start on August 1st of the year prior to their officially assigned year. The
three-year span during which we capture the recalls then refers to the 36 months following the
August of the prior year. For example, we consider the 2014 Chevrolet Cruz to having been sold
to consumers starting on August 1, 2013. During the following 36 months (i.e., August 1, 2013, to
July 31, 2016), General Motors issued eight recalls for this vehicle year – five from August 1,
2013, until July 31, 2014, one from August 1, 2014, until July 31, 2015, and two from August 1,
2015, to July 31, 2016. In the main analysis, we use the three-year forward measure of the number
of recalls, which captures, on average, 42.7% of the total number of recalls a vehicle experienced
during its life span.
Complaints. The measurement approach for Complaints is consistent with the approach we took
to measure recalls. Specifically, we use a three-year forward measure for this variable using the
number of complaints that a vehicle of a specific year received within a three-year span, which
starts from August 1st of the year prior to the official launch.
2. Independent Variables
Localization. Localization is a binary variable with Localization = 1 indicating the automaker
localizing the assembly of the vehicle in the given year, and Localization = 0 indicating the location
30
of the assembly not changing. We obtained this information from the Wards final assembly plant
location data for North America and verified the information using the press releases on the plant
relocation and plant opening from the automakers’ websites and other reliable resources. There
was one vehicle (MITSUBISHI OUTLANDER SPORT) that was relocated to foreign plants after
being localized. We delete it from our sample to avoid the confounding effect of multiple
localization decisions.
Introduction. Introduction is a binary variable. Specifically, Introduction = 1 indicates that a
localized vehicle was produced in the year it was introduced, Introduction = 0 indicates otherwise.
Since some vehicles were assembled in more than one plant in a given year, it is possible that the
introduction of localized vehicles only happened in one plant, while the quality impacts for “other”
vehicles manufactured in that plant could be captured in aggregate across all plants in which the
same vehicle is manufactured. This can create a problem in isolating the assessment of the impact
of localization of a vehicle on other vehicles manufactured in the plant. However, in Panel B, our
data suggests that none of the models that had multiple assembly plants experienced the
introduction of localized vehicles.
3. Control Variables
We include a number of variables in our analysis to control for heterogeneity at the vehicle level.
We also control for vehicle fixed-effects and year fixed-effects. These data were obtained from the
Wards sales data. Each of these variables are described below:
Sales. Prior literature suggests that a high level of production volume leads to an increase of
quality-related issues, with production volume capturing economies of scales (Ball et al., 2018;
Shah et al., 2017). In this study, we use the sales volume as a proxy for production volume. Sales
data are available for each vehicle year. Specifically, we aggregated the monthly data into yearly
31
data to match the measurement period of other variables. As noted above, it is common practice
in the auto industry that dealers start to sell a vehicle model before its assigned year. Therefore,
the sales record of the prior year may capture part of the sales of a given vehicle year. To align the
aggregated data with the sales practice, we made the same assumption as the one we presented in
the creation of recall variable, i.e., that the sale of a vehicle always starts on August 1st of the year
prior to their officially assigned year. As such, the yearly sales data capture the number of vehicles
sold during the following 12 months. For example, the sales of the 2010 Honda Accord refer to
the volume sold from August 1, 2019, until July 31, 2020. We performed a log transformation for
this variable before including it into the analysis.
Cum Sales. Cumulative production volume is also related to quality issues, such as recalls or
complaints, because it measures the extent to which manufacturers can learn from the experience
to reduce quality concerns (Haunschild & Rhee, 2004). We use the cumulative sales volume as a
proxy of cumulative production volume. We construct the variable by calculating the number of
vehicles sold for a vehicle from its launch, or from the earliest record in our sample until the last
year of record. We performed a log transformation for the one-year lagged variable before
including it in the analysis.
32
REFERENCES
33
REFERENCES
Badiru, A. B. (1998). Quality Improvement through Learning Curve Analysis. In Handbook of
Total Quality Management (pp. 87–107). Springer US.
Bailey, D., and De Propris, L. (2014). Manufacturing Reshoring and Its Limits: The UK
Automotive Case. Cambridge Journal of Regions, Economy and Society, 7(3), 379–395.
Baker, A. C., Larcker, D. F., and Wang, C. C. Y. (2022). How Much Should We Trust Staggered
Difference-in-Differences Estimates? Journal of Financial Economics, 144(2), 370–395.
Ball, G. P., Shah, R., and Wowak, K. D. (2018). Product Competition, Managerial Discretion,
and Manufacturing Recalls in the U.S. Pharmaceutical Industry. Journal of Operations
Management, 58–59(1), 59–72.
Bandyopadhyay, J. K., and Jenicke, L. O. (2007). Six Sigma Approach to Quality Assurance in
Global Supply Chains: A Study of United States Automakers. International Journal of
Management, 24(1), 101–107.
BDO. (2021). 2021 BDO manufacturing CFO outlook survey.
https://www.bdo.com/BDO/media/CFO-Outlook-Survey/IND_2021-Manufacturing-CFO-
Outlook-Survey_WEB.pdf.
Ben-Ner, A., and Siemsen, E. (2017). Decentralization and Localization of Production: The
Organizational and Economic Consequences of Additive Manufacturing. California
Management Review, 59(2), 5–23.
Bock, S. (2008). Supporting Offshoring and Nearshoring Decisions for Mass Customization
Manufacturing Processes. European Journal of Operational Research, 184(2), 490–508.
Bray, R. L., Serpa, J. C., and Colak, A. (2019). Supply Chain Proximity and Product Quality.
Management Science, 65(9), 4079–4099.
Brennan, L., Ferdows, K., Godsell, J., Golini, R., Keegan, R., Kinkel, S., Srai, J. S., and Taylor,
M. (2015). Manufacturing in the World: Where Next? International Journal of Operations
and Production Management, 35(9), 1253–1274.
Brun, A., and Zorzini, M. (2009). Evaluation of Product Customization Strategies through
Modularization and Postponement. International Journal of Production Economics, 120(1),
205–220.
Callaway, B., and Sant’Anna, P. H. C. (2021). Difference-in-Differences with Multiple Time PP.
Journal of Econometrics, 225(2), 200–230.
Cassia, F. (2020). ‘Manufacturing Is Coming Home’: Does Reshoring Improve Perceived
Product Quality? The TQM Journal, 32(6), 1099–1113.
34
Cengiz, D., Dube, A., Lindner, A., and Zipperer, B. (2019). The Effect of Minimum Wages on
Low-Wage Jobs*. The Quarterly Journal of Economics, 134(3), 1405–1454.
Cheng, Y., Farooq, S., and Johansen, J. (2015). International Manufacturing Network: Past,
Present, and Future. International Journal of Operations and Production Management,
35(3), 392–429.
Cherney, M. (2020, December 27). Firms Want to Adjust Supply Chains Post-Pandemic, but
Changes Take Time. Wall Street Journal (Online).
http://ezproxy.msu.edu/login?url=https://www.proquest.com/newspapers/firms-want-adjust-
supply-chains-post-pandemic/docview/2472846898/se-2?accountid=12598.
Clark, K. B., and Fujimoto., T. (1992). Product Development Performance: Strategy,
Organization, and Management in the World Auto Industry. Harvard Business School Press.
CNN. (2015). Chinese-owned Volvo to open its first U.S. car plant.
https://money.cnn.com/2015/03/30/news/companies/volvo-new-factory-u-s-/index.html.
Curran, E. (2021, May 21). APEC Sees Challenges in Reshoring as Pandemic Hits Supply
Chains. Bloomberg Newsletter. https://www.bloomberg.com/news/newsletters/2021-05-
27/supply-chains-latest-weighing-costs-and-benefits-of-reshoring.
da Silveira, G. J. C. (2014). An Empirical Analysis of Manufacturing Competitive Factors and
Offshoring. International Journal of Production Economics, 150, 163–173.
Dana, L. P., Hamilton, R. T., and Pauwels, B. (2007). Evaluating Offshore and Domestic
Production in the Apparel Industry: The Small Firm’s Perspective. Journal of International
Entrepreneurship, 5(3–4), 47–63.
de Treville, S., Bicer, I., Chavez-Demoulin, V., Hagspiel, V., Schürhoff, N., Tasserit, C., and
Wager, S. (2014). Valuing Lead Time. Journal of Operations Management, 32(6), 337–346.
Dunning, J. H. (1980). Toward an Eclectic Theory of International Production: Some Empirical
Tests. Journal of International Business Studies, 11(1), 9–31.
Ellram, L. M., Tate, W. L., and Petersen, K. J. (2013). Offshoring and Reshoring: An Update on
the Manufacturing Location Decision. Journal of Supply Chain Management, 49(2), 14–22.
Foerstl, K., Kirchoff, J. F., and Bals, L. (2016). Reshoring and Insourcing: Drivers and Future
Research Directions. International Journal of Physical Distribution & Logistics
Management, 46(5), 492–515.
Foldy, B. (2020, April 2). Auto-Parts Suppliers Teeter as Car Production Halts. The Wall Street
Journal. https://www.wsj.com/articles/auto-parts-suppliers-teeter-as-car-production-halts-
11585828803.
Fratocchi, L., Ancarani, A., Barbieri, P., Di Mauro, C., Nassimbeni, G., Sartor, M., Vignoli, M.,
and Zanoni, A. (2016). Motivations of Manufacturing Reshoring: An Interpretative
35
Framework. International Journal of Physical Distribution & Logistics Management, 46(2),
98–127.
Goodman-Bacon, A. (2021). Difference-in-Differences With Variation in Treatment Timing.
Journal of Econometrics, 225(2), 254–277.
Gopal, A., Goyal, M., Netessine, S., and Reindorp, M. (2013). The Impact of New Product
Introduction on Plant Productivity in the North American Automotive Industry.
Management Science, 59(10), 2217–2236.
Grappi, S., Romani, S., and Bagozzi, R. P. (2020). The Effects of Reshoring Decisions on
Employees. Personnel Review, 49(6), 1254–1268.
Gray, J. V., Roth, A. V., and Leiblein, M. J. (2011). Quality Risk in Offshore Manufacturing:
Evidence from the Pharmaceutical Industry. Journal of Operations Management, 29(7–8),
737–752.
Gray, J. V., Skowronski, K., Esenduran, G., and Johnny Rungtusanatham, M. (2013). The
Reshoring Phenomenon: What Supply Chain Academics Ought to Know and Should Do.
Journal of Supply Chain Management, 49(2), 27–33.
Gryta, T., and Cutter, C. (2021, November 1). Farewell Offshoring, Outsourcing. Pandemic
Rewrites CEO Playbook. Wall Street Journal (Online).
http://ezproxy.msu.edu/login?url=https://www.proquest.com/newspapers/farewell-
offshoring-outsourcing-pandemic-rewrites/docview/2590139909/se-2?accountid=12598.
Haex, P., and Buck, R. (2022). Global reshoring & footprint strategy.
https://bciglobal.com/uploads/9/artikelen/global-reshoring-and-footprint-strategy-2022.pdf.
Haunschild, P. R., and Rhee, M. (2004). The Role of Volition in Organizational Learning: The
Case of Automotive Product Recalls. Management Science, 50(11), 1545–1560.
Held, M., Marian, A., and Reaves, J. (2018). The auto industry’s growing recall problem—and
how to fix it.
https://www.alixpartners.com/media/14438/ap_auto_industry_recall_problem_jan_2018.pdf
HMMA. (2006). Hyundai celebrates grand opening of its first US plant - Hyundai Motor
Manufacturing Alabama, LLC (HMMA). https://www.hmmausa.com/hyundai-celebrates-
grand-opening-of-its-first-us-plant/.
Honda. (2006, September 26). 2007 Honda CR-V. U.S. Production. https://hondanews.com/en-
US/photos/photo-547a298d6a2fe6a21f1369004c350649-2007-honda-cr-v-u-s-
production?firstResultIndex=40&channelsConstraint=channel-3012.
Imai, K., and Kim, I. S. (2021). On the Use of Two-Way Fixed Effects Regression Models for
Causal Inference with Panel Data. Political Analysis, 29(3), 405–415.
Isidore, C. (2015, February 4). GM’s total recall cost: $4.1 billion. CNNMoney.
36
https://money.cnn.com/2015/02/04/news/companies/gm-earnings-recall-costs/index.html.
Isidore, C., and Valdes-Dapena, P. (2021, February 25). Hyundai’s Recall of 82,000 Electric
Cars Is One of the Most Expensive in History. CNN Business.
https://www.cnn.com/2021/02/25/tech/hyundai-ev-recall/index.html.
ITA. (2018). Automotive Industry Spotlight. https://www.selectusa.gov/automotive-industry-
united-states.
Johansson, M., and Olhager, J. (2018). Comparing Offshoring and Backshoring: The Role of
Manufacturing Site Location Factors and Their Impact on Post-Relocation Performance.
International Journal of Production Economics, 205, 37–46.
Ketokivi, M., Turkulainen, V., Seppälä, T., Rouvinen, P., and Ali-Yrkkö, J. (2017). Why Locate
Manufacturing in a High-Cost Country? A Case Study of 35 Production Location
Decisions. Journal of Operations Management, 49–51(1), 20–30.
Lacetera, N., and Sydnor, J. (2015). Would You Buy a Honda Made in the United States? The
Impact of Production Location on Manufacturing Quality. Review of Economics and
Statistics, 97(4), 855–876.
Li, G., and Rajagopalan, S. (1998). Process Improvement, Quality, and Learning Effects.
Management Science, 44(11 PART 1), 1517–1532.
Ma, C., Hauck, M., and Matava, D. (2021). State of North America manufacturing 2021 annual
report. https://f.hubspotusercontent00.net/hubfs/242200/UA Files/State of North American
Manufacturing 2021 Annual Report v1.3.pdf.
Moradlou, H., Backhouse, C., and Ranganathan, R. (2017). Responsiveness, the Primary Reason
behind Re-Shoring Manufacturing Activities to the UK: An Indian Industry Perspective.
International Journal of Physical Distribution and Logistics Management, 47(2–3), 222–
236.
NielsonIQ. (2020). COVID-19 concerns are a likely tipping point for local brand growth.
https://nielseniq.com/global/en/insights/analysis/2020/covid-19-concerns-are-a-likely-
tipping-point-for-local-brand-growth/.
Raza, W., Grumiller, J., Grohs, H., Essletzbichler, J., and Pintar, N. (2021). Post Covid-19 value
chains: options for reshoring production back to Europe in a globalised economy (Issue
March). https://www.europarl.europa.eu/thinktank/en/document/EXPO_STU(2021)653626.
https://doi.org/10.2861/118324
Reuters. (2016). Volkswagen targets East Africa with Kenya car assembly plant.
https://www.reuters.com/article/us-volkswagen-kenya/volkswagen-targets-east-africa-with-
kenya-car-assembly-plant-idUSKCN11D1YM.
Rosenbaum, E. (2021, December 2). CEOs across Economy Agree on One 2022 Prediction:
More Volatility, No End to Covid. CNBC. https://www.cnbc.com/2021/12/02/ceos-across-
37
economy-agree-on-one-big-2022-prediction-more-volatility.html.
Schwartz, N. D. (2022, February 18). Supply Chain Woes Prompt a New Push to Revive U.S.
Factories. International New York Times. https://link-gale-
com.proxy1.cl.msu.edu/apps/doc/A688930535/GIC?u=msu_main&sid=bookmark-
GIC&xid=e0e2c2cc.
Shah, R., Ball, G. P., and Netessine, S. (2017). Plant Operations and Product Recalls in the
Automotive Industry: An Empirical Investigation. Management Science, 63(8), 2439–2459.
Simchi-Levi, D., and Simchi-Levi, E. (2020). Building Resilient Supply Chains Won’t Be Easy.
Harvard Business Review.
Srai, J. S., and Ané, C. (2016). Institutional and Strategic Operations Perspectives on
Manufacturing Reshoring. International Journal of Production Research, 54(23), 7193–
7211.
Staeblein, T., and Aoki, K. (2015). Planning and Scheduling in the Automotive Industry: A
Comparison of Industrial Practice at German and Japanese Makers. International Journal of
Production Economics, 162, 258–272.
Stentoft, J., Mikkelsen, O. S., Jensen, J. K., and Rajkumar, C. (2018). Performance Outcomes of
Offshoring, Backshoring and Staying at Home Manufacturing. International Journal of
Production Economics, 199, 199–208.
Sun, L., and Abraham, S. (2021). Estimating Dynamic Treatment Effects in Event Studies with
Heterogeneous Treatment Effects. Journal of Econometrics, 225(2), 175–199.
Tech Explore. (2020). Tesla to build “world’s largest” battery plant near Berlin.
https://techxplore.com/news/2020-11-tesla-world-largest-battery-berlin.html.
Tesla. (2020). Gigafactory Berlin-Brandenburg. https://www.tesla.com/gigafactory-berlin.
The White House. (2021). Building resilient supply chains, revitalizing American manufacturing,
and fostering broad-based growth. https://www.whitehouse.gov/wp-
content/uploads/2021/06/100-day-supply-chain-review-report.pdf.
TheEconomist. (2013, January 17). Here, There and Everywhere. The Economist, 406(8819).
Vahlne, J. E., and Johanson, J. (2017). From Internationalization to Evolution: The Uppsala
Model at 40 Years. Journal of International Business Studies, 48(9), 1087–1102.
Vernon, R. (1966). International Investment and International Trade in the Product Cycle. The
Quarterly Journal of Economics, 80(2), 190.
Wellener, P., Hardin, K., and Beckoff, D. (2022). 2022 manufacturing industry outlook.
https://www2.deloitte.com/content/dam/Deloitte/us/Documents/energy-resources/us-2021-
manufacturing-industry-outlook.pdf.
38
Wiesmann, B., Snoei, J. R., Hilletofth, P., and Eriksson, D. (2017). Drivers and Barriers to
Reshoring: A Literature Review on Offshoring in Reverse. European Business Review,
29(1), 15–42.
39
CHAPTER 3 – Firms' Policy Engagement on Climate Change: Taxonomy Development and
Validation
3.1. Introduction
With the increasing awareness of climate change as a pressing sustainability challenge and a major
global issue in the last few decades, the governments of many countries and regions have
formulated and enacted climate change policies to restrict and reduce greenhouse gas emissions.
Up to July 2022, the International Energy Agency’s (IEA, 2022) policies database recorded 6,782
regulations, policies, and measures related to the reduction of carbon emission, the improvement
of energy efficiency, and the development and deployment of renewable and other clean energy
technologies that are in force in 37 countries. The World Bank’s Carbon Pricing Dashboard (The
World Bank, 2022) has tracked down 68 carbon pricing initiatives, one of the key economic policy
instruments, that were implemented or scheduled for implementation in 46 nations and 36
subnational areas, which covers 23% of global greenhouse gas emissions. Many jurisdictions have
adopted renewable energy targets, led by the EU’s binding target that 27% of the overall energy
consumption of the EU should come from renewable sources by 2030 (EEU, 2018), and the U.S.’s
state renewable portfolio standards that require a different degree of clean energy consumption in
different states (RSP, 2021).
Those emerging public policies or policy changes exert regulatory and policy pressures on
firms and have urged businesses to take action. On the one hand, the increased scrutiny motivates
firms to improve policy compliance and commit to climate change mitigation and adaptation.
Consequently, firms have adopted a variety of approaches to transform or tweak internal
operations and engage in sustainable supply chain management to minimize the climate impact.
Investigations of those business practices led to a stream of climate change-related research in the
40
operations and supply chain management (OSCM) field, which has identified and discussed
approaches that firms incorporate in procurement and sourcing strategies, manufacturing
operations, and logistic and transportation management to make or enhance their commitment to
the reduction of carbon emissions (Ghadge et al., 2020), including supply chain collaboration and
coordination for carbon transparency and other engagements (Jira and Toffel, 2013; Theißen et al.,
2014; Tidy et al., 2016; Villena and Dhanorkar, 2020), carbon footprint mapping (K.-H. Lee, 2011;
Rizet et al., 2012), low-carbon product and packaging design (Ji et al., 2014; Oglethorpe and Heron,
2010), low-carbon manufacturing strategies (Dadhich et al., 2015; Oglethorpe and Heron, 2010),
and low-carbon supply chain strategies (C.-M. Chen, 2017; Jin et al., 2014). These studies note
climate change policies as an external context and concentrate on how firms survive the climate
change scrutiny by adopting sustainable OSCM.
On the other hand, a context of strict enforcement of climate change policies prompts firms
to reversely influence policymaking to redirect or evade regulatory scrutiny. This effort has been
evident especially when emerging policies influence the dynamics of an industry and firms
residing in an industry (Grover and Dresner, 2022). As an example, consider automakers’ varying
attempts to reshape the action plans of the targets to gradually phase out sales of new internal
combustion engine (ICE) vehicles to achieve zero-emission by 2050 at the latest, which have been
specified in 17 national or regional governments’ official policies or strategy documents
(Automotive World, 2020). While General Motors, Ford, and several other automakers commit to
a carbon-neutral stance and advocate for a more aggressive goal of eliminating gas and diesel
vehicles by 2035 (Boudette and Davenport, 2021), the remaining automakers led by BMW lobby
against the pressing enforcement of ICE bans (Hetzner, 2021). As another example serves the
coexisting support and resistance in the oil and gas industry to policies related to decarbonization
41
(Domonoske, 2021; McCarthy, 2019), which are believed to be the unavoidable but costly path
energy companies should take to achieve the goal of The Paris Agreement (Pee et al., 2018).
In contrast to reactively responding to policy changes with improved compliance and
adaptation, firms engaging in influence actions seek to reshape or amend the policy and regulatory
environments in which they embed through close interactions with government departments,
participation in ad-hoc coalitions, lobbying, and imposing market influences (Clapp and Meckling,
2013). Such engagement has been captured by the literature as part of corporate political activities
(CPAs) (Bumpus, 2015; Eberlein and Matten, 2009; Okereke and Russel, 2010; Oliver and
Holzinger, 2008). However, these issues are understudied from an OSCM perspective.
One exception is a recent conceptual study by Grover and Dresner (2022). They extend the
typology of political actions proposed by management researchers (Oliver and Holzinger, 2008),
and discuss the applicability of influence-oriented political actions for supply chain risk
management. Specific to the context of climate change, only Cory et al.'s (2021) research justifies
the relevance of OSCM to firms’ attempts to influence policymaking. They empirically
demonstrate that firms’ embeddedness in the supply chain ecosystem motivates firms to reshape
climate change policies in favor of the less sustainable segments of their supply chain. Specifically,
the greenhouse gas intensity of the extended supply chain is positively associated with the focal
firms’ participation in carbon coalitions and lobbying.
The scarcity of discussions on firms’ engagement in influencing climate change policies
(EICCP) is due to the unbalance in climate change-related research – most existing studies have
regarded climate change policies as external conditions to which firms should adapt, while very
few studies have paid attention to the impact of firms’ political actions on climate change policies
(Aragón-Correa et al., 2020; Greiner and Kim, 2021). Yet, the investigation of how firms reshape
42
the regulatory context to solve the problem of practice-policy decoupling and obtain legitimacy
under scrutiny over climate impact is important to further the understanding of firms’ overall
sustainability strategy. Specifically, identifying and examining different strategies of EICCP sets
a foundation for the exploration of the dynamics between public policies and sustainable OSCM,
responding to the call for more investigation on policy issues from an OSCM perspective (Tokar
and Swink, 2019). Further, prior literature indicates that the discourse of sustainability has been
marginalized in OSCM to align with firms’ intentions to make incremental changes instead of a
fundamental shift in practices as well as their concentration on profitability over sustainability
(Hardy et al., 2020; Pagell and Shevchenko, 2014). I contend that the investigation of firms’
EICCP provides insights into the role of business in the formation of the current discourse of
sustainability in the OSCM context and facilitates the revamping of the discourse that leads to
more sustainable supply chains. Overall, there is a need to complement climate change-related
OSCM studies by incorporating the conceptualization of EICCP and identifying different types of
engagements. Within this context, I specifically focus on two research questions: 1) Which criteria
need to be considered to differentiate EICCP strategies? and 2) What kind of EICCP strategies do
firms adopt? How can these strategies be captured?
To answer those questions, I first introduce the concept of EICCP. Building on the
literature on CPA and environmental politics, I also propose a typology for EICCP that
differentiates firms’ engagement strategies from three dimensions: firms’ value perspective and
engagement level, firms’ participation level in EICCP, and the type of resources devoted to EICCP.
Second, I validate the typology by identifying strategies of EICCP adopted in the real world by
performing text analytics with automated machine-learning techniques on firms’ self-disclosure in
CDP Climate Change data. I also evaluate the measures and discuss future research opportunities.
43
3.2. Literature Review and Theory Development for EICCP
3.2.1. Literature Review
Researchers in economics, political science, and management have long recognized the business
responses to government regulations, which is referred to as CPA. At the micro or firm level, CPA
has been defined as “strategies to employ an organization’s resources to integrate objectives and
to undertake coherent actions directed towards the political, social, and legal environment to secure
either permanent or temporary advantage and influence over other actors in the process” (Mahon,
1983, pp. 51-52). From a managerial perspective, prior literature posits that firms engage in the
public policy process to strengthen their organizational legitimacy in a broader social system as
well as to obtain competitive advantages over their competitors (Shaffer, 1995). Due to the
heterogeneity in political resources and capabilities, firms perceive the impact of policies
differently (Hillman et al., 2004). Consequently, firms choose different ways to engage in policies,
with expectations of investments and potential gains (Bonardi et al., 2005). The various ways to
engage mainly depend on the structure of firms and industries, the characteristics of different
political issues, and the institutional features (Hillman et al., 2004; Lawton et al., 2013).
The multiplicity of firms’ engagement strategies enlightens a stream of literature that
focuses on developing a taxonomy of CPA to further the understanding of CPA types. With the
initial attempts only being to differentiate proactive behaviors from reactive behaviors (Blumentritt,
2003; Meznar and Nigh, 1995), later research proposes various criteria seeking to develop a more
holistic taxonomy following different theoretical underpinnings. Representative work includes
Hillman and Hitt (1999), which categorizes CPA engagement strategies based on firms’
approaches to CPA (relational vs. transactional engagement), participation level (engage as a
leader vs. as a follower), and fundamental resources exchanged (information, financial incentives
44
vs. constituency building). The authors apply resources dependence theory and institutional theory
to consider the firms’ resources or resource constraints, as well as the institutional differences at
the country- or subnational-level as important factors for political strategy formulation. Dahan
(2005) extends the typology of resources exchanged by proposing other political resources used
for CPA. Building on the resource-based view and dynamic capabilities view, Oliver and
Holzinger (2008) propose that the effectiveness of firms’ CPA strategies differ due to the different
dynamic capabilities that firms have developed in the political environment. To achieve effective
political management, firms ground the CPA strategies in their value perspectives, which delineate
whether firms aim to maintain or create value, and strategic orientation, which suggests whether
firms need to influence or comply with policies. Integrating the two criteria yields a typology that
is comprised of four strategies: reactive (value maintenance with compliance), anticipatory (value
creation with compliance), defensive (value maintenance with influence), and proactive strategies
(value creation with influence). Grover and Dresner (2022), presenting an integrated model of
CPA and supply chain risk management strategies, extend Oliver and Holzinger’s (2008)
taxonomy specifying a competitive dynamics perspective. They argue that there exist two
additional strategic orientations besides compliance and influence. One is moderation, which refers
to firms “acting to moderate their political environments to improve or defend their competitive
advantage” (Grover and Dresner, 2022, pp. 53). The other is neutral, which describes firms’
attempt to “adopt a free-riding political stance or submit to the political environment” to maintain
competitive advantage (Grover and Dresner, 2022, pp. 53).
Discussions on general types of CPA provide a theoretical foundation for the further
exploration of firms’ political engagement. However, these taxonomies may have limited
explanatory power if focused on a specific context such as climate change. This is because the
45
ways firms engage in policymaking depend on the issues and the stages of the policymaking
process (Hillman et al., 2004; Keim and Zeithaml, 1986). For example, prior literature contends
that the types and intensity of firms’ CPA are associated with the degree of saliency for a political
issue, which refers to the importance that individuals such as voters place on a certain issue (Keim
and Bonardi, 2005; Moniz and Wlezien, 2020). While issues such as healthcare, foreign policy,
and abortion, have different levels of saliency (Doherty et al., 2020), firms can adopt different
CPA strategies to respond to policy changes on those issues. Therefore, there is a need to provide
issue-specific discussions when exploring the potential types of CPA engagements. This study
contributes to this discussion by investigating firms’ different engagement strategies in climate
change-related policymaking.
In addition, while the existing typologies examine several general types of CPA strategies,
I further investigate the various engagement efforts within a certain strategy type to further the
understanding of firms’ responses to climate change policies. This is especially necessary when
the exploration of different strategy types are unbalanced, with the influence-oriented engagement
in policymaking for climate change issues receiving less academic attention than the adaptation-
oriented engagement (Aragón-Correa et al., 2020; Greiner and Kim, 2021). Thus, an examination
of the distinct approaches that firms adopt to influence climate change policies is needed.
Accordingly, this essay contributes to the development of a holistic framework of business
responses to the climate change issue and develops a taxonomy specific to EICCP.
3.2.2. Typologies of EICCP strategies
I elaborate on the literature on general CPA to develop the concept of EICCP. Specifically, I define
EICCP as strategic actions firms perform to influence climate change policymaking processes,
aiming at shaping policies or promoting policy changes in favor of their interests. Since EICCP
46
focuses on influence, it falls into the influence-type of political actions that Oliver and Holzinger
(2008) and Grover and Dresner (2022) define. Although extant literature has not studied the types
of EICCP strategies, I found studies that focus on business engagement in environmental public
policies relevant given that environmental policies embrace climate change policies. My literature
review yields two typologies of firms’ engagement strategies in environmental policies.
The first taxonomy by Tienhaara (2013) categorizes firms’ engagement strategies based on
different forms of corporate power firms exercise in the engagement process: structural power,
instrumental power, discursive power, and institutional power. Firms that execute structural power
move business activities away from the regulated regions, which seek to implicitly influence
policymaking by imposing economic sanctions on the region. Instrumental power allows firms to
showcase their organizational strength in expertise or resources to shape policy-related decision-
making. Discursive power enhances firms’ political legitimacy and helps firms couch
policymakers’ preferences through constituency building or research funding. Firms imposing
institutional power possess global corporations’ ability to shift environmental issues from
regulatory institutions to enabling institutions of trade and investment to avert regulatory risks.
Different dimensions of corporate power also inspire the classification of engagement in
environmental policies by influence channels corporate power can wield. Clapp and Meckling
(2013), for instance, identify and differentiate lobbying, the execution of market influence, rule-
setting practices, and issue-framing activities as four distinct ways for firms to engage in
environmental policymaking. Although the classification based on firms’ power over regulatory
agencies provides rationales for the implementation of several political activities including evasion,
lobbying, and political advertising, it has also drawbacks in supporting the follow-up empirical
investigation. This is specifically due to the differences among the proposed forms of power being
47
obscure (Barnett and Duvall, 2004). The inconsistent interpretation of power and the inherent
interrelation among distinct forms of power have therefore resulted in different classification
models and the increased difficulty of model applications (Tienhaara, 2013).
The second taxonomy by Meckling (2015) proposes four engagement strategies that are
more exclusive from the other types. Combining regulatory pressure and distributional effect as
the basis of classification, this typology identifies opposition, hedging, support, and non-
participation as four types of business strategies firms take when engaging in environmental
policies. Regulatory pressure derives from firms’ interpretation of “its multi-layered institutional
environment and its mixed signals with regard to demand for regulatory action on a given
environmental issue” (Meckling, 2015, p. 22), which considers the impact of external factors on
firms’ decision-making. The distributional effect captures the cost versus the benefits of the
engagement considering "firms’ heterogeneity due to market position and technology portfolio”
(Meckling, 2015, p. 21). Since this classification method concentrates on firms’ perceptions of
external and internal risks associated with influencing environmental policymaking, I can measure
the engagement strategies using firms’ self-disclosure on environmental policy engagement as the
source of data. Also, the classification criteria are compatible with the general typologies of CPA
by Oliver and Holzinger (2008) and Grover and Dresner (2022) to further the understanding of
influence-oriented engagement, which is the focus of this study. Among the variety of typologies
of corporate political actions, Oliver and Holzinger’s (2008) model builds on the dynamic
capabilities perspective, which is a crucial theory in OSCM research, to explain the motives of
engagement strategies. Specifically, firms will take the perspective of value maintenance if
protecting the established value base delivers more benefits to them than adapting to policy
changes. In contrast, they will adopt the perspective of value creation if they can promote
48
regulatory advancements to raise the value of their resources. The influence actions under the
perspectives of value maintenance and value creation follow a defensive and a proactive strategy,
respectively. Grover and Dresner (2022) extend Oliver and Holzinger’s (2008) model specifying
a competitive dynamics perspective and developing an integrated model of political actions and
supply chain strategies. In this typology, firms that seek to defend their competitive position
embrace a defensive influence strategy while firms aiming at improving their competitive position
choose a proactive influence strategy. Although elaborating on a different theoretical concept, the
essence of obtaining positive distributive effects and achieving a favorable competitive position is
to obtain more output than input by engaging in policymaking processes. Therefore, I conclude
that Meckling (2015) shares the theoretical emphasis on the value of engagements with Oliver and
Holzinger (2008), with Grover and Dresner (2022) providing additional insights on the regulatory
conditions.
The compatibility between the typologies of Meckling (2015), Oliver and Holzinger (2008),
and Grover and Dresner (2022) allows me to integrate them to develop a taxonomy for EICCP.
Building on these typologies, I further propose to consider firms’ resources as criteria to classify
influence-oriented engagement strategies in the context of climate change following the resource-
based view. The resulting taxonomy for EICCP is based on three factors: firms’ value perspective
and engagement level, the level of participation, and resources devoted to EICCP, as shown in
Figure 2.1.
49
Figure 2.1 Taxonomy of EICCP
Criteria
Engagement level: High Hedging
Value perspective:
Strategic Value maintenance
decisions Engagement level: Low Opposition
based on
external Engagement level: High Promotion
conditions Value perspective:
Value creation
Engagement level: Low Support
Individual
Participation level
Strategic
Collective
decisions
based on
internal Knowledge resources
resources
Types of resources devoted
Financial resources
I select these factors as the classification criteria because the engagement strategies I
identify through this taxonomy have different implications on firms’ performance and competitive
advantage, which is relevant to OSCM research. Although the examination of performance
outcomes of EICCP strategies is beyond the scope of this study, I will briefly discuss those
implications to inspire future research.
First, EICCP can be motivated by different value perspectives and subsequently classified
into different engagement levels. The value perspective depicts firms’ motives for influencing
policy processes. Following Oliver and Holzinger (2008) and Grover and Dresner (2022), I
differentiate the value maintenance perspective from the value creation perspective. The value
maintenance perspective reflects firms’ intention to maintain the status quo to protect their value
50
base when they currently occupy an advantageous competitive position. In contrast, the value
creation perspective consists of firms’ motives for changing the status quo to create value by
confronting or anticipating climate policy changes that undermine the firms’ competitive
advantage. Further, I propose to divide EICCP strategies under each value perspective into two
engagement levels. I derive the factor of the engagement level from Meckling’s (2015) typology
that considers firms’ perceived regulatory pressures. Specifically, I posit that firms’ engagement
level reflects firms’ willingness to engage considering external conditions that include regulatory
pressures. A high engagement level is associated with a higher level of scrutiny that firms perceive,
while a low engagement level is related to a lower level of scrutiny.
Combining value perspectives and engagement levels, I classify EICCP strategies into four
categories: opposition, hedging, support, and promotion strategies. The value maintenance
perspective with a low engagement level leads to an opposition strategy, with which firms negate
the eligibility of certain climate change policies or regulatory initiatives and seek to veto them.
The value maintenance perspective with a high engagement level leads to the hedging strategy,
with which firms seek to level the pressure of policy development and compliance across a global
industry by challenging the application scope of certain climate change policies or regulatory
initiatives. The engagement level in the hedging strategy is relatively higher than in the opposition
strategy because the hedging strategy includes a mixture of defensive and proactive actions to
strategically accommodate the political demand for climate change and make a self-interested
contribution to climate change policies at the same time. In addition, the opposition strategy
usually concentrates on a single or a small set of policies while the hedging strategy focuses on
regulatory changes that influence the competitive dynamics of the international market. Serving
as an example is the utility companies’ advocacy to establish a comprehensive renewable energy
51
policy that applies equally to all energy sectors worldwide. Instead of directly opposing the climate
policies that affect their business, these utility companies propose a more complicated regulatory
option, which will prolong the policymaking process. At the same time, this regulatory option
allows them to maintain competitive advantages by advocating the same or even more stringent
regulations on their global competitors (Meckling, 2015).
The value creation perspective with a low engagement level results in the support strategy,
with which firms participate in climate change policymaking to promote policy compliance
without pursuing radical changes in the policy. Common practices of support strategies include
participation in climate change-related workshops, voluntary disclosure, communication, and
information sharing with other entities about the reduction of carbon emissions and other climate
change impacts. The value creation perspective with a high engagement level yields the promotion
strategy, with which firms seek to advance climate change policymaking by proposing more
stringent policy schemes, setting standards or measures for further policy development, enhancing
incentives, or using specific technologies to expand their regulatory influence. The core of the
promotion strategy is that firms attempt to redefine the current policies or their legitimacy (Oliver
and Holzinger, 2008).
The four engagement strategies described above have different implications for the
sustainability of firms’ competitive advantage. Specific to the criterion of value perspectives, firms
with a value maintenance motive are more defensive to environmental changes. However, such
changes cannot be avoided in the long run. Therefore, the competitive advantage of those firms
will be undermined over time. In comparison, firms with a value creation motive are proactively
shaping external environments to leverage the firm’s strengths and interests. Therefore, firms
employing the promotion and support strategies have more potential in developing sustainable
52
competitive advantage than firms relying on opposition and hedging strategies, since the
perspective of value creation reflects firms’ intention and capabilities to develop new
competencies that accommodate new environmental conditions. Within the value creation
strategies, the promotion strategy leads to more sustainable competitive advantages than the
support strategy. This is because firms adopting the promotion strategy, through reshaping climate
change policies, develop superior capabilities to combat climate change, when compared to firms
that adopt a support strategy, and can thus obtain more favorable regulatory conditions that fit the
firms’ strengths. In comparison, firms with a support strategy need to obtain competitive
advantages by outperforming their competitors and creating a first mover advantage.
While the value perspective and the engagement levels capture firms’ strategic decisions
based on external conditions, I propose other criteria to categorize EICCP strategies considering
firms’ internal resources. Prior literature has indicated the critical role of firms’ internal resources
in promoting their environmental strategies (Menguc et al., 2010; Paulraj, 2011). As such, the
extent to which firms pursue environmental strategies depends on the availability of firm-specific
resources (Lee et al., 2018). Specific to the climate change issue and extending firms’ efforts from
adaptation-oriented strategies to influence-oriented strategies, I argue that the heterogeneity in
firms’ internal resources can also lead to different EICCP strategies. In my taxonomy, I consider
two aspects of internal resources: resource availability, which is reflected by firms’ level of
participation in EICCP, and the types of resources firms devote to EICCP, following Hillman and
Hitt (1999), who incorporate resources into their framework of proactive corporate political
strategy as decision variables.
Specifically, first, capturing resource availability, firms’ EICCP can fall into two levels of
participation, the individual level, and the collective level. EICCP at the individual level means
53
that a firm directly interacts with climate change policymakers on behalf of itself, while engaging
at the collective level indicates that a firm participates in groups, such as climate change coalitions
and associations, with the group performing activities that are integral to influencing policymaking.
Hillman and Hitt (1999) argue that firms may or may not possess requisite resources for
independent actions and the ones with resource constraints tend to engage in collective actions to
consolidate resources. Building on this logic, the level of participation implies firms’ resource
availability. Specifically, firms adopting the strategy of individual engagement are likely to be
industrial leaders and possess more political resources than firms relying on collective engagement.
Extending Hillman and Hitt’s (1999) arguments from an OSCM perspective, I posit that
different levels of participation have different implications on firms’ reputational risks and
environmental performance. On the one hand, firms involved in collective engagement have a
better chance to shape climate policies in favor of their businesses without exposing themselves to
a higher level of scrutiny (Brulle, 2014; Cory et al., 2021). On the other hand, collective
engagement can potentially lead to better environmental performance. This is because collective
engagement fosters inter-organizational information sharing and allows firms to benchmark
against other members. Many climate change-related coalitions also help members set ambitious
goals of reducing carbon emissions while serving as a platform for the promotion of best practices.
With several analyses showing that trade associations have become the dominant sources of
lobbying expenditures in the U.S. (Brulle, 2018; Drutman, 2015), I believe empirically examining
performance outcomes of different levels of participation can be valuable for future research.
And second, EICCP can rely on different types of resources that firms devote to
engagement activities. Specifically, Hillman and Hitt (1999) identified information, financial
incentives, and constituency-building as three different resources firms can use to influence
54
policymaking, among which information and financial incentives apply to direct policy
engagement, while constituency-building work in indirect policy engagement. Since this study
focuses on direct policy engagement in climate change, I elaborate on information and financial
incentives and identify knowledge resources and financial resources as resources firms can devote
to EICCP. I argue that in the climate change context, firms use not only information, which is
defined as “firms’ preferences for policy or policy positions” (Hillman and Hitt, 1999, pp. 834),
but also the expertise in mitigating climate impact as resources to influence the decisions of
policymakers. Such expertise can take the form of technological know-how needed for
implementing or developing carbon reduction techniques. I refer to them as knowledge resources.
The use of knowledge resources is relevant to OSCM research given that it is tacit and accumulates
through firms’ sustainable practices. Firms that rely on knowledge resources in EICCP are likely
to be leading firms in sustainable operations and are thus expected to have good environmental
performance (Schmidt et al., 2017). Financial resources refer to political action funds firms spend
to influence climate change policymaking, as per Hillman and Hitt’s (1999) definition. Prior
literature on CPA indicates that the expenditure on policy engagement is negatively associated
with firms’ environmental performance, especially for industries that are under great
environmental scrutiny (Cho et al., 2006).
Identifying the different types of resources devoted to EICCP helps to further understand
the different performance implications of different EICCP strategies. I argue that knowledge
resources and financial resources, although both are related to firms’ environmental performance,
may yield different performance outcomes since they have different levels of transferability.
Specifically, knowledge resources that a firm devotes to EICCP are not consumed but strengthened
through consistent engagement activities, while the political action funds devoted to an
55
engagement activity, or a specific climate-related issue, cannot be spent elsewhere. Following the
resource-based view, firms devoting knowledge resources to EICCP should be able to develop a
greater ability to reshape climate change policy in favor of their businesses than firms relying on
financial resources in the long run. As such, firms can take advantage of favorable policy
conditions to make their emission figures look better without reducing carbon emissions.
Consequently, using knowledge resources for EICCP allows firms to achieve better environmental
performance.
3.3. Measuring EICCP strategies
To validate the taxonomy that I propose for EICCP strategies, I seek to investigate distinct EICCP
strategies firms adopt. However, identifying firms’ EICCP strategies is challenging given the
scarcity of information and the lack of established measures for EICCP. Investigations by non-
profit organizations have shown that large firms and trade associations resist political disclosure
(Levinthal, 2016), and thus, the mandatory disclosure of CPA, which can be a stable source of
information, cannot be established (Werner, 2017). To solve this problem, I follow the literature
on voluntary CPA disclosure (e.g., Goh et al., 2020; Lei et al., 2019) to analyze firm-level self-
disclosed data from the Carbon Disclosure Project’s (CDP) Climate Change (CC) dataset. The data
is unstructured text, which includes firms’ descriptions of their direct and indirect policy
engagement. To extract information from the texts to measure EICCP strategies, I employ the
method of text analytics. Specifically, I use Natural Language Processing (NLP) with random
forest classification, a supervised learning approach, to automate the text mining processes. While
text analytics has been extensively applied in OSCM research, the machine learning approach for
text analysis is relatively novel (Bansal et al., 2020). Recent research has shown that the machine
learning approach performs better than other text analysis approaches, such as the dictionary
56
method. Therefore, I adopt the machine learning approach in this study. Below, I describe the data
and the method I use.
3.3.1. Data
The CDP CC dataset collects firms’ public responses to an information request sent by CDP CC
on behalf of their signatory investors. Since 2013, CDP CC has included in the questionnaire a
series of questions about firms’ engagement in climate change policies. For this study, I use the
responses to the open-ended question: “On what issues have you been engaging directly with
policymakers?” The text data for analysis combine firms’ statements on details of engagement and
proposed legislative solutions to form a complete response to an issue. Firms can submit multiple
answers if they engage in more than one issue, which results in multiple data entries for a firm
year. I construct panel data using responses from 2013 to 2019 from the CDP CC datasets, which
include 1,309 firms and 8,529 firm-year responses.
3.3.2 Text Analytics Using Natural Language Processing with Random Forest Classification
To validate the typology of EICCP strategies I proposed, I apply Natural Language Processing
(NLP) techniques to the policy engagement disclosure data from the CDP CC dataset. Specifically,
I follow the supervised learning approach to employ the random forest model as the major
classifier to train the classification model and use it to automatically identify and classify different
engagement strategies. The application of automated learning for text analysis has substantially
progressed over the last decades in business research due to the empirical evidence that the well-
designed algorithms are gradually closing the gap between automated classification and manual
classification in precision, and that automated classification outperforms the manual classification
in efficiency, objectivity, statistical power and replicability (e.g., Donovan et al., 2021; Frankel et
al., 2016; Huang et al., 2018). Among different classification models for supervised learning, I
57
choose to use the random forest model due to the recent evidence that the assessment of text
disclosures using the random forest method yields “the least measurement error relative to
measures based on alternative machine-learning methods such as support vector regression and
supervised-latent-Dirichlet allocation” (Frankel et al., 2021, p. 16).
A random forest is an ensemble of a predefined number of individual decision trees. It
relies on the decision tree that yields the best prediction results to do the classification. Therefore,
the random forest model inherits the merits of the decision tree model but possesses additional
advantages. First, the outcome of the random forest model is less sensitive to the selection of
training data and is more accurate than the output of a single decision tree (Ali et al., 2012). In
each decision tree in a random forest model, the analysis follows a hierarchical framework to
divide the training data consecutively by a set of word features. The model training starts from
identifying and using the most important word feature among all the selected features to divide the
training data into two parts, then, advances to the next hierarchy in which each part is further
divided by another word feature that ranks in the second place in terms of importance. The model
training continues until all the training data are properly classified. Since the analysis involves
word features in the order of importance, I need to first evaluate the importance of each word
feature using metrics such as entropy, information gain, gain ratio, and Gini index. This evaluation
captures the relative importance of words in a specific text and is not robust to changes in the
training data. The random forest model overcomes this deficiency by introducing a voting
mechanism to train the best model. Instead of using the original training data, it draws random
samples of a predefined size from training data with replacement and trains decision tree models
separately on each random sample. When incorporating several decision trees for the analysis and
using the most voted decision tree results for classification, the random tree model considers a
58
larger set of word features and generates a better training model. The iterative approach also
increases the robustness of the prediction.
Second, model overfitting is less likely for the random forest model than for the decision
tree model. Besides the calculation of the importance of word features, the decision tree model
training relies on decisions including how to set a threshold for importance level to select word
features and how many end nodes should be in the final decision tree. Those arbitrary decisions
can be decisive for the training outcomes – a model with more word features and more end nodes
yields more accurate results but might be overfitted and affected by the noise in the data. In
comparison, the random forest model depends less on parameter settings. With running decision
tree models iteratively, the random forest model allows using word features with a relatively lower
level of importance, as well as a flexible number of features and end nodes, without jeopardizing
the accuracy of the final prediction. Figure 2.2 provides illustrations of the presumptive results of
the decision tree model and random forest model.
I detail my application of the NLP techniques with random forest classification below.
Figure 2.2 Presumptive results of decision tree model and random forest models
Root Root Start with the most
Root
important feature
among a random
Start with a feature selected
sample of all features
from the pool of features
node1 node2 node1 node2 …
node1 node2
…… ……
…
…… …… …… ……
……
…… …… Decision tree result 1 Decision tree result 2 …………
Decision tree result Random forest result
59
3.3.3 Manual Labelling
The data inputs for the text analytics are unstructured and usually comprise a set of sentences and
paragraphs. Before training the data to develop a prediction model, I need to construct labels and
conduct feature engineering to identify features that have the highest predictive power of the label.
Since the concept and the typology of EICCP are novel, there is no established data that I can use
to label different strategies. Therefore, I perform manual labeling to prepare the training data for
further analysis. First, I randomly pick 10% of firm responses, which equals 850 texts, to form the
training sample. Second, I import the training sample into Atlas.ti, qualitative data analysis, and
research software, and assign labels to texts (see Figure 2.3 for the interface of Atlas.ti). Labeling
is a manual process and relies on researchers’ judgment. I read and manually assign labels to each
text if I believe it falls in a category of an engagement strategy. One text can have multiple labels
if it presents more than one engagement strategy. Although the process is subjective, I endeavor
to achieve consistency by developing standards for labeling and engaging other raters for
validation. In Table 2.1 I briefly discuss the standards for labeling for each strategy and provide
examples.
Figure 2.3 The interface of Atlas.ti
Note. I can access the original text on Atlas.ti and assign self-defined labels to the text.
60
Table 2.1 Illustrations of manual labeling standards and text examples for each strategy
Strategy Standards Text example
Opposition The appearance of only dissenting On May 27, 2015, Apple voiced objections to the energy policy language of
strategy views on a specific policy North Carolina bill H332, which we believed, if passed, would have had a
significant negative impact on the availability of a clean and diversified
energy supply in the state. We partnered with other technology companies to
jointly write to the North Carolina legislature. The bill was thereafter
defeated. Enabling a clean and diversified energy supply.
Hedging strategy The appearance of concerns on a In France, the environmental legislation Grenelle II and the Decree 2011-
specific policy, or the coexistence 1336 request the implementation of mandatory carbon reporting on shipment
of favorable and dissenting views level by October 2011. We generally support this approach but have been
on a specific policy, followed by actively engaged through consultations, talks, and presentations at relevant
justifications that the opposition meetings to request that not an isolated national approach is taken, but also
originates from the concerns over international methodologies will be considered valid. With the publication of
unfair competition, profit loss, and the EN 16258 standard, the transport sector has its first official standard for
instability of the market; usually carbon calculations on a product level. This standard and the tools and
mentions an alternative policy methods proposed therein should be recognized as a valid methodology
within the French legislation.
61
Table 2.1 (cont’d)
Strategy Standards Text example
Support strategy The appearance of only favorable Pendal is a signatory and investor participant in the global Climate Action
views on a specific policy; no 100+ initiative. It has directly supported engagement with an Australian oil
evidence of proactively advancing and gas company. Through Regnan, Pendal has supported submissions to
the policy or guiding the the Australian Stock Exchange (ASX) consultation on the latest edition of
policymaking into a new direction its Principles and Recommendations, a framework for listed company
reporting. We encouraged more detailed guidance relating to the
consideration of climate change as a material business risk. Further, that
companies should disclose material climate-related risks in their main
corporate filings.
Promotion The appearance of only favorable Consumers Energy participated in the Midwest Collaborative to develop a
strategy views on a specific policy, with regional influence on EPA’s effort to craft GHG regulations via Section 111
proactive practices in of the Clean Air Act (CAA). The Midwest Collaborative is a voluntary
policymaking such as proposing group effort, consisting of industry, state regulatory, and environmental
new policies advocacy representatives. One of the group’s primary goals is to develop
stakeholder consensus on a policy framework for any upcoming regulations
implemented under Section 111 of the CAA. Participation is based on
periodic conference calls with some face-to-face meetings. The primary
work product is to develop a straw man proposal to present to EPA before
its completion of draft regulations. Consumers Energy supports a
representative stakeholder process developing consensus-driven guidance
for submittal to EPA to influence rulemaking processes.
62
Table 2.1 (cont’d)
Strategy Standards Text example
Collective strategy Statements on firms’ affiliation or Same as above
membership of one or multiple
groups, followed by the
descriptions of how the group(s)
engage in climate change
policymaking
Individual strategy A firm instead of a group is the Cabot engaged with the United States Environmental Protection Agency to
subject of the policy engagement; better understand the application of the clean power program to a proposed
no evidence of engagement cogeneration project at its Franklin, Louisiana vacuity. Cabot routinely
through groups or affiliations communicated with staff at the Agency’s Research Triangle Park offices.
Cabot will continue to advocate for clarification of the applicability of the
Clean Power program to ensure it encourages the capture of conversion of
waste energy.
63
Table 2.1 (cont’d)
Strategy Standards Text example
Use of knowledge The use of the information such as As one of the largest insurance and asset management companies in the
resources policy recommendations, world, AXA engages in a set of policy and regulatory issues that may affect
understanding of the technologies, the Group's strategy over the short and long term. On top of key prudential,
etc., are mentioned consumer protection-related, and digital economy-related topics, the Group
engages in the various policy and regulatory initiatives related to the long-
term financing of the economy (EU or French projects) in connection also to
sustainability issues and climate change. AXA contributed to the EU High-
Level Expert Group on Sustainable Finance, which developed
recommendations on how sustainability could be placed in the European
Union's core financial processes, how different participants in the financial
system could act on it, and how to mobilize capital more effectively for a
sustainable economy. Sustainable finance offers Europe a powerful tool for
achieving its goals of economic prosperity, social inclusion, and
environmental regeneration.
Note. I removed the label for the use of financial resources due to limited observations. This table does not present a text sample for the
strategy “Use of financial resources” because I did not find any disclosure about this strategy in the training data. Since the use of
financial resources is theoretically justified, failing to find relevant descriptions may attribute to two reasons. First, the current training
data does not successfully capture the description of this strategy. Second, firms choose not to report their use of financial resources. It
is a limitation of the current study that I have not expanded the size of the training data to test those assumptions. I will consider
investigating this strategy in future research.
64
3.3.4 Feature Engineering
In this step, I identify and evaluate word features using the labeled training data. Feature
engineering is the process of deriving useful features from unstructured data using the manual label
as a reference. In text analytics, word features that are useful for the prediction meet the following
standards. First, they should significantly correlate with the manual labels either in a positive or
negative relationship. Second, they should not have a high correlation with all manual labels. This
criterion rules out stop words or other common vocabularies that exist in nearly all text responses
but do not deliver meaningful information to guide classification. Third, useful word features
should not be redundant and duplicated. This criterion requires the stemming of words to unify
inflected or derived words to the same word root by eliminating the suffixes and prefixes of words.
To prepare word features that meet those requirements, I follow the standard data processing
procedure for text analysis specified in the three steps in Figure 2.4, similar to prior text analytics
research (e.g., Frankel et al., 2016, 2021).
Figure 2.4 Data pre-processing for feature engineering
qualiti must improv
(1) “Quality 99” is a
(1) quality must (1) qualiti must
must (1) 0.34 0.05 0
(2) amd must (2) amd must
(2) Amd must do
improvement improv (2) 0 0.05 0.21
improvement
(3) …… (3) ……
(3) …… (3) …… …… ……
Step 1: data preprocessing: Step 3: build training data
Step 2: word stemming
(1) Number is filtered. (1) Derive both individual and
(1) Unify singular and plural
(2) Punctuation mark is filtered. combined word features.
forms.
(3) Lower case is applied. (2) Split the sentence into word
(2) Unify the word tense.
(4) Company name is filtered. features.
(3) Extract the word root.
(5) Alphabet letter is filtered. (3) Calculate TF-IDF.
In Step 1, I first eliminate the 13 non-English responses. Then, I pre-process each sentence
of the remaining responses by deleting the numbers, punctuations, company names, alphabet
65
letters, and a list of common stop words used in the English language. Also, I changed all capital
letters into lower cases since it reduces the inconsistency among words without affecting the
meaning of the documents. In Step 2, I perform word stemming using the Porter Stemming
Algorithm (Porter, 1980) to unify different word forms including the singular and plural forms,
and distinct word tense and part of speech. In Step 3, I further derive the individual and combined
word features. Individual word features refer to only a single stemmed word, while combined word
features refer to a combination of two consecutive stemmed words. The combined word feature is
more informative than the individual word feature because it can capture the coexistence features
of words in the sentence and further differentiate the word used for the development of a better
classification model. For instance, the stemmed word “qualiti” is neutral in sentiment but the
combined stemmed word “high qualiti” expresses a positive sentiment. In this study, I use both
individual and combined word features as candidate features for each text response. After
constructing the word features, I remove the sentence boundary and eliminate duplicate stemmed
words or stemmed word groups. This sub-step results in a set of unique word features derived from
all text responses. The final sub-step is to evaluate and quantify the importance of each word
feature. To do so, I use the term frequency-inverse document frequency (TF-IDF) measure. The
TF-IDF measure is the multiplication of the term frequency, which is captured by the count of a
specific word feature over the total number of word features in a text response, and the inverse
document frequency, which is the log of the count of a specific word feature in all the text
responses over the count of this word feature in the present text response. TF-IDF is an established
measure widely used in NLP since it not only captures the importance of the word features in one
text response but also offsets the effects of the useless high-frequency word feature that nearly all
66
text responses share (Gentzkow et al., 2019). The data pre-processing procedure converts
unstructured data to structured data, which supports further feature identifications.
While the structured data generated by the previous procedure provides interpretable
information for machine learning, it is highly dimensional, with tens of thousands of word features
in the data, making it difficult to incorporate into statistical analyses and to train a predictive model.
When the number of word features exceeds the number of text responses, the training model will
face the overfitting problem. Therefore, I need to conduct feature selection to extract word features
that have the highest predictive power. In this step, I use the biserial correlation metric to measure
the association between manual labels and word features, following the text mining literature. A
high correlation coefficient between a word feature and a manual label indicates that the word
feature is qualified for model training. In this study, I use the words that rank in the 75% quantile
in the correlation coefficient to ensure that the number of overall word features does not exceed
the number of observations in the training data and thus, avoid the overfitting problem. I used the
70% quantile and 80% quantile for sensitivity tests and found consistent prediction results. Table
2.2 shows the top-ranked word features in the biserial correlation metric for the manual label
“promotion strategy” as an illustration.
67
Table 2.2 Word features with the highest biserial correlation coefficients by manual label
Word features Original words Biserial correlation
with the manual label
benefit benefits, benefit, benefitting 0.32926985
condit conditions 0.31306848
include + carbon included + carbon, including + carbon 0.30337897
infrastructur infrastructure, infrastructures 0.28793145
car car, cars 0.28597564
germani germany 0.28348094
german + govern german + government 0.28313922
million +electror million + electric 0.26013577
damag damage 0.26011967
reduct reduction, reductions 0.25823276
3.3.5 Model Training
I use the structured training data from previous steps to train random forest models. A random
forest model can iterate the training process by including many decision trees, a predefined number
of randomly selected word features in each decision tree, and a predefined condition to stop
partition. In my analysis, I use 500 decision trees for each random forest iteration. The more
decision trees are included in a random forest model, the more stable the training model will be,
but the less efficient the computation. I use 300 and 400 decision trees for the robustness check
and find that changing the setting does not significantly influence model accuracy. Then, I set the
range of 1 to 50 as the number of word features used in a decision tree, which determines the
number of word features each partition will consider. For example, if I set 50 as the number of
word features applied, in each partition, the model will randomly select 50 word features and
choose the word feature with the highest biserial correlation coefficient as the decision node, then
move on to the next partition to identify another most relevant word feature among the 50 random
selected word features. While literature suggests that a lower number of word features used
decreases the correlations among decisions trees and yields more stable predictions (Probst et al.,
68
2019), I choose to start from a relatively large range of word features and rely on the prediction
accuracy to empirically define the optimal number of word features for model training. The
partition in each decision tree will end as no more partitions are needed. The common conditions
for stopping partition include a threshold of least improved entropy and the number of partitions,
whose values are automatically chosen by the algorithm. Besides those stopping criteria, the model
usually requires another threshold for the minimum sample size in the terminal node. This
threshold applies to prevent overfitting when the stopping criteria are not met (Segal, 2004). I
arbitrarily predefined the number equivalent to 5% of the training texts as the minimal size of a
terminal node. However, I found the actual partition in our data all stopped when there was at least
more than 10% of the training texts. That means the decision trees met the stopping criteria before
triggering the threshold of the minimum sample size in the terminal node. That is to say, the model
training does not rely on the arbitrary number we defined.
3.4. Evaluation and Results of EICCP Measures
The evaluation of the random forest model is different from other classification models in which
an n-fold cross-validation approach is applied to get an unbiased estimate of the model accuracy
(Segal, 2004). Since each decision tree in the random forests randomly selects a predefined number
of word features to train the model, I can configure the maximum number of selected word features
to construct an important metric for the selected word features. The validation process for this
metric is similar to a three-fold cross-validation approach. Specifically, each iteration in the
random forest model employs a different bootstrapped sample from the training data, in which
about one-third of the training data is left out and is not used in the construction of the decision
trees. Once all decision tree models are trained using the remaining two-thirds of the training data,
the majority vote of those models can be generated. The accuracy of the random forest model is
69
then evaluated by applying this majority-vote model to the one-third of training data that has been
left out and calculating the accuracy of label classifications. The model accuracy metric will
stabilize as the number of decision trees in the random forest model increase but usually will not
improve significantly when the number of decision trees passes a certain threshold (Probst et al.,
2019). Therefore, I used a different number of trees for the robustness check.
Table 2.3 shows the accuracy, the optimal number of the word features for the random
forest model training, and the total number of labeled text responses in the training data. I was not
able to model the strategy featured by the use of financial resources because of too few numbers
of labeled texts for this strategy. The accuracy of random forest models for all other models is over
80%, with the prediction of hedging strategies having the highest accuracy and the use of
knowledge resources in EICCP ranking second in terms of accuracy. The optimal number of word
features needed for random forest model training captures the efficiency of this predictive model.
The best models use as few as eight word features to identify the hedging strategy or to differentiate
individual engagements from collective engagements. In general, incorporating more word
features will stabilize the model’s accuracy. Figure 2.5 illustrates the relationship between the
number of word features used for model training and the resulting model accuracy for the
promotion strategy. The reported number of manually labeled texts shows that the distribution of
EICCP strategies is not balanced. The least adopted strategy is the opposition strategy. The label
classification of the opposition strategy requires more word features and is likely to be more
sensitive to the changes in the training data.
Table 2.4 presents a list of ten top-ranked word features for the training of random forest
models for each strategy. The importance value is scaled to a range of 1 to 100 for illustrative
purposes. The higher the importance value, the more useful a word feature for the label prediction.
70
Although I include the combined word features into the pool of word features, the results show
that the importance of the single-word features outweighs the importance of the combined word
features. I also observe that some single-word features rank top in importance for more than one
label. This is normal in random forest classifications since the model not only considers the
information delivered by the individual word feature but also the interrelations among word
features for the prediction.
71
Table 2.3 Results of the random forest models by manual labeling
Promotion Support Hedging Opposition Individual vs. Use of
strategy strategy strategy strategy Collective knowledge
engagement resources
Accuracy
83.52% 88.06% 95.53% 85.83% 88.14% 91.76%
The optimal number of
word features for random 15 18 8 24 8 12
forest model training
Total number of labeled
279 461 129 42 447 461
texts in the training data
Table 2.4 Top-ranked word features with their importance value in the random forest model by manual labeling
Label Stemmed word Original word Importance value
benefit benefits, benefit, benefitting 100
deliv delivering, deliver, delivered 95.76604651
car car, cars 85.82345359
reduct reduction, reductions 82.30050453
Promotion refer reference 81.1160511
strategy recycl recyclable, recycling 80.11132255
damag damage 54.16039484
german german 39.51177473
germani germany 38.53420378
prior prior 26.90623168
72
Table 2.4 (cont’d)
Label Stemmed word Original word Importance value
like like, likely 100
support support, supporting, supports, supported, supportive 87.35595215
carbon carbon 76.4263096
effici efficiency, efficient, efficiently, efficiencies 69.73425017
avoid avoid 65.67096615
Support
reduct reduction, reductions 60.99042573
strategy
see see, seeing, sees 60.07590989
tax taxes, tax, taxing 53.22218388
report report, reporting, reports, reported 52.34498386
commitment, committed, commitments, committed, commit,
commit 50.30464546
committing, commits
Label Stemmed word Original word Importance value
base based, base 100
tax taxes, tax, taxing 95.26947338
price price, pricing, prices 94.91548668
competit competition, competitive, competitiveness 88.85893806
Hedging adjust adjusted, adjust 85.85833515
strategy avoid avoid 81.19206705
carbon carbon 60.82896735
equal equal, equally, equality 51.1678536
leakag leakage 41.26519366
view view, views 40.99033926
73
Table 2.4 (cont’d)
Label Stemmed word Original word Importance value
reject reject, rejecting 100
notic noticeable 22.42896593
seem seems 18.65198389
Opposition
shortcom shortcomings 17.0592976
strategy
situat situation 15.25507852
shortag shortage 15.1787334
unlik unlike 13.87098634
Label Stemmed word Original word Importance value
insur insurance, insurer, insurers 100
fight fight 73.81066551
issu issued, issues, issue 72.22356913
Individual economi economy 68.47477488
vs. council council 66.79671497
Collective coordin coordinates, coordinated, coordination, coordinate 64.96486061
engagement aim aims, aiming, aim, aimed 64.3818241
associ associated, association, associations 63.61627332
industri industrial, industry, industry’s, industries 47.4586111
research research 42.19198728
74
Table 2.4 (cont’d)
Label Stemmed word Original word Importance value
team teams, team 100
calcul calculation, calculator, calculate 72.28758843
guidanc guidance 69.32843009
rail rail 67.11812223
Use of lead lead, leading 55.9354086
knowledge
resources urbanis urbanization 53.56842282
rapid rapid 49.64596555
locki lockie 41.55400531
profil profile 26.72260299
scarciti scarcities 13.66049445
Note. The tables show only the ten top-ranked word features based on the importance of each strategy. For label
prediction, I included more word features in the analysis. The original words for each word stem share the same word
root but have different forms of tense and form. I extract them from the original documents and reassign them to each
word stem after the model training.
75
Figure 2.5 The relationship between the accuracy and the number of randomly selected word
features for the model training for the promotion strategy
Accuracy
0.80
Accuracy (Repeated Cross-Validation)
0.75
0.70
0 10 20 30 40 50
number of
Number of randomly
randomly selected
selected word
data features
features
3.5. Discussion
In this study, I propose a taxonomy for EICCP strategies based on firms’ value perspectives and
engagement levels, levels of participation, and resources devoted to EICCP. Examining EICCP
strategies using firms’ self-disclosed data, I found that there is an unbalanced application of
different strategies and provide an overview of the real-world EICCP below.
3.5.1. Overview of EICCP
The adoption of EICCP strategies has clear trends as shown in Figure 2.6. Focusing on the four
strategies classified by value perspective and engagement level, the implementation of the support
strategy has declined while the usage of the promotion, hedging, and opposition strategies has
increased from 2013 to 2019. However, the support strategy remains the most adopted strategy
among these four strategies (see Figure 2.7). There are very few firms that take the opposition
strategy. The application of the promotion strategy is slightly more than that of the hedging strategy,
76
but both have a significantly lower level of implementation than the support strategy. These
observations are consistent with the notion that when the climate change policy process advances
and provokes a higher level of public pressure, firms tend to enhance their engagement (Hillman
and Hitt, 1999; Meckling, 2015) – moving from low engagement levels to high engagement levels,
such as moving toward the promotion or the hedging strategies. The only exception here is the
opposition strategy, which is a strategy with a low engagement level according to the taxonomy I
propose, but which has been implemented increasingly over time. A potential interpretation is that
many nations have only started the policy process recently and are thus expected to stay on the
agenda-setting stage for a long time. At this stage, the opposition strategy is efficient in reducing
compliance costs for them, so it experienced an increase in adoptions.
In addition, the level of individual engagement was lower in 2019 than it was in 2013,
suggesting that firms increasingly adopt collective engagements over time. This also meets the
political notion that firms increasingly engage through coalitions, associations, or other groups to
consolidate resources and make a greater impact on policymaking without exposing themselves to
a higher level of reputational risks and scrutiny (Cory et al., 2021; Hillman and Hitt, 1999). Finally,
the use of knowledge resources in EICCP is substantially less in 2019 than in 2013. This might be
attributed to the larger knowledge base about climate change and research on climate change that
policymakers can access through enhanced intergovernmental collaborations (IPCC, 2022).
Consequently, the business sector provides less input.
77
Figure 2.6 Implementation trends for different EICCP strategies
Likelihood Likelihood Likelihood
Likelihood Likelihood Likelihood
Figure 2.7 Longitudinal comparison of four EICCP strategies classified by value perspectives
and engagement levels
78
3.5.2. Contributions
My research is positioned at the interface between public policy and OSCM. Heeding the calls for
OSCM research that “consider regulatory policy uncertainty as a driver of decisions and business
performance, and how it shapes exchange” (Tokar and Swink, 2019, pp. 76), I sought to investigate
firms’ responses to climate change policies. Prior literature has identified firms’ various
approaches to obtain organizational legitimacy under regulatory scrutiny but also indicated that
academic attention to firms’ influence-oriented actions, which refers to firms’ political activities
aiming at shaping governmental policy or process, is still scarce (Grover and Dresner, 2022). The
lack of consideration for business factors in influencing policymaking processes leads to an
unrealistic assumption that the policy and regulatory environment is an objective and
uncontrollable context that is external to firms’ strategic decisions, operations, and activities.
Consequently, extant research on climate change focuses mainly on policy compliance and
adaptation, with very few studies exploring business efforts in reshaping the policies in favor of
firms’ interests (Cory et al., 2021; Greiner and Kim, 2021).
To further the disciplinary understanding of firms’ influence-oriented actions in climate
change, I proposed the concept of EICCP, investigated firms’ various strategies of EICCP, and
proposed a typology for engagement strategies. Validating my typology using firms’ self-
disclosure on direct or indirect engagement with policymakers in climate change issues from CDP
data, I empirically demonstrated that firms adopt various approaches to influence policy and
regulatory environments for climate change issues. This study provided insights into the
multiplicity of the firms’ influence-oriented responses to climate change policies. I further
discussed the implications of performance and competitive advantage for different engagement
79
strategies, setting a foundation for future research on the outcomes of the complex decision-
making in EICCP.
My research also adds to prior research methodologically. While recent research suggests
that the machine learning method performs better than a dictionary-based method in accuracy and
robustness for text analytics, machine learning approaches have not been sufficiently implemented
in business literature (Frankel et al., 2021; F. Li, 2010). I followed machine learning methods to
convert unstructured and qualitative disclosure data into structured word features via feature
engineering. Then, I performed text analytics on the converted and structured data using the
random forest model, a supervised learning method that overperformed other methods in sentiment
analysis (Frankel et al., 2021), to identify firms’ engagement strategies.
This study yielded measures of EICCP engagement strategies that can be used for future
research. Using firms’ self-disclosed information, my measures captured the extent to which firms
engage in each type of strategy. They are available for 1,309 global firms spanning from 2013 to
2019. Future studies can use the measures developed for empirical investigations on antecedents
and outcomes of EICCP.
80
REFERENCES
81
REFERENCES
Ali, J., Khan, R., Ahmad, N., and Maqsood, I. (2012). Random Forests and Decision Trees.
IJCSI International Journal of Computer Science Issues, 9(5), 272–278.
Aragón-Correa, J. A., Marcus, A. A., and Vogel, D. (2020). The Effects of Mandatory and
Voluntary Regulatory Pressures on Firms’ Environmental Strategies: A Review and
Recommendations for Future Research. Academy of Management Annals, 14(1), 339–365.
Automotive World. (2020). Growing momentum: Global overview of government targets for
phasing out sales of new internal combustion engine vehicles.
https://www.automotiveworld.com/news-releases/growing-momentum-global-overview-of-
government-targets-for-phasing-out-sales-of-new-internal-combustion-engine-vehicles/.
Bansal, P., Gualandris, J., and Kim, N. (2020). Theorizing Supply Chains with Qualitative Big
Data and Topic Modeling. Journal of Supply Chain Management, 56(2), 7–18.
Barnett, M., and Duvall, R. (2004). Power in Global Governance. In Power in Global
Governance. Cambridge University Press.
Blumentritt, T. P. (2003). Foreign Subsidiaries’ Government Affairs Activities: The Influence of
Managers and Resources. Business & Society, 42(2), 202–233.
Bonardi, J. P., Hillman, A. J., and Keim, G. D. (2005). The Attractiveness of Political Markets:
Implications for Firm Strategy. Academy of Management Review, 30(2), 397–413.
Boudette, N. E., and Davenport, C. (2021, January 28). G.M. Will Sell Only Zero-Emission
Vehicles by 2035. The New York Times, 28–30.
https://www.nytimes.com/2021/01/28/business/gm-zero-emission-vehicles.html.
Brulle, R. J. (2014). Institutionalizing Delay: Foundation Funding and the Creation of U.S.
Climate Change Counter-Movement Organizations. Climatic Change, 122(4), 681–694.
Brulle, R. J. (2018). The Climate Lobby: A Sectoral Analysis of Lobbying Spending on Climate
Change in the USA, 2000 to 2016. Climatic Change, 149(3–4), 289–303.
Bumpus, A. G. (2015). Firm Responses to a Carbon Price: Corporate Decision Making under
British Columbia’s Carbon Tax. Climate Policy, 15(4), 475–493.
Chen, C.-M. (2017). Supply Chain Strategies and Carbon Intensity: The Roles of Process
Leanness, Diversification Strategy, and Outsourcing. Journal of Business Ethics, 143(3),
603–620.
Cho, C. H., Patten, D. M., and Roberts, R. W. (2006). Corporate Political Strategy: An
Examination of the Relation between Political Expenditures, Environmental Performance,
and Environmental Disclosure. Journal of Business Ethics, 67(2), 139–154.
82
Clapp, J., and Meckling, J. (2013). Business as a Global Actor. In The Handbook of Global
Climate and Environment Policy (pp. 286–303). John Wiley & Sons Ltd.
Cory, J., Lerner, M., and Osgood, I. (2021). Supply Chain Linkages and the Extended Carbon
Coalition. American Journal of Political Science, 65(1), 69–87.
Dadhich, P., Genovese, A., Kumar, N., and Acquaye, A. (2015). Developing Sustainable Supply
Chains in the UK Construction Industry: A Case Study. International Journal of Production
Economics, 164, 271–284.
Dahan, N. (2005). A Contribution to the Conceptualization of Political Resources Utilized in
Corporate Political Action. Journal of Public Affairs, 5(1), 43–54.
Doherty, C., Kiley, J., Asheer, N., and Jordan, C. (2020). Election 2020: Voters Are Highly
Engaged, but Nearly Half Expect To Have Difficulties Voting.
https://www.pewresearch.org/politics/2020/08/13/important-issues-in-the-2020-election/.
Domonoske, C. (2021). Big Oil (Probably) Isn’t Going Away Anytime Soon. But It’s Definitely
Changing. Npr.
Donovan, J., Jennings, J., Koharki, K., and Lee, J. (2021). Measuring Credit Risk Using
Qualitative Disclosure. Review of Accounting Studies, 26(2), 815–863.
Drutman, L. (2015). The Business of America Is Lobbying: How Corporations Became
Politicized and Politics Became More Corporate. Oxford Scholarship Online, 15(1), 583–
605.
Eberlein, B., and Matten, D. (2009). Business Responses to Climate Change Regulation in
Canada and Germany: Lessons for MNCs from Emerging Economies. Journal of Business
Ethics, 86(S2), 241–255.
EEU. (2018). Renewable energy targets 2020 and 2030. https://eeueuropa.eu/renewable-energy-
targets-2020-2030/.
Frankel, R., Jennings, J., and Lee, J. (2016). Using Unstructured and Qualitative Disclosures to
Explain Accruals. Journal of Accounting and Economics, 62(2–3), 209–227.
Frankel, R., Jennings, J., and Lee, J. (2021). Disclosure Sentiment: Machine Learning vs.
Dictionary Methods. Management Science.
Gentzkow, M., Kelly, B., and Taddy, M. (2019). Text as Data. Journal of Economic Literature,
57(3), 535–574.
Ghadge, A., Wurtmann, H., and Seuring, S. (2020). Managing Climate Change Risks in Global
Supply Chains: A Review and Research Agenda. International Journal of Production
Research, 58(1), 44–64.
Goh, L., Liu, X., and Tsang, A. (2020). Voluntary Disclosure of Corporate Political Spending.
83
Journal of Corporate Finance, 61, 101403.
Greiner, M., and Kim, J. (2021). Corporate Political Activity and Greenwashing: Can
CPA Clarify Which Firm Communications on Social & Environmental
Events Are Genuine? Corporate Social Responsibility and Environmental Management,
28(1), 1–10.
Grover, A. K., and Dresner, M. (2022). A Theoretical Model on How Firms Can Leverage
Political Resources To Align With Supply Chain Strategy for Competitive Advantage.
Journal of Supply Chain Management, 58(2), 48–65.
Hardy, C., Bhakoo, V., and Maguire, S. (2020). A New Methodology for Supply Chain
Management: Discourse Analysis and Its Potential for Theoretical Advancement. Journal of
Supply Chain Management, 56(2), 19–35.
Hetzner, C. (2021). Automakers Blast Europe’s Proposed Ban on New Combustion Engine Cars
by 2035. Fortune.Com, N.PAG.
Hillman, A. J., and Hitt, M. A. (1999). Corporate Political Strategy Formulation: A Model of
Approach, Participation, and Strategy Decisions. Academy of Management Review, 24(4),
825–842.
Hillman, A. J., Keim, G. D., and Schuler, D. (2004). Corporate Political Activity: A Review and
Research Agenda. Journal of Management, 30(6), 837–857.
Huang, A. H., Lehavy, R., Zang, A. Y., and Zheng, R. (2018). Analyst Information Discovery
and Interpretation Roles: A Topic Modeling Approach. Management Science, 64(6), 2833–
2855.
IEA. (2022). IEA’s Policies and Measures Database. https://www.iea.org/policies.
IPCC. (2022). Special Report: Global Warming of 1.5 C- Summary for Policymakers. In Global
Warming of 1.5°C. Cambridge University Press.
https://www.cambridge.org/core/product/identifier/9781009157940%23prf2/type/book_part
. https://doi.org/10.1017/9781009157940.001
Ji, G., Gunasekaran, A., and Yang, G. (2014). Constructing Sustainable Supply Chain under
Double Environmental Medium Regulations. International Journal of Production
Economics, 147, 211–219.
Jin, M., Granda-Marulanda, N. A., and Down, I. (2014). The Impact of Carbon Policies on
Supply Chain Design and Logistics of a Major Retailer. Journal of Cleaner Production, 85,
453–461.
Jira, C. (Fern), and Toffel, M. W. (2013). Engaging Supply Chains in Climate Change.
Manufacturing & Service Operations Management, 15(4), 559–577.
Keim, G. D., and Bonardi, J. P. (2005). Corporate Political Strategies for Widely Salient Issues.
84
The Academy of Management Review, 30(3), 555–576.
Keim, G. D., and Zeithaml, C. P. (1986). Corporate Political Strategy and Legislative Decision
Making: A Review and Contingency Approach. The Academy of Management Review,
11(4), 828.
Lawton, T., McGuire, S., and Rajwani, T. (2013). Corporate Political Activity: A Literature
Review and Research Agenda. International Journal of Management Reviews, 15(1), 86–
105.
Lee, J. W., Kim, Y. M., and Kim, Y. E. (2018). Antecedents of Adopting Corporate
Environmental Responsibility and Green Practices. Journal of Business Ethics, 148(2),
397–409.
Lee, K.-H. (2011). Integrating Carbon Footprint into Supply Chain Management: The Case of
Hyundai Motor Company (HMC) in the Automobile Industry. Journal of Cleaner
Production, 19(11), 1216–1223.
Lei, L. (Gillian), Li, Y., and Luo, Y. (2019). Social Media and Voluntary Nonfinancial
Disclosure: Evidence from Twitter Presence and Corporate Political Disclosure. Journal of
Information Systems, 33(2), 99–128.
Levinthal, D. (2016). Trade Groups to Top Corporations: Resist Political Disclosure.
https://publicintegrity.org/politics/trade-groups-to-top-corporations-resist-political-
disclosure/.
Li, F. (2010). The Information Content of Forward-Looking Statements in Corporate Filings-A
Naïve Bayesian Machine Learning Approach. Journal of Accounting Research, 48(5),
1049–1102.
Mahon, J. F. (1983). Corporate Political Strategies: An Empirical Study of Chemical Firm
Responses to Superfund Legislation. Research in Corporate Social Performance and
Policy, 5, 143–182.
McCarthy, N. (2019). Oil And Gas Giants Spend Millions Lobbying To Block Climate Change
Policies. Forbes. https://www.forbes.com/sites/niallmccarthy/2019/03/25/oil-and-gas-
giants-spend-millions-lobbying-to-block-climate-change-policies-
infographic/?sh=6797aac87c4f.
Meckling, J. (2015). Oppose, Support, or Hedge? Distributional Effects, Regulatory Pressure,
and Business Strategy in Environmental Politics. Global Environmental Politics, 15(2), 19–
37.
Menguc, B., Auh, S., and Ozanne, L. (2010). The Interactive Effect of Internal and External
Factors on a Proactive Environmental Strategy and Its Influence on a Firm’s Performance.
Journal of Business Ethics, 94(2), 279–298.
Meznar, M. B., and Nigh, D. (1995). Buffer or Bridge? Environmental and Organizational
85
Determinants of Public Affairs Activities in American Firms. Academy of Management
Journal, 38(4), 975–996.
Moniz, P., and Wlezien, C. (2020). Issue Salience and Political Decisions. In Oxford Research
Encyclopedia of Politics. Oxford University Press.
Oglethorpe, D., and Heron, G. (2010). Sensible Operational Choices for the Climate Change
Agenda. The International Journal of Logistics Management, 21(3), 538–557.
Okereke, C., and Russel, D. (2010). Regulatory Pressure and Competitive Dynamics: Carbon
Management Strategies of UK Energy-Intensive Companies. California Management
Review, 52(4), 100–124.
Oliver, C., and Holzinger, I. (2008). The Effectiveness of Strategic Political Management: A
Dynamic Capabilities Framework. Academy of Management Review, 33(2), 496–520.
Pagell, M., and Shevchenko, A. (2014). Why Research in Sustainable Supply Chain
Management Should Have No Future. Journal of Supply Chain Management, 50(1), 44–55.
Paulraj, A. (2011). Understanding the Relationships Between Internal Resources and
Capabilities, Sustainable Supply Management and Organizational Sustainability. Journal of
Supply Chain Management, 47(1), 19–37.
Pee, A. de, Pinner, D., Roelofsen, O., Somers, K., Speelman, E., and Witteveen, M. (2018).
Decarbonization of industrial sectors: the next frontier. In McKinsey & Company (Issue
June). https://www.mckinsey.com/~/media/McKinsey/Business Functions/Sustainability
and Resource Productivity/Our Insights/How industry can move toward a low carbon
future/Decarbonization-of-industrial-sectors-The-next-frontier.ashx.
Porter, M. F. (1980). An Algorithm for Suffix Stripping. Program, 14(3), 130−137.
Probst, P., Wright, M. N., and Boulesteix, A. L. (2019). Hyperparameters and Tuning Strategies
for Random Forest. In Wiley Interdisciplinary Reviews: Data Mining and Knowledge
Discovery (Vol. 9, Issue 3).
Rizet, C., Browne, M., Cornelis, E., and Leonardi, J. (2012). Assessing Carbon Footprint and
Energy Efficiency in Competing Supply Chains: Review – Case Studies and Benchmarking.
Transportation Research Part D: Transport and Environment, 17(4), 293–300.
RSP. (2021). State Renewable Portfolio Standards and Goals.
https://www.ncsl.org/research/energy/renewable-portfolio-standards.aspx.
Schmidt, C. G., Foerstl, K., and Schaltenbrand, B. (2017). The Supply Chain Position Paradox:
Green Practices and Firm Performance. Journal of Supply Chain Management, 53(1), 3–25.
Segal, M. R. (2004). Machine Learning Benchmarks and Random Forest Regression. In
Biostatistics.
86
Shaffer, B. (1995). Firm-Level Responses to Government Regulation: Theoretical and Research
Approaches. Journal of Management, 21(3), 495–514.
The World Bank. (2022). Carbon Pricing Dashboard.
https://carbonpricingdashboard.worldbank.org/.
Theißen, S., Spinler, S., and Huchzermeier, A. (2014). Reducing the Carbon Footprint within
Fast-Moving Consumer Goods Supply Chains through Collaboration: The Manufacturers’
Perspective. Journal of Supply Chain Management, 50(4), 44–61.
Tidy, M., Wang, X., and Hall, M. (2016). The Role of Supplier Relationship Management in
Reducing Greenhouse Gas Emissions from Food Supply Chains: Supplier Engagement in
the UK Supermarket Sector. Journal of Cleaner Production, 112, 3294–3305.
Tienhaara, K. (2013). Corporations: Business and Industrial Influence. In P. G. Harris (Ed.),
Routledge Handbook of Global Environmental Politics (pp. 164–175). Taylor & Francis
Group.
Tokar, T., and Swink, M. (2019). Public Policy and Supply Chain Management: Using Shared
Foundational Principles to Improve Formulation, Implementation, and Evaluation. Journal
of Supply Chain Management, 55(2), 68–79.
Villena, V. H., and Dhanorkar, S. (2020). How Institutional Pressures and Managerial Incentives
Elicit Carbon Transparency in Global Supply Chains. Journal of Operations Management,
66(6), 697–734.
Werner, T. (2017). Investor Reaction to Covert Corporate Political Activity. Strategic
Management Journal, 38(12), 2424–2443.
87
CHAPTER 4 – Supply Network Complexity, Regulatory Risks, and Firms’ Engagement in
Influencing Climate Change Policies
4.1. Introduction
Firms’ climate impact is facing unprecedentedly increasing scrutiny by stakeholders, investors,
and regulators, with the annual world carbon emission hitting a record in 2019, and the setting of
the goal in the 2020 Paris Agreement to limit global warming to well below 2°C (ideally to 1.5°C)
in the following 10 years. Such scrutiny not only exposes firms to climate risks (Engel et al., 2015)
but also makes it challenging for firms to assert leadership on climate change. As such, adapting
to existing climate policies is not enough to distinguish firms as climate leaders; more proactive
efforts are required (Reichart, 2019). To enhance organizational legitimacy and competitive
advantage, many firms have strived to go beyond policy compliance and engage in influencing
climate change policies. As examples serve more than 1,600 businesses that have committed to
ambitious emission reduction targets through the Science Based Targets initiative (SBTi)
(Ambitious Corporate Climate Action - Science Based Targets, 2022), and around 360 corporates
that have committed to 100% renewable electricity through RE100 initiatives to advocate the
world’s transition to a zero-carbon economy (RE100, 2022). These endeavors demonstrate firms’
more active advocacy of climate policies. Other examples include oil firms spending millions of
dollars per year on lobbying against climate-motivated policies that impose carbon taxes or prices
on carbon emissions, which reflect those firms’ opposition to regulatory restrictions (McCarthy,
2019).
As firms’ engagement in influencing climate change policies (EICCP) becomes frequent
and potentially impactful, it attracts the attention of different stakeholders in climate change
policies to keep track of or even monitor firms’ engagements. On the one hand, non-governmental
88
organizations (NGOs) have started to track big corporates’ deployment of political capital on
climate change to examine the impact of their engagement (InfluenceMap, 2021). On the other
hand, investors have urged firms to establish governance and disclosure procedures for their
engagement (Veena Ramani, 2020). However, firms’ EICCP has not received much academic
attention. Existing research mainly deals with climate change policies as external conditions to
which firms need to adapt, rather than regulatory environments that firms can shape or manipulate
in favor of their business (Greiner and Kim, 2021; Grover and Dresner, 2022). In Chapter 3, I
discussed the importance of investigating EICCP in understanding firms’ overall environmental
strategies and proposed and validated different strategies of EICCP. In this chapter, I continue
furthering the understanding of EICCP by investigating the antecedents of EICCP from a supply
chain perspective.
Prior literature has studied the antecedents of firms’ general political engagement (e.g.,
Clapp and Meckling, 2013; Hillman et al., 2004; Sadrich and Annavarjulia, 2002). However, those
discussions are not specifically concentrating on firms’ engagement in climate change policies. In
different regulatory contexts, firms’ engagement decisions can be heterogeneous (e.g., firms may
or may not engage, or may engage at different levels), considering the varying regulatory risks to
which they are exposed and the difficulties of exerting influences on policymaking. Therefore,
discussions on the antecedents of CPA need to be context specific. In addition, extant studies focus
on the impact of firm-level, industry-level, and country-level factors on the degree of firms’
engagement (Lux et al., 2011), while attention on the influence of supply chain-level factors is
scarce. From a network perspective, I argue that firms’ engagement in climate change policies
under regulatory risks is contingent upon the features of their supply network, given that the
characteristics of firms’ network ties have implications on the extent to which the network
89
members can benefit from the focal firms’ strategic decisions (Ojala and Hallikas, 2006;
Tachizawa and Wong, 2015).
In this study, I investigate the moderating effects of supply network complexity on the
relationship between firms’ regulatory risks and EICCP. Extant literature has identified regulatory
risks as key drivers of corporate political activities (CPA) (Clapp and Meckling, 2013; Lux et al.,
2011). However, some empirical studies have not found evidence for such a link between
regulatory risks and CPA (e.g., Lenway and Rehbein, 1991; Martin, 1995). I suggest that these
inconsistent results can be attributed to firms’ different perceptions of CPA’s efficacy. While CPA
has the potential to reshape the regulatory environment in favor of firms (Greiner and Kim, 2021;
Hillman et al., 2004), it requires substantial financial inputs and ultimately an investment decision
for engaging firms (Mitchell et al., 1997). As firms perceive the high level of uncertainty and
difficulty in influencing climate change policies, they may not choose to engage due to the
potentially low return on investments. I argue that supply network complexity constrains firms’
capabilities of engaging in climate change-related policymaking and increases the uncertainty of
the engagement outcomes. Therefore, examining regulatory risks and firms’ network complexity
as interrelated factors is critical for the understanding of firms’ engagement decisions. Following
this logic, I seek to answer the following research questions: 1) Does a higher level of regulatory
risks lead to a higher level of EICCP? 2) Does supply network complexity serve as a contingent
factor for the relationship between regulatory risks and EICCP? Specifically, does supply network
complexity negatively moderate the regulatory risks-EICCP link, such that this relationship
becomes weaker when supply network complexity is higher?
To address these questions, I adopt the measures of EICCP generated in Chapter 3 and
compile a panel spanning from 2013 to 2019 for firm-level regulatory risks and firms’ supply
90
networks. I find that regulatory risks have different impacts on firms’ EICCP under different levels
of supply network complexity. Further, distinct dimensions of supply network complexity have
different moderating effects on the regulatory risk-EICCP link.
4.2. Literature Review
4.2.1 Theoretical Underpinnings for EICCP Studies
Following the conceptualization of EICCP in Chapter 3, I define EICCP as strategic actions firms
perform to influence climate change policymaking processes, aiming at shaping policies or
promoting policy changes in favor of their interests. While no literature exists that has studied the
antecedents of EICCP, CPA research can provide some theoretical foundations for my study, since
EICCP is a special type of CPA applied to climate change issues. Literature on CPA adopts several
theoretical underpinnings to study the antecedents of firms’ policy engagement, which I
summarize below.
One theory used to understand CPA is legitimacy theory, which builds on the idea that
firms always attempt to obtain legitimacy from different reference groups in society and that CPA
provides firms with opportunities to be increasingly legitimate (R. Gray et al., 1995). Research on
CPA contends that engaging in CPA gives firms opportunities to communicate with and convince
policymakers that they are willing to comply, as well as to create a positive reputation when
pursuing constituency building through advertising campaigns and strategic public relations
(Banerjee and Venaik, 2018). Due to these potentials, prior literature contends that the intention
of enhancing organizational legitimacy motivates firms to engage in CPA (Lux et al., 2011).
A further theory that has been used is the institutional theory, which contends that
institutional forces push firms toward isomorphism (DiMaggio and Powell, 1983). Specific to the
91
adoption of CPA, institutional theory suggests that firms may be forced to engage in CPA because
the industry leaders or their competitors do so (e.g., Kim, 2008; Schuler et al., 2002).
Yet a third theory that justifies firms’ engagement in CPA is the resource dependence
theory, which is based on the tenet that business depends on public policy (Pfeffer and Salancik,
1978). Specifically, firms should engage more in CPA as the magnitude of their dependence on
public policy increases. That is to say, firms confronting an increased level of regulatory scrutiny
or more constraining and costly regulations are more motivated to manage such dependency
through CPA (Hart, 2001; Mitchell et al., 1997).
This literature review suggests that firms’ CPA engagement is a complicated strategic
decision and can be explained from different theoretical perspectives. To strengthen the
understanding of a particular type of CPA such as EICCP, incorporating several theories and
integrating different theoretical underpinnings are helpful.
4.2.2 Supply Chain Complexity Research
Supply chain complexity research has thrived over the last two decades with the popularity of
network research in the supply chain field. My literature review yields numerous empirical,
analytical, and conceptual pieces that investigate complexity as the main construct. I summarize
those studies in Table 3.3 of the Appendices. Since the present study is an empirical study, I
primarily focus on prior empirical and conceptual works to briefly discuss the concept, the metrics,
and the application of supply chain complexity in sustainability literature below.
First, the earliest conceptualization of supply chain complexity traces back to the seminal
work of Choi and Hong (2002), which defines it as the load on the network system that requires
coordination – “the higher the differentiation and the loose coupling among the elements in the
system, the higher the load required to coordinate the system” (Choi and Hong, 2002, p. 471).
92
Building on this, Skilton and Robinson (2009) specify supply network complexity as “a function
of the number of participants in the whole chain of relationships that ultimately connect consumers
to the means of production for specific goods and services, the level of differentiation between
participants, and the level and types of interrelationships that exist between participants” (Skilton
and Robinson, 2009, p. 42). Although later studies further differentiate structural complexity,
which refers to the “number and variety of elements defining the system” (Bode and Wagner, 2015,
p. 216), and dynamic complexity, which refers to the interactions among those elements (Bode
and Wagner, 2015), their conceptualization of supply network complexity is consistent with
Skilton and Robinson (2009). We, therefore, follow Skilton and Robinson’s (2009)
conceptualization in this study.
Second, although prior literature has consensus on the concept of supply chain complexity,
there is no agreement about its dimensions. A popular multi-faceted metric that has been applied
is the one by Choi and Hong (2002), who consider horizontal, vertical, and spatial complexity as
the three dimensions of supply chain complexity (Adhikary et al., 2020; Bode and Wagner, 2015).
The other metrics that prior literature has used include a single-facet metric focusing on the number
of nodes in the supply chain or the network size (Blackhurst et al., 2011; Kim et al., 2011; Wiedmer
et al., 2021), a two-dimensional metric addressing the number of nodes together with the number
of flows (Craighead et al., 2007), and a multi-dimensional metric capturing a series of
characteristics of upstream and downstream operations that include the number of nodes, the
differentiation among nodes, and the dispersion of nodes, among others (Bozarth et al., 2009;
Brandon-Jones et al., 2015). Bode and Wagner (2015) attribute those differences to the different
scopes of the studies, suggesting that some studies focus on the entire supply chain while others
are only interested in certain parts or segments of the supply chain. In this study, we focus on the
93
supply network, which is upstream of the supply chain. With this emphasis, we measure the three
dimensions with the prevalent metrics that consider the number, the types, and the differentiations
of relational ties between buyers and suppliers.
Third, most studies focus on the structural aspect of complexity (Adhikary et al., 2020; Lu
and Shang, 2017; Sharma et al., 2020). Lu and Shang (2017) justify this decision by contending
that structural complexity provides explicit measures, while dynamic complexity can possess
many different dimensions as the variety of interactions increases with the number and type of
elements in the supply chain. Further, Lu and Shang (2017) propose a characteristic to extend the
dimensions of structural complexity, which is the visibility of structural links. The authors argue
that horizontal, vertical, and spatial complexity are visible structural dimensions, while eliminative
and cooperative complexity, which measure the level of connections between the first-tier
suppliers and the focal buyer’s customers, and the level of connections among first-tier suppliers,
respectively, are not-so-visible dimensions. Given that this study focuses only on the relationship
ties between buyers and suppliers, such dynamic complexity is beyond the scope of our discussion.
I specifically investigate how firms’ EICCP is contingent on structural complexity of their supply
networks.
Fourth, empirical research on supply chain complexity covers a variety of topics, with
significant attention being paid to risks and disruptions. While several studies explore how supply
chain complexity is associated with plant- or firm-level operational, financial, environmental and
innovation performance (Adhikary et al., 2020; Bozarth et al., 2009; Lu and Shang, 2017; Sharma
et al., 2020), over half of the studies reviewed investigate the influence of supply chain complexity
on supply chain disruptions and resiliency, including the severity, frequency, impact and recovery
from disruptions (Blackhurst et al., 2011; Bode and Wagner, 2015; Brandon-Jones et al., 2015;
94
Craighead et al., 2007; Handley and Benton, 2013; Wiedmer et al., 2021). However, most of these
studies specifically focus on disruptions, without attending to other types of risks. To extend this
stream of research, I investigate the associations between supply network complexity and
regulatory risks.
A further stream of research, on which I rely primarily in this essay, studies sustainability
and environmental issues from the perspective of complexity. However, Adhikary et al. (2020) is
the only empirical piece that falls into this category, which focuses on structural complexity as
well as network embeddedness as antecedents to a focal firm’s greenhouse gas emissions. Another
relevant study is conceptual (Tachizawa and Wong, 2015), in which the authors discuss how
supply network complexity, specifically the number of suppliers, the number of interactions, and
the level of the interrelationship among suppliers, affect focal firms’ environmental performance.
Overall, the literature view shows that supply chain complexity has not been sufficiently
considered in sustainability research. Although a few studies examine supply network complexity
as antecedents of firms’ performance outcomes (Adhikary et al., 2020; Tachizawa and Wong,
2015), how supply network complexity influences firms’ overall sustainability strategy has not
been investigated. Given that EICCP constitutes a critical decision in sustainable operations, I seek
to understand, in this exploratory study, how supply network complexity serves as a contextual
factor for firms’ policy engagement when firms confront regulatory risks. This study not only
extends sustainability literature by clarifying the boundary conditions in which firms engage in
influence-oriented responses instead of adaptation-oriented responses to climate change policies
but also contributes to complexity literature by investigating the association between supply
network complexity to risks beyond disruptions.
95
4.3. Hypothesis Development
4.3.1 Regulatory Risks and EICCP
Regulatory risks, in the context of this essay, refer to risks to which firms are exposed due to the
development of climate change-related regulations and policies. These risks can originate from
regulatory uncertainty that is intrinsic to policy formulation as well as the high level of regulatory
stringency resulting from established and strict policies (Söderholm et al., 2015).
High levels of regulatory uncertainty motivate firms to take actions that either reduce the
uncertainty or protect them from the uncertainty, following the tenet of resource dependence
theory (Drees and Heugens, 2013). EICCP can be considered to be such actions. Prior literature
suggests that engaging in climate change policymaking at the early stage of the political process
not only offers firms opportunities to shape the policies but also grants these firms first-mover
advantages that facilitate policy compliance (Banerjee and Venaik, 2018). When the climate
change policies in development align with the interests of a firm, the firm can advocate the policies
to facilitate the strengthening of organizational legitimacy and obtain competitive advantages. In
contrast, when the current climate policy is not conducive to promoting the firm’s competitive
advantage, the firm can attempt to reshape it through lobbying or constituency building. Given that
firms are able to reduce regulatory uncertainty in the long run through EICCP, EICCP can be
positively associated with regulatory risks triggered by a high level of policy uncertainty.
High levels of regulatory stringency, however, may discourage firms to engage in
influencing policymaking. While regulatory uncertainty is higher in the agenda-setting stage,
which is an early stage of the political process, regulatory stringency rises as the political process
advances to the policy formulation stage. At this stage, the opportunities to change the established
policy instrument through EICCP are limited and require more investments (Meckling, 2015).
96
Prior literature has indicated that CPA does not lead to favorable outcomes when firms’ influence
on policies and the impact of the ultimately established policies are uncertain (Hadani et al., 2017).
Therefore, regulatory risks triggered by a high level of policy stringency should be negatively
related to EICCP.
In the context of climate change, recent research on policy processes has indicated that the
stages of agenda setting and policy formulation for climate change policies are intertwined
(Leppänen and Liefferink, 2022). The back-and-forth climate change policymaking signifies the
advancement from agenda setting to policy formulation (Bromley-Trujillo and Holman, 2020).
Therefore, I argue that regulatory risks associated with climate change policies originate from
regulatory stringency rather than from regulatory uncertainty. Since the high level of regulatory
stringency is negatively related to firms’ policy engagement, I theorize the following:
H1: Regulatory risks are negatively related to firms’ EICCP.
4.3.2 Moderating Effects of Supply Network Complexity on the Regulatory Risks-EICCP Link
The relationship between regulatory risks and EICCP is not only constrained by the essence of
regulatory risks, but also by firms’ capabilities of implementing firm-level political actions. The
lack of capabilities can nullify firms’ decision to engage in policymaking even when the regulatory
risks are high. In this study, I take a network perspective and investigate how supply network
complexity constrains firms’ EICCP capabilities. I examine three structural complexity
dimensions—horizontal complexity, vertical complexity, and spatial complexity—following prior
literature on supply chain complexity (Adhikary et al., 2020; Bode and Wagner, 2015; Lu and
Shang, 2017). According to the conceptualization of Lu and Shang (2017), horizontal complexity
measures the number of first-tier suppliers a firm has, which captures the width of the supply base;
vertical complexity measures the average number of second-tier suppliers each first-tier supplier
97
has, which reveals the depth of the supply base; and spatial complexity measures the number of
countries in which the firms’ supplier are located, which reflects the geographical spread of the
supply base. High supply network complexity in those dimensions indicates that focal firms should
consider numerous stakeholders that are experiencing different stages of the policymaking process
and residing in different regulatory contexts, and thus, are very likely to require a variety of
changes in climate change policies. On the one hand, the divergent demand requires greater
investment in EICCP, which makes the engagement decision less favorable for the focal firm. On
the other hand, the high level of supply network complexity makes the outcomes of EICCP
unpredictable. Engagement efforts that favor some suppliers may further increase the regulatory
pressure on others. Therefore, a highly complex supply network drains firms’ capabilities of
political engagement and thus demotivates firms’ EICCP, even when the regulatory risks are high.
Therefore, I propose the following hypotheses:
H2a. Horizontal complexity negatively moderates the regulatory risks-EICCP relationship
such that the higher the horizontal complexity, the more negative the regulatory risks-EICCP
relationship.
H2b. Vertical complexity negatively moderates the regulatory risks-EICCP relationship
such that the higher the vertical complexity, the more negative the regulatory risks-EICCP
relationship.
H2c. Spatial complexity negatively moderates the regulatory risks-EICCP relationship
such that the higher the vertical complexity, the more negative the regulatory risks-EICCP
relationship.
Figure 3.1 presents the theoretical model of this study.
98
Figure 3.1 Theoretical model
Supply network complexity
• Spatial complexity
• Vertical complexity
• Horizontal complexity
Regulatory risk exposure EICCP
4.4. Methodology
4.4.1. Data
To test my hypotheses, I compile the following datasets to construct a panel:
CDP Climate Change Dataset. As introduced in Chapter 3, this dataset collects firms’ public
responses to an information request sent by CDP on behalf of their signatory investors regarding
climate change issues. For this study, I continue to use firms’ responses to the open-ended
question— “On what issues have you been engaging directly with policymakers?” as the main data
source of firm-level engagements in influencing climate change policies. Firms can provide more
than one response to detail different types of engagement. Therefore, I aggregate responses at the
firm level for each year of observation. Each firm-year response includes firms’ descriptions of
their engagement and their proposed legislative solutions.
FactSet Revere-Supply Chain Relationship Dataset. This dataset offers a detailed mapping of a
firm’s various relationships with other stakeholders in the supply chain, including customers,
suppliers, partners, and competitors. I focus on firms’ direct suppliers and their suppliers’ suppliers
to construct a multi-tier supply network for each firm in the dataset. To do so, I take the following
99
steps to process the data: 1) I extract all the relationships tagged as supplier to obtain a list of firms’
first-tier suppliers. 2) I then extract all the relationships tagged as customer to complement and
cross-validate the list I obtained in step (1). Specifically, I reverse all these relationships to obtain
another list of the first-tier suppliers for the listed customers. 3) I then merge the two lists and keep
the unique data entries to construct a complete list of uniquely defined buyer-supplier relationships.
Treating buyers as focal firms, I count the number of first-tier suppliers for each focal firm. 4) I
identify the second-tier suppliers by searching for the suppliers of each first-tier supplier I found
in step (3). Then I aggregate the number of second-tier suppliers at the level of focal firms. 5) I
identify the country in which each first and second-tier supplier is located and count the total
number of countries existing for the focal firm level. Through these five steps, I obtain the number
of first-tier suppliers, the number of second-tier suppliers, and the number of countries in which
the suppliers in the first two tiers of a focal firm are located. I will detail how I construct the
variables of the supply network complexity in the next section.
Firm-Level Climate Change Exposure Dataset. This data was created by Sautner et al. (2022) to
quantify firm-level risk exposure related to opportunity and physical and regulatory shocks of
climate change. In this study, I use the metrics that capture firms’ exposure to regulatory risks in
climate change. To construct this risk variable, Sautner et al. (2022) conduct text analytics on firms’
earnings conference calls using a machine learning approach with the keyword discovery
algorithm. This method starts with predefining a short list of climate change bigrams as a training
library. The training library then serves as the input to the algorithm that calculates the probability
of classifying a certain sentence into the climate change-related category. Using 80% of the
probability as a threshold, Sautner et al. (2022) obtained over 700,000 sentences that potentially
mention climate change content and about 70 million sentences that do not. At this point, they
100
expand the initial list of climate change bigrams by discovering bigrams that appear frequently
and only in climate change-related sentences. They further classify climate change-related bigrams
into different categories based on two criteria. The first criterion focuses on whether a bigram
indicates opportunities or risks. The second criterion concentrates on topics including technologies,
regulations, and physical climate aspects. Combining the two criteria yields several lists of bigrams
including the one that the authors use to further calculate the regulatory risks. Specifically, the
authors divide the total number of potential bigrams in a text by the number of bigrams related to
a regulatory risk to construct a measure of regulatory risks.
Compustat Fundamental Dataset. This dataset offers information about firms’ profiles and
financial status, which serve as controls in this study.
The data compilation started with the CDP dataset, which yielded a panel that includes
1,309 firms and 8,529 firm-year responses spanning from 2013 to 2019. We then merged the panel
with the FactSet, Firm-Level Climate Change Exposure, and Compustat datasets to obtain network,
regulatory risk, and financial information, respectively, for firms in the panel. The associated
variables are discussed in the next section.
4.4.2. Variables
Dependent Variable. My dependent variable is EICCP, which is a count variable that uses the
number of strategies a firm pursues out of four basic EICCP strategies (promotion, support,
hedging, and opposition) in a given year as a proxy of the firm’s level of engagement. This
measurement is consistent with prior sustainability studies that use the count of sustainability
strategies to measure firms’ engagement in green practices (e.g., Chen and Ho, 2019; Peters et al.,
2019). I construct this variable based on the machine learning predictions obtained in Chapter 3.
Specifically, I conduct the text analysis on firms’ self-disclosure of EICCP using the random forest
101
approach, following the recent empirical finding that the random forest approach outperforms the
dictionary-based approach in text classification (Frankel et al., 2021). The prediction models yield
the probabilities that firms pursue one of the four basic EICCP strategies. I use 50% of the
probability as a threshold to identify the strategies a firm took. For example, if the machine learning
prediction outcomes show that in the year 2019, the probability that firm A pursued a promotion
strategy is 75%, the probability of pursuing a support strategy is 55%, a hedging strategy is 35%,
and an opposition strategy is 1%; this scenario would suggest that firm A engaged through
promotion and support strategies, but not through hedging or opposition strategies. The value of
firm A’s EICCP strategies in 2019 is therefore 2. Among the 22,314 firm-year observations
spanning from 2013 to 2019 in my sample, 4,720 firm-year observations feature one engagement
strategy, 642 are characterized by two strategies, and 7 by three strategies. The number of
observations for each case vary when I used a different probability as a threshold. I report those
numbers and present the results of the sensitivity analysis in Table 3.4 of the Appendices. To
briefly summarize the results here, applying a higher probability as the threshold (60% or 70%)
yields consistent regression results.
Independent variables. All independent variables were lagged by one year for my econometric
modeling and estimation (i.e., using observations from 2012 to 2018), considering that firms’
current strategies of EICCP depend on firms’ past perceptions of risks and resources. This is
consistent with extant CPA literature (Lux et al., 2011).
I follow prior literature to measure supply network complexity using three variables (Bode
and Wagner, 2015; Lu and Shang, 2017): horizontal complexity, which is measured by the number
of firms’ first-tier suppliers; vertical complexity, which is measured by the average number of the
102
second-tier suppliers per firms’ first-tier supplier; and spatial complexity, which is measured by
the total number of countries the first and second-tier suppliers come from.
To capture firms’ regulatory risks in climate change, I use a measure established by Sautner
et al. (2022), which accounts for the “relative frequency with which bigrams that capture regulatory
shocks related to climate change occur in the transcripts of analyst conference calls” by “counting
the number of such bigrams and divide by the total number of bigrams in a transcript” (Sautner et
al., 2022; p. 35).
Control variables. I include several time-varying variables in my model to control for potential
heterogeneity in firms’ EICCP strategies, consistent with CPA literature. These control variables
include: 1) firms’ annual sales and 2) the number of employees, which both capture firm size; these
are included since prior literature suggests that larger firms are more likely to engage intensively
in influencing policymaking (Kim, 2008; Schuler and Rehbein, 1997). 3) Firms’ sales growth, 4)
net income, 5) return on assets, and 6) market share, which all capture firms’ economic opportunity;
these are included since extant research posits that economic factors, including firms’ revenue
growth and profitability, can also lead to higher levels of policy engagement (Kim, 2008; Taylor,
1997; Zardkoohi, 1985). 6) Firms’ slack, measured by using firms’ current ratio, which captures
the availability of firms’ financial resources and provides financial support for CPA (Lenway and
Rehbein, 1991; Schuler and Rehbein, 1997); and 7) firms’ sector, which is a categorical variable
that controls for sector level differences in policy engagement. I also controlled for firm and year
fixed effects. Table 3.1 provides details of the variables I use in this study.
4.4.3. Econometric Models
Due to the count nature of the dependent variable, I fit a Poisson regression with panel data to
investigate the effects of supply network complexity and regulatory risks, as well as their
103
interactions on firms’ EICCP. In comparison to negative binomial regression analysis, which is
another model applicable to count variables, Poisson regression analysis relies on fewer
assumptions for the correct specification of the dispersion and has robust properties for the
estimation (Wooldridge, 2010). Specifically, while negative binomial regression requires the
occurrence of overdispersion as a function of the mean to produce consistent results, Poisson
regression makes no assumptions on the dispersion and can be robust to scenarios of both
underdispersion and overdispersion. Therefore, I apply Poisson regression to our dependent
variable EICCP. The regression model is as follows:
!"##$!" = '# ()*+,)-./0 #)12034+.5!,"%# + '& 73*.+8/0 #)12034+.5!,"%# +
'' 92/.+/0 #)12034+.5!,"%# + '( :3;<0/.)*5 :+=> !42)=<*3!,"%# +
') ()*+,)-./0 #)12034+.5!,"%# × :3;<0/.)*5 :+=> !42)=<*3!,"%# +
'* 73*.+8/0 #)12034+.5!,"%# × :3;<0/.)*5 :+=> !42)=<*3!,"%# +
'+ 92/.+/0 #)12034+.5!,"%# × :3;<0/.)*5 :+=> !42)=<*3!,"%# +
', #)-.*)0=!,"%# + '- 938.)*! + '#. @3/*!" + A!" ,
in which Controlsi,t-1 is a vector of all the time-varying control variables.
For model estimation, I first apply the random-effects approach. The likelihood-ratio test,
which compares the panel estimator with the pooled estimator, indicates that the random-effects
model is significantly different from the pooled model. In this case, the fixed-effects specification
can provide a better model fit (StataCorp, 2019). Therefore, I adopt the fixed-effect approach for
model estimation. To account for the potential overdispersion and heteroskedasticity, I also use
robust standard errors.
104
Table 3.1 Summary statistics and data sources
Variables Description Mean SD Data Source
EICCP Categorical, number of four basic strategies of EICCP 0.270 0.506 CDP Climate
(promotion, support, hedging, and opposition) a firm Change Dataset
take
Supply Network Complexity
Horizontal Complexity Number of tier-1 suppliers 20.706 31.609 FactSet Revere
Vertical Complexity Number of tier-2 suppliers per tier-1 supplier 18.554 22.124 Buyer-Supplier
Spatial Complexity Number of countries in which tier-1 and tier-2 9.947 10.770 Relationship Dataset
suppliers locate
Firms’ Regulatory Risk Number of such bigrams representing the regulatory 8.43 3.218 Firm-level Climate
Exposure risk of climate change divided by the total number of × 10-5 × 10-4 Change Exposure
bigrams in the text Dataset
Controls
Sales Log transformation of firms’ annual sales (in millions) 8.679 1.397
Employees Log transformation of the number of employees 2.933 1.296
Sales Growth Percentage change of the sales of the current year 0.275 19.785
compared to the sales of the previous year
Net Income Log transformation of firms' annual net income (in 10.051 0.143
millions) Compustat
Fundamental
ROA The ratio of firms’ net income and the average total 0.122 0.071
Dataset
assets
Market Share Firms’ annual sales divided by the total sales of the 0.216 0.285
industry over the same period
Slack Firms’ current ratio, which is their current assets 1.666 1.414
divided by their current liabilities
105
4.5. Results
I report the regression results in Table 3.2. Model 1 examines the main effects of regulatory risk
exposure on EICCP. I found no statistical evidence that regulatory risk exposure and EICCP are
significantly associated. H1 was therefore not supported. In Model 2, I added the interaction terms
between regulatory risk exposure and supply network complexity. The results show a negative
interaction between horizontal complexity and regulatory risk exposure (β = -0.060, p < 0.05),
supporting H2a. The absence of the main effect and the significant interaction indicate crossover
moderating effects of horizontal complexity, such that when the level of horizontal complexity is
high, regulatory risk exposure is negatively related to EICCP; however, when the level of
horizontal complexity is low, the association is positive. Similarly, the negative interactions
between vertical complexity and regulatory risk exposure (β = -0.078, p < 0.05) suggest that the
regulatory risk exposure is negatively related to EICCP when the level of vertical complexity is
high, and vice versa when the level of vertical complexity is low. H2b is therefore supported. An
examination of the interaction of spatial complexity and regulatory risk exposure yields different
results, suggesting that regulatory risk exposure and EICCP are positively related when the level
of spatial complexity is high (β = 0.209, p < 0.01). Therefore, H2c was rejected.
I plot the marginal effects of regulatory risk exposure on EICCP at different values of the
three supply network complexity aspects in Figure 3.2. The plots show that with the value of
horizontal complexity one standard deviation above the mean, regulatory risk exposure is
negatively related to EICCP (β = -1.535, p < 0.05); with the value of vertical complexity one
standard deviation above the mean, regulatory risk exposure is also negatively related to EICCP
(β = -1.990, p < 0.05); in contrast, with the value of spatial complexity one standard deviation
above the mean, regulatory risk exposure is positively related to EICCP (β = 1.289, p < 0.05).
106
Table 3.2 Results of the Poisson regression
Model 1 Model 2
coef se coef se
Independent Variables (one-year lagged)
Horizontal Complexity -0.000 (0.002) 0.000 (0.002)
Vertical Complexity -0.000 (0.002) 0.001 (0.002)
Spatial Complexity -0.004 (0.005) -0.007 (0.005)
Regulatory Risks -0.150 (0.344) -0.044 (0.479)
Horizontal Complexity
-0.060* (0.027)
× Regulatory Risks
Vertical Complexity
-0.078* (0.035)
× Regulatory Risks
Spatial Complexity
0.209** (0.071)
× Regulatory Risks
Control Variables (one-year lagged)
Sales -0.077 (0.181) -0.099 (0.181)
Employees -0.034 (0.159) -0.021 (0.155)
Sales Growth 0.057 (0.041) 0.060 (0.038)
Net Income -0.077 (0.258) -0.088 (0.258)
ROA -0.254 (0.755) -0.174 (0.739)
Market Share -0.488* (0.202) -0.494* (0.195)
Slack -0.066 (0.049) -0.066 (0.048)
Observations 1,624 1,624
Number of Firms 402 402
Log Pseudolikelihood -1012.2267 -1011.089
Note. Robust standard errors in parentheses. ** p<0.01, * p<0.05
107
Figure 3.2 Effects of regulatory risks on EICCP at different values of supply network complexity
Average marginal effects of one-year lagged regulatory Average marginal effects of one-year lagged regulatory
risk exposure with 95% CI risk exposure with 95% CI
Average marginal effects of one-year lagged regulatory
risk exposure with 95% CI
4.6. Discussion and Conclusion
In this study, I show that regulatory risks, supply network complexity, and their interactions
influence firms’ EICCP. I found neither the main effects of regulatory risks on EICCP nor the
main effects of supply network complexity on EICCP, suggesting that those factors do not
individually influence EICCP. However, I found the interactions between two dimensions of
supply network complexity, horizontal and vertical complexity, and regulatory risks are negatively
related to EICCP. This provides empirical evidence that considering more suppliers, or a bigger
108
extended supply chain, reduces firms’ propensity to engage in policymaking under regulatory
pressures. The interaction between spatial complexity and regulatory risks, in contrast, is positively
related to EICCP, contradicting my hypothesis. One potential interpretation is that a high level of
spatial complexity, reflected by a greater number of countries in which suppliers reside, increases
the regulatory risks firms perceive and thus, leads to a higher level of EICCP. Another
interpretation is that firms’ that have a higher level of complexity are likely to be large and
multinational companies that have more resources or capabilities to influence climate change
policymaking.
My research makes several contributions. First, this study builds on Chapter 3 to
econometrically assess EICCP, which provides face validity to the concept of EICCP. Although I
did not find any main effects of regulatory risks on EICCP, I found a negative sign for the
coefficient, indicating that regulatory risks are potentially negatively related to EICCP.
Second, as the first study in the supply chain field that empirically investigates the
antecedents of political engagement in a specific context, I explore supply network complexity
dimensions as contingency factors for firms’ EICCP under regulatory risks. The main effects of
the three dimensions of supply network complexity were missing, while the signs of the coefficient
of horizontal and vertical complexity are positive and the sign of the coefficient of spatial
complexity is negative. Despite the absent direct link, supply network complexity has significant
moderating effects on the regulatory risks-EICCP link. Integrating resource dependency theory
and network perspectives, I argue that network complexity constrains firms’ capabilities in
political engagement. My endeavor responds to the call for more policy-related studies in our field
from a supply chain perspective (Tokar and Swink, 2019).
109
This paper also extends the supply chain complexity literature by demonstrating the
relevance of supply network complexity for firms’ sustainability-related strategic decisions. While
prior literature on supply chain complexity focuses on the risk of supply chain disruptions, this
essay extends this stream of study by discussing the synergy between supply network complexity
and the regulatory risks that firms confront on EICCP. My work contributes to the risk
management literature contending that supply chain complexity causes more than one type of risk.
My research also has limitations. First, this study is exploratory in nature and only focuses
on one set of network-level factors when investigating the antecedents of EICCP. Future research
can integrate firm-level, industry-level, and country-level factors to construct a more
comprehensive framework. Interesting questions to ask include whether EICCP varies in different
industries or countries; and whether industrial factors and supply network factors constitute three-
way moderating effects on the link between supply network risks and EICCP. Second, while I
focus on the intensity of engagement in this study, the types of engagement strategies are also
worthwhile to study. For example, further studies can be conducted to examine whether different
risk perceptions and network characteristics lead to different types of EICCP engagement
strategies. Cluster analysis can be applied for this purpose.
110
APPENDICES
111
Table 3.3 Summary of Existing Research on Supply Chain Complexity
Study Research Metrics of network Method Data Source Dependent Independent
settings complexity variables variables
Risk Green
Empirical
Choi and No No Horizontal, vertical, Multi-case Interview of Formalization, Antecedents of
Hong, 2002 and spatial complexity study firm managers, centralization and supply network
measures, which refer firm complexity of structure,
to, respectively, the documents and supply network including:
average number of observations Formulized rules,
entities across all tiers, of plant visits norms and
the average number of policies;
entities in all possible Cost
vertical supply chains, consideration;
and the average Centralized
geographical distance approach;
between companies in Expensive product
the top two tiers in the lines;
network Use of core
supplier list;
etc.
112
Table 3.3 (cont’d)
Study Research Metrics of network Method Data Source Dependent Independent
settings complexity variables variables
Risk Green
Craighead Yes No Total number of nodes Multi-case Interview of Severity of supply Supply chain
et al., 2007 within a supply chain; study executives and chain disruptions density, supply
Total number of supply chain chain complexity,
forward, backward, and professionals node criticality;
within-tier materials of multiple Supply chain
flows in a supply chain supply chain mitigation
entities of a capabilities:
U.S.-based recovery
automobile capability and
manufacturer warning capability
(moderators)
Bozarth et No No Upstream complexity, Structrual Survey data of Plant-level Supply chain
al., 2008 including number of equation seven performance upstream
suppliers, long supplier modeling developed complexity,
lead times, supplier countries internal
delivery unreliability, manufacturing
and percentage of complexity, and
purchases imported; downstream
Downstream complexity
complexity, including
number of customers,
customer
heterogeneity, short
product life cycle, and
demand variability
113
Table 3.3 (cont’d)
Study Research Metrics of network Method Data Source Dependent Independent
settings complexity variables variables
Risk Green
Kim et al., No No Network size; Case Interview of Node-level Material flow or
2011 Network density, study with firm managers, network contractual
including network social firm characteristics relationship
density, core size, core network documents and including five network type
density, core to analysis observations centrality
periphery (CTP) of plant visits measures; network
density, and periphery level
to core (PTC) density characteristics
including
centralization and
complexity
measures
Blackhurst, Yes No Number of nodes in Multi-case Interview of Supply chain Resiliency
Dunn and supply chain study executives and resiliency enhancers
Craighead, supply chain including human
2011 professionals capital,
of multiple organizational and
supply chain interorganizational
entities of a capital, and
U.S.-based physical capital;
automobile Resiliency
manufacturer reducers including
flow activities
(e.g. number of
nodes in supply
chain), flow units,
and source of flow
units
114
Table 3.3 (cont’d)
Study Research Metrics of network Method Data Source Dependent Independent
settings complexity variables variables
Risk Green
Hadley and Yes No Location-specific OLS Survey data of Inter- Task-specific
Benton, complexity including regression U.S. public organizational complexity
2013 geographic dispersion, companies management costs including scale of
geographic distance, including control service, breadth of
and cultural distance costs and tasks, and service
coordination costs customization;
Location-specific
complexity
including
geographic
dispersion,
geographic
distance, and
cultural distance
Bode and Yes No Supply network Negative Survey data of Frequency of Supply network
Wager, horizontal, vertical and multinomial firms in supply chain horizontal,
2015 spatial complexity regression German, disruptions vertical and
Austria and spatial complexity
Switzerland,
and archival
data
115
Table 3.3 (cont’d)
Study Research Metrics of network Method Data Source Dependent Independent
settings complexity variables variables
Risk Green
Brandon- Yes No Scale as the number of OLS Survey data of Frequency of Scale complexity,
Jones et al., suppliers; regression manufacturing supply chain differentiation
2015 Differentiation as the firms in the disruptions; complexity,
degree of difference in United Plant performance delivery
size and technical Kindom complexity, and
capability between geographic
suppliers; dispersion
Delivery reliability by complexity;
on-time performance Production
and lead-time; capacity, safety
Geographic dispersion stock at suppliers
as an index developed and at plant, and
by Stock, Greis, and visibility
Kasarda (2000). (moderators)
Lu and No No Supply network OLS Archival data Focal firm's Supply network
Shang, horizontal, vertical, regression financial horizontal,
2017 spatial, eliminative and performance vertical, spatial,
cooperative complexity eliminative and
cooperative
complexity
116
Table 3.3 (cont’d)
Study Research Metrics of network Method Data Source Dependent Independent
settings complexity variables variables
Risk Green
Sharma et No No Supply network Control Archival data Focal firm's Focal firm's
al., 2020 horizontal, vertical and function innovation supply network
spatial complexity instrumental horizontal,
variable vertical and
panel spatial
regression complexity; focal
firm's strategic
emphasis and
influence over the
network
Adhikary et No Yes Supply network Control Archival data Focal firm's green Focal firm's
al., 2020 horizontal, vertical and function (Bloomberg house gas supply network
spatial complexity instrumental SPLC, FA and emissions horizontal,
variable ESG, GRI, vertical and
panel and CDP spatial
regression database) complexity; focal
firm's
betweenness
centrality and
reach
Wiedmer et Yes No Supply complexity as Difference- Archival data Supply network Supply
al., 2021 nodes in the network in- (Panjiva, resilience complexity (nodes
differences import and including in the network),
models export data of disruption impact logistics
automotive and disruption complexity (arcs
industry) recovery in the network),
and product
complexity
117
Table 3.3 (cont’d)
Study Research Metrics of network Method Data Source Dependent Independent
settings complexity variables variables
Risk Green
Simulation and Modeling
Basole and Yes No Network topology: Simulation N/A Risk propogation Supply network
Bellamy, Random; with OLS and recovery in structure (small-
2014 Small-world; regression supply network world vs. scale-
Scale-free free); network
structural
visibility; initial
level of healthy
entities in supply
network
Giannoccaro, No No Number of nodes in Simulation N/A Supply network Focal firm's
Nair and supply network; adaptive scope of control;
Choi, 2017 Supply interactions performance supply network
complexity,
including supply
interactions and
number of firms
Demirel et Yes No Horizontal complexity; Analytical Archival data Stability of supply Dynamics of
al., 2019 Vertical complexity; study of selected networks material flows
Degree heterogeneity; (generalized firms for case and inventory
Interrelatedness modeling study level
between suppliers method)
and case
study
118
Table 3.3 (cont’d)
Study Research Metrics of network Method Data Source Dependent Independent
settings complexity variables variables
Risk Green
Conceptual
Vachon and No No Network level: Conceptual N/A Defining supply
Klassen, Technological study chain complexity
2002 dimension of the from technological
supply chain; and information
Information processing processing
dimension of dimensions at
complexity network level
Choi and Yes No Supply base level: Conceptual N/A Number of
Krause, Number of suppliers; study suppliers, Degree
2006 Degree of of differentiation,
differentiation among Level of
these suppliers; interrelationships
Level of inter-
relationships among
the suppliers
Skilton and No No Number of suppliers; Conceptual N/A Traceability of Supply network
Robinson, Differentiation of study adverse events complexity,
2009 suppliers; degree of tight
Level of coupling and
interrelationship transparency
between suppliers
119
Table 3.3 (cont’d)
Study Research Metrics of network Method Data Source Dependent Independent
settings complexity variables variables
Risk Green
Pathak, Wu No No Network level: Conceptual N/A Co-opetitive
and Community supply study dynamics in four
Johnston, network; network
2014 Federal supply archetypes,
network; including
Consortium supply community,
network federation,
Hierarchical supply consortium and
network hierarchy supply
networks
Yan et al., No No Network level: Conceptual N/A Conceptualizing
2015 Operational nexus study and identifying
supplier; three types nexus
Monopolistic nexus suppliers, which
supplier; influence focal
Informational nexus firms' operational
supplier performance
differently
Tachizawa No Yes Number of suppliers; Conceptual N/A Focal firm's Green SCM
and Wong, Number of interactions study environmental formal/ informal
2015 among suppliers; performance governance
Level of mechanism;
interrelationship supply network
between suppliers complexity;
centralization;
density
120
Table 3.4 Sensitivity Tests
Model 1 Model 2
coef se coef se
Independent Variables (one-year lagged)
Horizontal Complexity -0.000 (0.002) 0.001 (0.002)
Vertical Complexity 0.000 (0.002) 0.003 (0.002)
Spatial Complexity -0.007 (0.005) -0.004 (0.007)
Regulatory Risks -0.299 (0.571) 0.229 (0.749)
Horizontal Complexity
-0.128** (0.040) -0.094+ (0.054)
× Regulatory Risks
Vertical Complexity
-0.145* (0.067) -0.162+ (0.086)
× Regulatory Risks
Spatial Complexity
0.451*** (0.128) 0.368* (0.187)
× Regulatory Risks
Control Variables (one-year lagged)
Sales 0.094 (0.227) -0.314 (0.275)
Employees -0.324 (0.210) -0.260 (0.265)
Sales Growth 0.082* (0.039) 0.117** (0.041)
Net Income 0.161 (0.343) -0.162 (0.428)
ROA -1.099 (0.826) 0.096 (1.209)
Market Share -0.216 (0.313) -0.095 (0.437)
Slack -0.065 (0.061) 0.024 (0.070)
Observations 1,487 1,487
Number of Firms 365 310
Log Pseudolikelihood -865.612 -685.966
Note. Robust standard errors in parentheses. *** p<0.001, ** p<0.01, * p<0.05, + p<0.1. Model
1 refer to the case that I use 60% of the probability, instead of 50% used in the main analysis, as
a threshold to identify the strategies a firm took. With this threshold, among the 22,314 firm-year
observations spanning from 2013 to 2019 in my sample, 4,195 firm-year observations feature
one engagement strategy, and 358 are characterized by two strategies. Model 2 refer to the case
that I use 70% of the probability as a threshold to identify the strategies a firm took. With this
threshold, 3,386 firm-year observations feature one engagement strategy, and 174 are
characterized by two strategies. The results of both Model 1 and Model 2 are consistent with the
results of the main analysis reported in Table 3.2.
121
REFERENCES
122
REFERENCES
Adhikary, A., Sharma, A., Diatha, K. S., and Jayaram, J. (2020). Impact of Buyer-Supplier
Network Complexity on Firms’ Greenhouse Gas (GHG) Emissions: An Empirical
Investigation. International Journal of Production Economics, 230(July), 107864.
Ambitious Corporate Climate action - Science Based Targets. (2022).
https://sciencebasedtargets.org/.
Banerjee, S., and Venaik, S. (2018). The Effect of Corporate Political Activity on MNC
Subsidiary Legitimacy: An Institutional Perspective. Management International Review,
58(5), 813–844.
Blackhurst, J., Dunn, K. S., and Craighead, C. W. (2011). An Empirically Derived Framework of
Global Supply Resiliency. Journal of Business Logistics, 32(4), 374–391.
Bode, C., and Wagner, S. M. (2015). Structural Drivers of Upstream Supply Chain Complexity
and the Frequency of Supply Chain Disruptions. Journal of Operations Management, 36(1),
215–228.
Bozarth, C. C., Warsing, D. P., Flynn, B. B., and Flynn, E. J. (2009). The Impact of Supply
Chain Complexity on Manufacturing Plant Performance. Journal of Operations
Management, 27(1), 78–93.
Brandon-Jones, E., Squire, B., and Van Rossenberg, Y. G. T. (2015). The Impact of Supply Base
Complexity on Disruptions and Performance: The Moderating Effects of Slack and
Visibility. International Journal of Production Research, 53(22), 6903–6918.
Bromley-Trujillo, R., and Holman, M. R. (2020). Climate Change Policymaking in the States: A
View at 2020. Publius: The Journal of Federalism, 50(3), 446–472.
Chen, C. M., and Ho, H. (2019). Who Pays You To Be Green? How Customers’ Environmental
Practices Affect the Sales Benefits of Suppliers’ Environmental Practices. Journal of
Operations Management, 65(4), 333–352.
Choi, T. Y., and Hong, Y. (2002). Unveiling the Structure of Supply Networks: Case Studies in
Honda, Acura, and DaimlerChrysler. Journal of Operations Management, 20(5), 469–493.
Clapp, J., and Meckling, J. (2013). Business as a Global Actor. In The Handbook of Global
Climate and Environment Policy (pp. 286–303). John Wiley & Sons Ltd.
Craighead, C. W., Blackhurst, J., Rungtusanatham, M. J., and Handfield, R. B. (2007). The
Severity of Supply Chain Disruptions: Design Characteristics and Mitigation Capabilities.
Decision Sciences, 38(1), 131–156.
DiMaggio, P. J., and Powell, W. W. (1983). The Iron Cage Revisited Institutional Isomorphism
and Collective Rationality in Organizational Fields. American Sociological Review, 48(2),
123
147–160.
Drees, J. M., and Heugens, P. P. M. A. R. (2013). Synthesizing and Extending Resource
Dependence Theory. Journal of Management, 39(6), 1666–1698.
Engel, H., Enkvist, P.-A., and Henderson, K. (2015). How companies can adapt to climate
change. https://www.mckinsey.com/~/media/McKinsey/Business
Functions/Sustainability/Our Insights/How companies can adapt to climate change/How
companies can adapt to climate change.pdf?shouldIndex=false.
Frankel, R., Jennings, J., and Lee, J. (2021). Disclosure Sentiment: Machine Learning vs.
Dictionary Methods. Management Science.
Gray, R., Kouhy, R., and Lavers, S. (1995). Corporate Social and Environmental Reporting A
Review of the Literature and a Longitudinal Study of UK Disclosure. Accounting, Auditing
& Accountability Journal, 8(2), 47–77.
Greiner, M., and Kim, J. (2021). Corporate Political Activity and Greenwashing: Can
CPA Clarify Which Firm Communications on Social & Environmental
Events Are Genuine? Corporate Social Responsibility and Environmental Management,
28(1), 1–10.
Grover, A. K., and Dresner, M. (2022). A Theoretical Model on How Firms Can Leverage
Political Resources To Align With Supply Chain Strategy for Competitive Advantage.
Journal of Supply Chain Management, 58(2), 48–65.
Hadani, M., Bonardi, J.-P., and Dahan, N. M. (2017). Corporate Political Activity, Public Policy
Uncertainty, and Firm Outcomes: A Meta-Analysis. Strategic Organization, 15(3), 338–
366.
Handley, S. M., and Benton, W. C. (2013). The Influence of Task- and Location-Specific
Complexity on the Control and Coordination Costs in Global Outsourcing Relationships.
Journal of Operations Management, 31(3), 109–128.
Hart, D. M. (2001). Why Do Some Firms Give? Why Do Some Give a Lot? High-Tech PACs,
1977-1996. Journal of Politics, 63, 1230–1249.
Hillman, A. J., Keim, G. D., and Schuler, D. (2004). Corporate Political Activity: A Review and
Research Agenda. Journal of Management, 30(6), 837–857.
InfluenceMap. (2021). Big Tech and Climate Policy. https://influencemap.org/report/Big-Tech-
and-Climate-Policy-afb476c56f217ea0ab351d79096df04a.
Kim, J.-H. (2008). Business and Politics Corporate Lobbying Revisited Corporate Lobbying
Revisited *.
Kim, Y., Choi, T. Y., Yan, T., and Dooley, K. (2011). Structural Investigation of Supply
Networks: A Social Network Analysis Approach. Journal of Operations Management,
124
29(3), 194–211.
Lenway, S. A., and Rehbein, K. (1991). Leaders, Followers, and Free Riders: An Empirical Test
of Variation in Corporate Political Involvement. Academy of Management Journal, 34(4),
893–905.
Leppänen, T., and Liefferink, D. (2022). Agenda-Setting, Policy Formulation, and the EU
Institutional Context: The Case of the Just Transition Fund. European Policy Analysis, 8(1),
51–67.
Lu, G., and Shang, G. (2017). Impact of Supply Base Structural Complexity on Financial
Performance: Roles of Visible and Not-so-Visible Characteristics. Journal of Operations
Management, 53–56(March 2016), 23–44.
Lux, S., Crook, T. R., and Woehr, D. J. (2011). Mixing Business With Politics: A Meta-Analysis
of the Antecedents and Outcomes of Corporate Political Activity. Journal of Management,
37(1), 223–247.
Martin, C. J. (1995). Nature or Nurture? Sources of Firm Preference for National Health Reform.
American Political Science Review, 89(4), 898–913.
McCarthy, N. (2019). Oil And Gas Giants Spend Millions Lobbying To Block Climate Change
Policies. Forbes. https://www.forbes.com/sites/niallmccarthy/2019/03/25/oil-and-gas-
giants-spend-millions-lobbying-to-block-climate-change-policies-
infographic/?sh=6797aac87c4f.
Meckling, J. (2015). Oppose, Support, or Hedge? Distributional Effects, Regulatory Pressure,
and Business Strategy in Environmental Politics. Global Environmental Politics, 15(2), 19–
37.
Mitchell, N. J., Hansen, W. L., and Jepsen, E. M. (1997). The Determinants of Domestic and
Foreign Corporate Political Activity. Journal of Politics, 59(4), 1096–1113.
Ojala, M., and Hallikas, J. (2006). Investment Decision-Making in Supplier Networks:
Management of Risk. International Journal of Production Economics, 104(1), 201–213.
Peters, G. F., Romi, A. M., and Sanchez, J. M. (2019). The Influence of Corporate Sustainability
Officers on Performance. Journal of Business Ethics, 159(4), 1065–1087.
Pfeffer, J., and Salancik, G. (1978). The External Control of Organizations: A Resource
Dependence Perspective. In Harper & Row.
RE100. (2022). https://www.there100.org/.
Reichart, E. (2019). 3 Ways Business Must Use Political Influence to Champion Climate
Ambition. World Resources Institute. https://www.wri.org/blog/2019/04/3-ways-business-
must-use-political-influence-champion-climate-ambition.
125
Sadrich, F., and Annavarjulia, M. (2002). Antecedents of Corporate Lobbying Participation and
Intensity: A Review of the Literature. Public Administration Quarterly, 26(3/4), 465–502.
Sautner, Z., van Lent, L., Vilkov, G., and Zhang, R. (2022). Firm-level Climate Change
Exposure.
Schuler, D. A., and Rehbein, K. (1997). The Filtering Role of the Firm in Corporate Political
Involvement. Business & Society, 36(2), 116–139.
Schuler, D. A., Rehbein, K., and Cramer, R. D. (2002). Pursuing Strategic Advantage Through
Political Means: A Multivariate Approach. Academy of Management Journal, 45(4), 659–
672.
Sharma, A., Pathak, S., Borah, S. B., and Adhikary, A. (2020). Is It Too Complex? The Curious
Case of Supply Network Complexity and Focal Firm Innovation. Journal of Operations
Management, 66(7–8), 839–865.
Skilton, P. F., and Robinson, J. L. (2009). Traceability and Normal Accident Theory: How Does
Supply Network Complexity Influence the Traceability of Adverse Events? Journal of
Supply Chain Management, 45(3), 40–53.
Söderholm, K., Söderholm, P., Helenius, H., Pettersson, M., Viklund, R., Masloboev, V.,
Mingaleva, T., and Petrov, V. (2015). Environmental Regulation and Competitiveness in
the Mining Industry: Permitting Processes With Special Focus on Finland, Sweden and
Russia. Resources Policy, 43, 130–142.
StataCorp. (2019). Xtpoisson. In Stata 16 Base Reference Manual. TX: Stata Press.
Tachizawa, E. M., and Wong, C. Y. (2015). The Performance of Green Supply Chain
Management Governance Mechanisms: A Supply Network and Complexity Perspective.
Journal of Supply Chain Management, 51(3), 18–32.
Taylor, D. F. (1997). The Relationship between Firm Investments in Technological Innovation
and Political Action. Southern Economic Journal, 63(4), 888.
Tokar, T., and Swink, M. (2019). Public Policy and Supply Chain Management: Using Shared
Foundational Principles to Improve Formulation, Implementation, and Evaluation. Journal
of Supply Chain Management, 55(2), 68–79.
Veena Ramani, C. (2020). Blueprint for Responsible Policy Engagement on Climate Change.
The Harvard Law School Forum on Corporate Governance.
https://corpgov.law.harvard.edu/2020/08/03/blueprint-for-responsible-policy-engagement-
on-climate-change/.
Wiedmer, R., Rogers, Z. S., Polyviou, M., Mena, C., and Chae, S. (2021). The Dark and Bright
Sides of Complexity: A Dual Perspective on Supply Network Resilience. Journal of
Business Logistics, 1–24.
126
Wooldridge, J. M. (2010). Econometric Analysis of Cross Section and Panel Data. In MIT Press.
Zardkoohi, A. (1985). On the Political Participation of the Firm in the Electoral Process.
Southern Economic Journal, 51(3), 804.
127