THE INFLUENCE OF CONTEXT ON THE EFFECTIVENESS OF BOUNDARY SPANNING IN ENHANCING SUPPLY CHAIN RESILIENCE: DOES ONE SIZE FIT ALL?

THE INFLUENCE OF CONTEXT ON THE

EFFECTIVENESS OF BOUNDARY SPANNING IN

ENHANCING SUPPLY CHAIN RESILIENCE: DOES ONE

SIZE FIT ALL?

MSc (Res) Thesis

Mitchell J. van den Adel (s2319373)

Supervisors:

dr. T.A. (Thom) de Vries prof. dr. D.P. (Dirk Pieter) van Donk

University of Groningen Faculty of Economics and Business

MSc (Res) Economics and Business Operations Management and Operations Research

July 2017

ABSTRACT

boundary spanning for increasing SC resilience differs across contexts, we take a first step toward more critically reflecting upon the costs and inefficiencies of resilience practices. Moreover, we successfully integrate two research domains of differing levels of analysis – those of boundary spanning and SC resilience – that have previously developed largely independently of each other. Our use of mixed methods further allows us to contribute more quantitative tests to the emerging SC resilience literature while concurrently seeking to obtain a richer understanding of this complex phenomenon.

Keywords: Supply chain resilience, Boundary spanning, Disruption characteristics

1. INTRODUCTION

In 2016, seventy percent of the organizations experienced at least one supply chain (SC) disruption ranging from unplanned telecommunications outages and data breaches to adverse weather conditions and acts of terrorism. Over thirty percent of these organizations suffered losses in excess of one million EUR from a single incident (Alcantara and Riglietti, 2016). One way for organizations to reduce these losses is by purposefully collaborating across organizational boundaries with, for example, suppliers and customers (e.g., Bode et al., 2011; Scholten and Schilder, 2015; Tukamuhabwa et al., 2015). Through such cross-boundary collaboration or boundary spanning (Marrone, 2010), organizations can gain access to the external resources, information, and support they need for detecting and resolving SC disruptions (Bode et al., 2011; Quick and Feldman, 2014). As such, boundary spanning may be an important antecedent of SC resilience (i.e., the ability to prepare for, respond to, and recover from disruptions: Tukamuhabwa et al., 2015).

These benefits for SC resilience notwithstanding, there is, however, also empirical micro-level team research that has identified boundary spanning as particularly challenging and costly (e.g., De Geest et al., 2016; Fugate et al., 2012; Gibson and Dibble, 2013). Building and maintaining boundary spanning linkages with many suppliers and customers is, for example, extremely labor intensive. Boundary spanning can, as such, easily overload organizational members and distract them from other, potentially more important tasks (Ancona and Caldwell, 1992). Boundary spanning may, for instance, divert members from alternative (internal) resilience practices such as building up slack resources and increasing production flexibility (e.g., Bode et al., 2011; Mishra et al., 2016). By subsequently heightening the impact of SC disruptions, boundary spanning linkages may even “become a source of supply chain risk” (Jüttner, 2005, p. 122). Considering these costs and risks, there may also be situations or contexts in which boundary spanning hurts – rather than helps – SC resilience. Contemporary research on SC resilience and boundary spanning has, however, largely overlooked this potential “dark side” of boundary spanning (De Geest et al., 2016; Marrone, 2010).

boundary spanning outweigh its costs when organizations face SC disruptions with either high levels of complexity or uncertainty. When facing such disruptions, organizations need boundary spanning to make sense of and resolve these disruptions, thereby increasing SC resilience. In contrast, we propose that the costs of boundary spanning overshadow its benefits when organizations face less complex or uncertain SC disruptions. In such situations, organizations may be able to mitigate the impact of disruptions without exhibiting costly boundary spanning efforts that may potentially harm SC resilience. We adopt a mixed methods research design and largely find support for our predictions.

By exploring under which disruption characteristics boundary spanning will help or hurt SC resilience, we contribute to extant literature in several important ways. First and foremost, we are among the first to empirically develop a more nuanced perspective on strategies aimed at increasing SC resilience. We believe that a contingency framework is particularly appropriate for studying under which conditions such strategies are beneficial or, rather, harmful for SC resilience. Second, we contribute to the relatively limited empirical research examining the contingency aspects of the relationship between boundary spanning and its desired performance outcomes. Further, we illustrate how integrating interdisciplinary micro-level (team boundary spanning) and macro-micro-level (SC resilience) research enables important theoretical advances in Operations Management research. A final contribution is that our mixed method research design allows us to provide more quantitative tests of SC resilience while concurrently developing a richer understanding of this complex phenomenon.

The remainder of this thesis is structured as follows. In the following section, we explore and integrate the concepts of SC resilience and boundary spanning, leading to our hypotheses development. Next, we consider in more detail the context and approach of our mixed methods research design. Afterward, we present and reflect upon our findings. Finally, we conclude this thesis by discussing theoretical and managerial implications, presenting limitations, and identifying recommendations for future research.

2. THEORETICAL BACKGROUND AND HYPOTHESES

DEVELOPMENT

2.1 Supply Chain Resilience

focused on the activities that organizations undertake to become resilient. An illustrative example of this perspective is the research of Scholten and Schilder (2015), who explore the mechanisms through which collaboration may lead to a resilient SC.

Much less empirical research, however, has adopted an outcome perspective of SC resilience (Hohenstein et al., 2015; Kamalahmadi and Parast, 2016). An outcome perspective conceptualizes SC resilience by whether SCs are successful in maintaining and restoring performance levels after encountering actual disruptions (Britt, 1988) and are, thus, able to “bounce back” (Sheffi and Rice, 2005). With only limited (empirical) research adopting an outcome perspective of SC resilience, we are unable to determine whether the examined resilience strategies and antecedents indeed allow organizations to restore performance levels and, if so, in which timeframe. As such, there is a clear need for more empirical and quantitative research assessing and measuring SC resilience in order to advance our understanding of this complex phenomenon (Van der Vegt et al., 2015). To address this need in contemporary SC resilience literature, we adopt an outcome perspective and define SC resilience as the “length of time a system takes to recover to its pre-jolt level of performance after experiencing a drop off in performance” (Britt, 1988, p. 60).

2.2 Boundary Spanning

Boundary spanning involves a member’s efforts to establish linkages and manage interactions with individuals from different teams or organizations for task-related purposes (Marrone, 2010; Marrone et al., 2007). Boundary spanning efforts may, for example, be directed at searching for relevant outside information, externally representing and buffering the organization, or coordinating activities with other, interdependent organizations (e.g., Ancona and Caldwell, 1992; Joshi et al., 2009; Marrone, 2010). Boundary spanning, therefore, connects an organization with stakeholders (e.g., suppliers, customers, competitors) in its focal environment (Aldrich and Herker, 1977; Thompson, 1967). Because boundary spanning subsequently enables an organization to achieve its goals by attending to demands and changes in its external environment, boundary spanning is frequently associated with organizational learning, innovation, and performance (Joshi et al., 2009; Marrone, 2010).

In SC resilience literature, a principal form of boundary spanning, inter-organizational collaboration, is recognized as a primary strategy to increase SC resilience (e.g., Hohenstein et al., 2015; Scholten and Schilder, 2015; Tukamuhabwa et al., 2015). However, previous SC resilience research, which is mainly conducted at the macro- or organizational-level, is largely inconsiderate of micro-level team research that has empirically identified boundary spanning as particularly challenging and costly (e.g., Faraj and Yan, 2009; De Geest et al., 2016; Gibson and Dibble, 2013). Our aim is to integrate these interdisciplinary streams of micro-level (team boundary spanning) and macro-micro-level (SC resilience) research to advance our understanding of whether the “dark side” of boundary spanning affect its relationship with SC resilience.

2.3 The Relationship between Boundary Spanning and Supply Chain Resilience

performance levels after a disruption. On the one hand, boundary spanning may be beneficial for SC resilience by providing an organization with access to the external resources, information, and support they need for quickly detecting and resolving SC disruptions (Bode et al., 2011; Mishra et al., 2016; Quick and Feldman, 2014). Boundary spanning also assists an organization in mapping changes in internal and external demands (Ancona, 1990). This awareness of internal and external dynamics confines the impact of disruptions by preventing overreactions and unnecessary interventions (Christopher and Lee, 2004; Jüttner and Maklan, 2011). Boundary spanning further strengthens information processing and inter-organizational coordination (Marrone, 2010). This helps organizations in managing their interdependencies with diverse external stakeholders whose exact needs might be unknown, thus allowing organizations to better cope with complexities and uncertainties in their environment (De Geest et al., 2016).

On the other hand, boundary spanning is not free of costs and, in some contexts, these costs may overrule its benefits for SC resilience. Excessive boundary spanning may, for example, result in unnecessarily complex SC relationships as coordination costs and flows of information increase without corresponding benefits (Prater et al., 2001). Such complexity, in turn, increases the severity and frequency of SC disruptions (Bode and Wagner, 2015; Brandon-Jones et al., 2014). Building and maintaining boundary spanning linkages is, furthermore, extremely labor and resource intensive. Boundary spanning can, as such, easily overload (Aldrich and Herker, 1977) and stress (Marrone et al., 2007) organizational members, distracting them from other, potentially more important tasks (Ancona and Caldwell, 1992). Boundary spanning may, for instance, divert members from alternative (internal) resilience practices such as building up slack resources and increasing production flexibility (e.g., Bode et al., 2011; Mishra et al., 2016). Alternatively, an organization may inadvertently increase recovery time by using time-consuming boundary spanning to source for external information and support, while it has the appropriate resources and capabilities internally available for dealing with a disruption (Choi, 2002; De Geest et al., 2016).

Despite these conflicting implications of boundary spanning, few scholars have empirically examined the “dark side” of boundary spanning (De Geest et al., 2016; Marrone, 2010), especially in the context of resilience. Considering the subsequent ambiguous implications of boundary spanning for SC resilience, we ground our study in contingency theory and propose that the benefits of boundary spanning outweigh its costs in some contexts, but not in others. When, however, inconsiderate of the context in which boundary spanning is performed, its overall benefits and costs for SC resilience may cancel each other out. We therefore do not expect a direct relationship between boundary spanning and SC resilience. This idea that boundary spanning is not a universally effective strategy to increase resilience is consistent with early boundary spanning literature (e.g., Ancona, 1990; Ancona and Caldwell, 1992; Choi, 2002) that has identified context characteristics as important moderators in the relationship between boundary spanning and performance outcomes.

2.4 The Moderating Role of Disruption Characteristics

Thompson, 1967). Contingency theory holds that the effectiveness of organizational practices is contingent on environmental or situational characteristics (e.g., complexity, uncertainty, turbulence). Organizations, in turn, adopt internal structures and processes that fit the environment or situation to maximize organizational performance (Burns and Stalker, 1961; Donaldson, 2001). Similarly, when applied to the concept of SC resilience, contingency theory suggests that practices aimed at increasing resilience should align with the characteristics of the environment or disruption that an organization faces to maximize effectiveness. Because of its predominant focus on “economic efficiency and intentional form of rationality”, Sousa and Voss (2008, p. 703) propose that the development of the field of Operations Management has been heavily informed by a contingency paradigm (cf. Boer et al., 2017). It is, therefore, surprising that previous SC resilience research has devoted little attention to the implications of matching resilience practices with environmental or situational characteristics.

Adding to extant literature, we build on contingency theory to make sense of the conflicting implications of boundary spanning for SC resilience. That is, we argue that organizations should match their use of boundary spanning with the characteristics of the disruption they face in order for boundary spanning to aid SC resilience. Organizational and contingency scholars across a wide variety of domains have characterized contexts along dimensions of complexity, uncertainty, and stability (e.g., Drach-Zahavy and Somech, 2010; Duncan, 1972; Lawrence and Lorsch, 1967; Thompson, 1967). Because disruptions are inherently dynamic (i.e., they have a high rate of change or turbulence compared to regular business conditions; Kovach et al., 2015), we specifically focus on the moderating roles of complexity and uncertainty. In other words, we examine how disruption complexity and disruption uncertainty may determine the extent to which boundary spanning will help or hurt SC resilience. Our focus is comparable to previous research on boundary spanning that has examined the moderating roles of task complexity and uncertainty (e.g., Ancona and Caldwell, 1992; Choi, 2002).

2.4.1 Disruption Complexity

An organization’s expertise, equipment, and other key resources (e.g., time, effort, personnel) may be insufficient for resolving disruptions high in complexity (De Geest et al., 2016). As a result, efforts of individual organizations are frequently inadequate for adopting comprehensive and effective countermeasures to such disruptions. In resolving complex disruptions, organizations are thus forced to look beyond their own boundaries toward the tangible resources, information, and support residing at SC partners. To readily gain access to these external resources, organizations may establish and maintain boundary spanning linkages with other organizations (Bode et al., 2011; Quick and Feldman, 2014). These boundary spanning linkages further facilitate better inter-organizational coordination and joint problem-solving (Galbraith, 1977; Koufteros et al., 2005). This is imperative as resource allocation becomes increasingly important for responding to complex disruptions. That is, inter-organizational coordination and joint problem-solving prevent organizations from misdirecting and depleting resources and expertise on redundant or inefficient operations and processes (Craighead et al., 2007; Marrone, 2010). Through engaging in boundary spanning, organization can orchestrate quicker and more effective solutions to complex disruptions, thereby increasing SC resilience.

For disruptions low in complexity, however, organizations may be able to restore processes and performance levels without engaging in costly boundary spanning activities. Organizations may, for example, draw from standardized operational procedures (e.g., safety stock, backup capacity) to deal with less complex disruptions in a timely manner (Bode et al., 2011; Galbraith, 1977; Quick and Feldman, 2014). These standardized procedures necessitate less boundary spanning because they can be accomplished independent of SC partners and with few, if any, external resources, information, and support (Gibson and Dibble, 2013; Marrone et al., 2007). Furthermore, engaging in time-consuming boundary spanning in low complexity contexts may result in an excessive exposure to flows of irrelevant information and resources, unnecessarily increasing risk exposure (Prater et al., 2001) and decreasing SC resilience (Jüttner, 2005). Thus, under low disruption complexity, boundary spanning, on the one hand, distracts organizations from more effective and quicker (internal) standardized procedures and, on the other hand, exposes organizations to additional risks and coordination costs. We therefore hypothesize:

Hypothesis 1. The relationship between boundary spanning and supply chain resilience is moderated by disruption complexity. This relationship is positive when disruption complexity is higher, but negative when disruption complexity is lower.

2.4.2 Disruption Uncertainty

and problem solving (Ancona and Caldwell, 1992; Marrone, 2010). Such coordination assists organizations in coping with high levels of disruption uncertainty by orchestrating countermeasures that are well-targeted and broadly supported (Bode and Wagner, 2015; Jüttner and Maklan, 2011; Scholten and Schilder, 2015). Moreover, when faced with uncertain disruptions, boundary spanning flows let an organization sustain an accurate and sufficiently broad view over its SC. This enables organizations to easily identify and resolve emerging uncertainties and interdependencies in (disrupted) shared processes (De Geest et al., 2016; Gibson and Dibble, 2013).

When facing certain disruptions, in contrast, organizations have less need for access to external information and expertise (Duncan, 1972; Leifer and Huber, 1977). To resolve such disruptions, organizations may instead rely on internal information and expertise (Cantor et al., 2014; Flynn et al., 2010). That is, low uncertainty disruptions allow organizations to follow predesigned backup procedures that require minimal coordination and monitoring of SC partners (Flynn et al., 2016; Wong et al., 2011). In fact, engaging in boundary spanning with other organizations may even misdirect already limited resources that could have been invested more efficiently for other operations or more effective (internal) resilience practices such as creating redundancies in raw materials and production capacity (Bode et al., 2011; Mishra et al., 2016; Nooraie and Parast, 2016). Also, establishing and maintaining boundary spanning ties while facing certain disruptions may result in unnecessary information- and role-overload, negatively affecting viability and SC resilience (Aldrich and Herker, 1977; Marrone et al., 2007; Prater et al., 2001). Because of these drawbacks – and only limited benefits – of boundary spanning under low disruption uncertainty, we hypothesize:

Hypothesis 2. The relationship between boundary spanning and supply chain resilience is moderated by disruption uncertainty. This relationship is positive when disruption uncertainty is higher, but negative when disruption uncertainty is lower.

In conclusion, in our conceptual framework, which is portrayed in Figure 1, we set out to integrate the research domains of boundary spanning and SC resilience. Informed by a contingency paradigm, we propose that disruption complexity and disruption uncertainty moderate the relationship between boundary spanning and SC resilience. In doing so, we explore in which contexts boundary spanning may be beneficial or harmful for the ability of a SC to restore performance levels after encountering a disruption.

3. METHODOLOGY

Because we integrate interdisciplinary micro-level (team boundary spanning) and macro-level (SC resilience) research, we adopt a mixed method research design to test our hypotheses. That is, our integration of two research domains that have previously developed largely independently of each other – and are mainly conducted at different levels of analysis – warrants supplementing “quantitative tests with qualitative data [to] enable a fuller explanation of statistical relationships between variables” (Edmondson and McManus, 2007, p. 1166). In other words, we require observations from qualitative data to confirm that our statistical results regarding SC resilience are indeed attributable to boundary spanning arguments. Our purpose of complementing quantitative analyses with qualitative observations is, as such, not to engage in explorative research. Rather, the purpose of our supplementary interviews is to enable a better understanding of the obtained statistical relationships (Greene et al., 1989; Yauch and Steudel, 2003). Furthermore, these interviews contribute to our results’ validity through data triangulation (Edmondson and McManus, 2007; Yauch and Steudel, 2003). Because of these benefits that mixed method research designs offer, they are preferred for studying inherently complex SC phenomena (Boyer and Swink, 2008; Golicic and Davis, 2012), such as SC resilience (Kamalahmadi and Parast, 2016).

In addition to providing a fuller understanding, a mixed method research design is particularly appropriate to address the contemporary developments in the expanding resilience literature. That is, on the one hand, resilience scholars identify the “clear need to use these observations [from the abundance of valuable case studies] to build more general theories that can be quantitatively tested and used to equip decision makers with better models to base crisis preparation and responses upon” (Van der Vegt et al., 2015, p. 974; see also Hohenstein et al., 2015; Kamalahmadi and Parast, 2016). These – and other – scholars, on the other hand, also suggest that “the literature is not rich enough, and more research is needed to examine and validate the theoretical foundations of supply chain resilience” (Kamalahmadi and Parast, 2016, p. 130). Our mixed method research design, combining both quantitative and qualitative approaches, is particularly useful in concurrently addressing these two developments in the resilience literature.

3.2 Case Description

basis. In order to minimize the impact of these and other disruptions, the various organizations involved in operating the Dutch railway system need to continuously interact (i.e., engage in boundary spanning) to share information and coordinate efforts, for which they use a collective information exchange platform.

Preliminary observations and interviews at the railway organizations confirm the centrality of boundary spanning in dealing with disruptions to the Dutch railway system. One of the reasons for this centrality of boundary spanning is the time pressure: “And that is extreme on the railway network; the longer you leave a calamity unaddressed, the larger becomes the chaos and the larger becomes the chance that you will never even be able to present a plan.” (interviewee G). An associated reason is the inter-station interdependency (i.e., the risk that the effects of a disruption near one train station or rail trajectory spill over to affect other train stations or rail trajectories) that creates an inherently complex SC context (Craighead et al., 2007). Interviewees explained that, in addition to a shared information system, organizations that frequently must work together therefore maintain a ‘base level’ of boundary spanning. The level of boundary spanning is subsequently adapted depending on the characteristics of the disruption at hand. These preliminary observations and interviews, hence, further stress the Dutch railway system as an appropriate context in which to investigate when collaboration across organizational boundaries may help or hurt SC resilience.

3.3 Data Collection

We obtained quantitative data on all disruptions to the Dutch railway system over a one-month period in the winter season, which typically is among the most hectic and busiest, making it particularly suitable for our study. The obtained dataset is an exact transcript of the core information system that railway organizations use to manage disruptions. In case of a disruption, railway organizations use this system for sharing information and coordinating activities. In particular, organizations make entries into the system to communicate about their current actions, solicit for external support, and orchestrate collective countermeasures. Example entries include: “Request for contact with [general management of traffic control] because of uncertainty about procedures.” and “Crossing has been fully recovered. A printed circuit still needs to be replaced, but this does not affect functioning.” The dataset further contains information on the sort of disruption, the affected trajectory, and the starting and ending time of a disruption. The interviewees confirmed the prominence of the information system in managing disruptions, endorsing it as an accurate reflection of boundary spanning efforts and resilience outcomes.

appropriate diversity in tenure and functional perspectives (e.g., asset manager, functional director, IT support staff) at different hierarchical levels across the railway organizations. Hence, the interviews yielded diverse and supplementary perspectives on the collaboration among railway organizations in responding to disruptions (Voss et al., 2002).

The interview protocol followed a critical incident technique (CIT) approach (Butterfield et al., 2005; Flanagan, 1954). That is, interviewees were asked to respectively recall a disruption in which inter-organizational collaboration was effective and one in which it was not. For each of these ‘critical incidents’, interviewees were asked to reflect on, among other aspects, motivations to collaborate, outcomes of the collaboration, and barriers to the collaboration. The interviews were conducted in the months following the covered period in the obtained dataset, allowing interviewees to reflect upon disruptions that were included in the dataset. The recalled disruptions varied from bomb alerts and a fire in one of the decentralized control centers to equipment defects and adverse weather conditions. Overall, the structure and richness of the collected quantitative and qualitative data allow for a thorough testing of our hypotheses.

3.4 Measures

The measures that we adopted in our quantitative analyses are as follows. Following Britt (1988) and other outcome oriented conceptualizations of SC resilience (cf. Hohenstein et al., 2015), we measured SC resilience as the time until the railway system had recovered to its regular, pre-disruption train schedule and situation, the point at which the disruption is closed in the information system (i.e., the duration of a disruption). This operationalization allowed us to examine in what timeframe the railway organizations were successful in restoring performance levels after encountering a disruption, providing an objective measure of SC resilience. In line with prior research (e.g., Tsai, 2002; De Vries et al., 2014), we operationalized boundary spanning as the total number of unique teams interacting and collaborating with each other in solving a disruption, as registered within the core information system for disruption management. This operationalization closely follows from the distinct purpose of boundary spanning to integrate external knowledge, resources, and support – residing within (teams of) other organizations – to achieve collective, inter-organizational outcomes (Ancona and Caldwell, 1992; Marrone, 2010).

unpredictability of subsequent developments. These characteristics, therefore, provide an accurate reflection of disruption uncertainty.

To rule out potential alternative explanations for our findings, we included several control variables. These control variables were selected based on a priori reasoning, followed by an inspection of their corresponding correlations with the dependent variable of interest (Becker, 2005). First, we controlled for the severity of a disruption as approximated by the number of entries into the information system per disruption. Because the severity of a disruption significantly affects its duration and, thus, the time until recovery (Craighead et al., 2007), not controlling for this would substantially confound the moderating effects of disruption complexity and uncertainty. Second, we controlled for the organizations included in the response to a disruption, as the characteristics of these organizations may inherently affect the time until recovery. Indeed, four of the eleven organizations were significantly correlated with the time until recovery. Hence, we included dummy variables for each of these four organizations in our analyses.

Our quantitative data incorporates complete information on 483 disruptions spread across 86 terminal stations. Table 1 presents the descriptive statistics and bivariate correlations for the variables of interest. The high correlations between boundary spanning and disruption severity and Organization VII are not surprising given that the information system and associated organizations are specifically aimed at responding to and resolving disruptions. These correlations illustrate that, in line with Craighead et al. (2007), severe disruptions have a longer duration, necessitating more boundary spanning between organizations with specific expertise, such as Organization VII. Despite these high correlations, no problems with multicollinearity were found given that VIF-values ranged from 1.069 to 4.573 – well below the commonly accepted threshold value of 10.00, as well as below the more conservative threshold value of 5.00 – and considering the large sample size (Hair et al., 2013). Our complexity and uncertainty measures are completely uncorrelated, suggesting that they measure distinct aspects of disruptions. Also, our complexity and uncertainty measures only partially correlate with disruption severity, providing further support for controlling for the latter variable’s confounding effect on our relationships of interest. These correlations and VIF-values do not take into account the nonindependence of the data, however, and should be interpreted with caution.

4. FINDINGS

Table 1 – Descriptive statistics and correlations.

Mean s.d. 1 2 3 4 5 6 7 8 1 Time until recoverya _{0.165 1.252}

Table 2 – Results of multilevel regression analyses.

Time until recovery

Variables Model 0 Model 1 Model 2 Model 3 Model 4 Model 5

Intercept 0.136 (0.066)* 0.164 (0.084) 0.129 (0.093) 0.223 (0.081)** 0.161 (0.093) 0.250 (0.082)** Boundary spanning -0.093 (0.107) -0.071 (0.093) -0.099 (0.108) -0.069 (0.093) Disruption complexity 1.376 (0.109)** 1.343 (0.110)** Boundary spanning x Disruption complexity (H1) -0.444 (0.098)** -0.510 (0.102)** Disruption uncertainty -0.228 (0.209) -0.069 (0.183) Boundary spanning x Disruption uncertainty (H2) -0.303 (0.153)* -0.282 (0.137)* Control variables Disruption severity 0.680 (0.062)** 0.731 (0.086)** 0.610 (0.076)** 0.803 (0.090)** 0.668 (0.082)** Organization IV 0.256 (0.244) 0.269 (0.243) 0.274 (0.208) 0.222 (0.242) 0.237 (0.207) Organization VI -0.179 (0.227) -0.115 (0.238) -0.119 (0.206) -0.087 (0.237) -0.098 (0.205) Organization VII -0.184 (0.124) -0.118 (0.145) -0.100 (0.126) -0.122 (0.144) -0.114 (0.126) Organization IX 0.413 (0.177)* 0.458 (0.185) 0.404 (0.160)* 0.442 (0.184)* 0.399 (0.159)* Variance components

Within-terminal station variance 1.495 1.060 1.058 0.793 1.043 0.785 Intercept variance 0.066 0.071 0.070 0.046 0.067 0.044

Additional information

ICC 0.051

-2 Log Likelihood (FIML) 1582.463 1423.62** 1422.872** 1281.119** 1415.326** 1275.787** Estimated parameters 3 8 9 11 11 13

Pseudo R2 0.28 0.28 0.46 0.29 0.47

Note: FIML = Full information maximum likelihood estimation. N = 483 SC disruptions. Unstandardized regression coefficients are shown; standard errors

the terminal station). Because of minor (positive) skewedness in our SC resilience measure (i.e., time until recovery), we decided to log-transform this outcome variable.

Table 2 summarizes the results of our multilevel regression analyses. In our null model, we only estimated a random intercept. Model 1 subsequently includes all our control variables. In Model 2, we additionally regressed boundary spanning on time until recovery. In Models 3 and 4, we tested our hypotheses by respectively adding disruption complexity and uncertainty and their interaction effects with boundary spanning. Finally, we included both moderating relationships in Model 5. Table 2 additionally reports variance components and model fit (i.e., -2 Log Likelihood and pseudo R2). Each of the estimated models predicts significantly better than our null model, with our complete model explaining 47% of the variance in time until recovery. The random slopes were insignificant in each of the estimated models, indicating that our multilevel modeling estimates for time until recovery did not significantly differ across terminal stations. The models including random slopes did, therefore, not provide a significant increase in explanatory power and are not reported.

With regard to our control variables, disruption severity and Organization IX are significant in respectively five and four of the estimated models. The respective regression coefficients show that the time until recovery is positively associated with disruption severity and the involvement of Organization IX in the disruption response. Moreover, as we expected, we find no direct relationship between boundary spanning and SC resilience in any of our models (e.g., B = -.093, SE = .107, p = n.s.; Model 2), suggesting that, indeed, disruption characteristics may play an important moderating role in the relationship between boundary spanning and SC resilience.

Hypothesis 1 posits that the relationship between boundary spanning and SC resilience is moderated by disruption complexity. That is, the relationship is positive under high complexity and negative under low complexity. Our results support this hypothesis (B = -.444, SE = .098, p < .01; Model 3). The significant interaction effect is plotted in Figure 2 using the untransformed values of time until recovery. The untransformed values are plotted for a low degree of boundary spanning and disruption complexity (i.e., one standard deviation below the mean) and a high degree of boundary spanning and disruption complexity (i.e., one standard deviation above the mean). Indeed, under low disruption complexity there is a negative relationship between boundary spanning and SC resilience. In such contexts, engaging in more boundary spanning is associated with an increase in the time until recovery (slope parameters: B = .373, SE = .134, p < .01), thereby hurting resilience. Under high disruption complexity, on the contrary, there is a positive relationship between boundary spanning and SC resilience. In such contexts, engaging in more boundary spanning is associated with a decrease in the time until recovery (slope parameters: B = -.515, SE = .369, p < .01), thereby helping resilience.

Figure 2 – Moderation effect of disruption complexity (H1).

Because disruption uncertainty is a dummy variable, low uncertainty indicates a value of 0 and high uncertainty a value of 1. As hypothesized, under high disruption uncertainty there is a positive relationship between boundary spanning and SC resilience. In such contexts, engaging in more boundary spanning is associated with a decrease in the time until recovery (slope parameters: B = -.231, SE = .103, p < .05), thereby helping resilience. Under low disruption uncertainty, however, we find no support for a moderation effect (slope parameters: B = -.137, SE = .096, p = n.s.). In such contexts, engaging in boundary spanning does not significantly affect the duration of a disruption. Hence, we find only partial support for Hypothesis 2.

Figure 3 – Moderation effect of disruption uncertainty (H2).

marginal increase in pseudo R2 may that our dataset contained relatively few – but sufficient – disruptions with high uncertainty, which is understandable (i.e., equipment malfunctioning occurs more frequently than bomb alerts). Pseudo R2 values, however, do not take into account the source of explained variation and should thus be interpreted with caution (Aguinis et al., 2013; Snijders and Bosker, 1999). The low pseudo R2 for disruption uncertainty notwithstanding, the significance of both interaction terms is robust when combined into our full model (Model 5; Disruption complexity: B = -.510, SE = .102, p < .01; Disruption uncertainty: B = -.282, SE = .137, p < .05).

4.1.1 Supplementary Analyses

In our hypotheses development and multilevel regression analyses, we assumed that the moderation effects of disruption complexity and disruption uncertainty are additive rather than multiplicative or interactive. Specifically, we assumed that boundary spanning during a disruption characterized by both high (low) complexity and high (low) uncertainty does not increase (decrease) SC resilience more than the combined, independent effects of boundary spanning under either high (low) complexity or high (low) uncertainty. To check this assumption, we tested for a three-way interaction between boundary spanning, disruption complexity, and disruption uncertainty. Reflecting the insignificant correlation between disruption complexity and disruption uncertainty (see also Table 1), we did not find support for this way interaction (B = -.855, SE = 1.136, p = n.s.). The insignificance of a three-way interaction between boundary spanning, disruption complexity, and disruption uncertainty justifies testing both moderation effects independently in Models 3 and 4 prior to including them parallel in Model 5.

4.2 Qualitative Analyses

Overall, our quantitative analyses provide initial support for the moderating role of disruption characteristics in the relationship between boundary spanning and SC resilience. To ensure that these statistical results constitute “a valid analysis of the phenomenon rather than artifacts of measurement” (Edmondson and McManus, 2007, p. 1166), we rely on supplementary interviews to uncover whether the underlying mechanisms are indeed attributable to boundary spanning processes. In addition to being generally preferred for inherently complex SC phenomena (Boyer and Swink, 2008; Golicic and Davis, 2012), this combination of quantitative and qualitative analyses is required because we integrate interdisciplinary micro-level (team boundary spanning) and macro-micro-level (SC resilience) research (Edmondson and McManus, 2007). The purpose of our qualitative analyses is, therefore, not to engage in explorative research, but to offer a fuller understanding of the obtained statistical relationships (Greene et al., 1989; Yauch and Steudel, 2003).

codes were grouped and classified according to whether they identified benefits or detriments of boundary spanning for SC resilience. The final step involved juxtaposing these observed benefits and detriments with our quantitative analyses in order to uncover the mechanisms underlying our statistical results. Table 3 provides an illustration of the described procedure. To maximize the validity and reliability of our qualitative analyses, we closely followed the measures as defined for our quantitative analyses, as well as that we only translated interview excerpts to English when including them in the present manuscript.

Table 3 – Excerpt from coding scheme.

Data reduction (first-order codes) Interpretive categories (second-order codes) Helpful/harmful (third-order theme) “You have to see it from the perspective of the

customer. Is the customer waiting for the

information at the moment that you think he should be waiting on it?” (Interviewee A)

Information overload Harmful

“So, ambiguity and incomplete information about the situation outside, problems in coordination between [three affected organizations], with, as consequence, complete uncertainty in passenger-information. And we could not get that under control.” (Interviewee B)

Uncertainty necessitating coordination

Helpful

“Well, at that moment, it just stops for us because we just cannot continue with our work. We need more information.” (Interviewee D)

Dependence on outside information

Helpful

“But, like yesterday, I had to fulfil a director’s role for 30% of my time. My other, regular activities suffer from that because I could only do 70% of those activities.” (Interviewee D)

Distracts from regular activities

Harmful

“What breaks you up is that you have an awfully high amount of work from communicating a decision.” (Interviewee G)

Time-consuming Harmful

“But you can make the right decision, because of which you can substantiate it and also convince everyone outside that the decision was a deliberate one.” (Interviewee R)

Better solutions Helpful

and allow organizations to anticipate the needs of others already in an early stage. For more uncertain and complex disruptions, this becomes more difficult, making organizations more dependent upon the flow of outside information. This observation provides an initial explanation of why the benefits of boundary spanning are more pronounced for disruptions with high levels of complexity or uncertainty as opposed to disruptions with low levels of complexity or uncertainty. In the following sections, we explore more fully the benefits and detriments of boundary spanning for SC resilience as perceived by the organizations involved in operating the Dutch railway system.

4.2.1 The Benefits of Boundary Spanning

The railway organizations acknowledged to increasingly rely on boundary spanning as disruptions become more complex and uncertain because of various reasons. The most important of which is the dependence on outside information, as interviewee D formulated: “Well, at that moment [after making an entry into the central information system], it just stops for us because we just cannot continue with our work. We need more information.” This information serves to create a complete picture of the situation and, most importantly, to determine available approaches: “You need each other. Because I can come up with a good plan […], but I need to rely on the carrier that my plan is executable as well.” (interviewee P). Vice versa, limited access to outside information was frequently stressed to increase the impact and duration of a disruption.

In addition, the railway organizations engaged in boundary spanning to gain access to the outside knowledge and expertise required for responding to and resolving complex and uncertain disruptions: “If all of that does not work, then it comes to us, at back office, and we will start talking with the carriers. We look whether there is a car with technical support people available. They have more knowledge of the material.” (interviewee D). Besides gaining access to outside information and expertise, interviewees indicated that the railway organizations rely on boundary spanning to gain access to external resources, for evaluation purposes, and because of the inability of individual teams or organizations to deal with multiple, co-occurring disruptions.

Overall, these benefits of boundary spanning as perceived by the railway organizations provide a better understanding of the positive relationship between boundary spanning and SC resilience under high disruption complexity and uncertainty. The mechanisms underlying this relationship closely mirror the theoretical boundary spanning arguments that we relied on in the development of our hypotheses. Specifically, the railway organizations indicated to rely on boundary spanning for the timely access to outside information, expertise, and resources, all of which were imperative for determining available approaches to (emerging) disruptions. Boundary spanning was further positioned to shorten the time until recovery by assisting teams or organizations that lacked the capacity to deal with multiple, co-occurring (i.e., complex) disruptions. In other words, the organizations involved in operating the Dutch railway system purposefully engaged in boundary spanning to better and more quickly resolve complex and uncertain disruptions.

4.2.2 The Detriments of Boundary Spanning

Reflected in the negative relationship between boundary spanning and SC resilience for disruptions low in complexity, our supplementary interviews highlighted several detriments of boundary spanning for the resilience of the Dutch railway system. Most of these perceived detriments are not specific to disruptions low in complexity or uncertainty. Rather, under high disruption complexity and uncertainty these detriments are offset against more substantial benefits, resulting in an overall shorter time until recovery. Under low disruption complexity and uncertainty, these detriments of boundary spanning may, however, not be offset and result in an overall longer time until recovery. Based on the supplementary interviews, we could not, however, find a definite explanation for our insignificant results for the negative effects of boundary spanning under low disruption uncertainty. Notwithstanding the need for future research, we believe the following detriments of boundary spanning to result in a longer time until recovery under both low disruption complexity and low disruption uncertainty.

may result in further delays and frustrations: “But, anyway, they disrupt the process you are in. That man must only call after fifteen or twenty minutes.” (interviewee N).

In addition, the interviews revealed that, in some instances, boundary spanning resulted in information overload due to an excessive exposure to irrelevant information. Accordingly, various interviewees stressed the importance of anticipating a recipient’s information needs: “You have to see it from the perspective of the customer. Is the customer waiting for the information at the moment that you think he or she should be waiting on it?” (interviewee A). Because of similar reasons, excessive boundary spanning for the level of disruption complexity or uncertainty was described as ineffective and inefficient: “We have to look very carefully to effectiveness and efficiency. […] In which areas can you collaborate? I think that sometimes we want to know too much, without the information being of value to us.” (interviewee E).

A third detriment of boundary spanning observed by the railway organizations is that not all teams and organizations have functions purposefully designed for boundary spanning. The absence of such functions causes team or organizational members to take on boundary spanning responsibilities in addition to their own activities: “Well, at [organization], they have not created a director function. They have added that to an existing function. […] But you must acknowledge that. That is wrong here. I must apparently design an official role, because you must minimally deliver this and this information to your partner.” (interviewee C). Understandably, this consolidation of boundary spanning responsibilities distracts from a member’s regular working activities: “But, like yesterday, I had to fulfil a director’s role for 30% of my time. My other, regular activities suffer from that because I could only do 70% of those activities.” (interviewee D). When the added value of these boundary spanning responsibilities is not evident for the respective employee, he or she is unlikely to make the necessary investments at the expense of his or her own work. Also, several interviewees stressed the need for individuals buffering the focal team or organization from outside pressures and interferences to allow for fewer disturbances.

The final and potentially most important detriment of boundary spanning evident from our supplementary interviews is that not all boundary spanning relationships are effective. As a result, the time, effort, and resources invested in these relationships may simply be lost. One of the principal reasons for this is different interests among the affected organizations, each of which may have a different perspective on the cause, impact, and consequences of a disruption. Overcoming self-interests and aligning perspectives is a difficult and time-consuming task: “He is going to save his own skin and he will not include you in that process. […] Then it falls of his shoulders, then he has finished his task. But this comes back to him, and to me as well, because afterward you have a lot work if it is badly executable or if it simply was not a right decision. Then you must reverse everything.” (interviewee G). This may ultimately result in a substantial amount of time and effort being wasted on monitoring and disputing decisions and processes rather than on increasing SC resilience: “But then you have to keep checking whether your request is forwarded in the circle, you have to finish the circle. You have to look whether your request is being acknowledged and returned” (interviewee O).

agreements might be necessary: “Sometimes you have to say: ‘I understand that talk about protocols, but now no nonsense. We are here working on a product, understand? We are not here to keep ourselves busy. We are here to provide our customer with the best possible information and that is our priority.’” (interviewee A). Our interviewees recalled several disruptions for which organizations were unwilling to deviate from standard protocols or original agreements. Such unwillingness hampers the effectiveness of boundary spanning in increasing resilience. Examples of other reasons why boundary spanning may be ineffective for increasing resilience are no power or willingness in the boundary spanning relationship to make high impact decisions, unclear internal organizational roles, and lack of in-person communication.

Potentially because of these experienced detriments of boundary spanning, multiple interviewees emphasized that solutions to (emerging) disruptions are initially sought within the team or organization. When the knowledge or expertise necessary to respond to the disruption is not internally available, then these are sought externally through boundary spanning. Yet, in this escalation process, the impact or expected duration of a disruption plays an important role: “Because in the alarming process, I have a margin of half an hour. When I am convinced that it has ended within half an hour, then I do not have to alert.” (interviewee N). In other words, for at least several of the railway organizations, not engaging in boundary spanning after a disruption is a deliberate decision when the disruption is characterized by low impact, complexity, or uncertainty.

In conclusion, these detriments of boundary spanning provide insights into the mechanisms underlying the negative relationship between boundary spanning and SC resilience under low disruption complexity and uncertainty. As for the observed benefits of boundary spanning, these detriments closely mirror the theoretical arguments that we relied on in the development of our hypotheses. In particular, the organizations involved in operating the Dutch railway system perceive boundary spanning as a complex and time- and effort-consuming process that may result in an excessive exposure to irrelevant information. The supplementary interviews further reveal that not all boundary spanning linkages are effective, as well as that not all teams and organizations are equally equipped to take on boundary spanning responsibilities. These detriments of boundary spanning are surpassed by substantial benefits for disruptions high in complexity or uncertainty, explaining the subsequent reduction in time until recovery. For disruptions low in complexity or uncertainty, however, the detriments of boundary spanning overshadow corresponding benefits, resulting in a longer time until recovery. In responding to such disruptions, organizations may instead wish to focus on other, more effective (internal) resilience practices.

5. CONCLUSION AND DISCUSSION

“dark side” of boundary spanning, we take a contingency perspective and set out to explore in which contexts boundary spanning helps or hurts SC resilience. Specifically, we examine the moderating effects of disruption characteristics – complexity and uncertainty – on the relationship between boundary spanning and SC resilience. We adopt a mixed method research design that includes a large sample of disruptions to the Dutch railway system, supplemented by nineteen interviews among affected railway organizations.

We find that under high disruption complexity, boundary spanning is associated with a shorter time until recovery. In these adverse contexts, boundary spanning is thus an important and effective means to increasing SC resilience by, among other aspects, providing organizations with timely access to outside expertise and resources. For less complex disruptions, however, we find that boundary spanning is associated with a longer time until recovery. In these contexts, boundary spanning is harmful for SC resilience as it may misdirect already limited resources and divert attention away from alternative, potentially more effective resilience practices. Specifically, our interviews revealed that boundary spanning was perceived to be a complex and time- and effort-consuming process that yielded only limited benefits under low disruption complexity.

Similar to boundary spanning under high disruption complexity, we find that boundary spanning under high disruption uncertainty is also associated with a shorter time until recovery. For such disruptions, boundary spanning facilitates the sharing of complete and accurate information among SC partners. This increases SC resilience by providing a detailed picture of the situation and feasible countermeasures. We find no significant moderation effect of low disruption uncertainty, indicating that boundary spanning may be an ineffective, yet resource-intensive approach to responding to SC disruptions characterized by low uncertainty. Based on our supplementary interviews, however, we could not identify a definitive explanation for the insignificance of this moderation effect and recognize the need for further research.

In combination with the absence of a direct effect between boundary spanning and SC resilience, these findings lend support to recent micro-level team research emphasizing the dualistic implications of boundary spanning. In addition, our findings are indicative of the need to critically address the overly positive view of resilience practices (cf. Van der Vegt et al., 2015). That is, as we have illustrated, resilience practices might not be uniformly effective – or even be counterproductive – across diverse contexts. Scholars and practitioners alike should therefore more deliberately consider the costs and inefficiencies, and the role of context therein, associated with practices aimed at strengthening resilience.

5.1 Contributions

image” of resilience (Van der Vegt et al., 2015, p. 975). We take a first step toward more critically reflecting upon the costs and inefficiencies of resilience strategies, benefiting both research and practice. In doing so, we provide further support for the benefits of adopting outward-facing resilience strategies in adverse contexts (e.g., Kauppi et al., 2016; Scholten and Schilder, 2015). However, we also show that organizations adopting boundary spanning – one such outward-facing resilience strategy – in more favorable contexts may inadvertently end up hurting resilience. SC managers should, therefore, carefully anticipate disruption characteristics in selecting and implementing resilience practices.

Second, we have initiated research addressing the “notable and surprising” absence of literature examining the contingency aspects of boundary spanning (Marrone, 2010: 933). Specifically, although early research has identified several conditions under which boundary spanning is particularly beneficial or, rather, counterproductive (e.g., Ancona and Caldwell, 1992), limited (empirical) research has since been conducted toward examining the contingencies of boundary spanning (Marrone, 2010). Our findings illustrate that, in addition to task complexity and uncertainty (e.g., Choi, 2002; Faraj and Yan, 2009; Gibson and Dibble, 2013), disruption complexity and uncertainty may be decisive for whether boundary spanning results in its desired performance outcomes – in our case, increased SC resilience. We encourage future research to further explore how disruption complexity and uncertainty moderate the relationship between boundary spanning and desired performance outcomes other than SC resilience (e.g., profitability, innovativeness).

A third contribution is our integration of two research domains – those of boundary spanning and SC resilience – that have previously developed largely independently of each other. Our combination of quantitative data and supplementary interviews illustrates that such an integration of micro-level (team boundary spanning) and macro-level (SC resilience) research enables important theoretical advances in Operations Management research. The relative absence of boundary spanning literature in the context of SC resilience is surprising because collaboration has been broadly recognized as one of the primary antecedents of SC resilience (e.g., Hohenstein et al., 2015; Tukamuhabwa et al., 2015). Because of the broader relevance of boundary spanning arguments for SC management, we believe that related research domains involving, for instance, SC integration and buyer-supplier relationships may likewise benefit from a better integration of boundary spanning literature. For example, social capital is seen as an important antecedent of boundary spanning (e.g., Marrone, 2010; Tsai, 2000). Boundary spanning may, as such, be one of the mechanisms sought by Villena et al. (2011) through which social capital affects the performance of buyer-supplier relationships.

in general SC Management literature (Golicic and Davis, 2012) and SC resilience literature in particular (Kamalahmadi and Parast, 2016). We hope that our study contributes to encouraging future (SC) research in adopting this instrumental research design.

5.2 Limitations and Directions for Future Research

The present study has several important strengths, among which a mixed method research design including a large, detailed sample of disruptions to the Dutch railway system. Having access to such a large sample is unique because few organizations maintain this type of data, especially at this level of detail. Nevertheless, our research has some limitations that need to be addressed. First, we could not infer causality from our sample. Albeit our theoretical framework is in line with previous research examining the moderating role of context in the relationship between boundary spanning and its desired outcomes (e.g., Choi, 2002; De Geest et al., 2016; Gibson and Dibble, 2013), future experimental or longitudinal research is needed before causal interference is justified.

A second limitation is that we were unable to measure and quantitatively test the micro-level mechanisms that we identified in our supplementary interviews. That is, because of the structure of our quantitative data, we were only able to assess the effectiveness of boundary spanning – i.e., whether it increased or decreased the time until recovery. We could not distinguish, for example, whether a decrease in time until recovery was caused by improved coordination or access to external resources, or both. Neither could we determine which of the identified mechanisms mostly strongly affected SC resilience. Although our inability to measure these micro-level mechanisms is less problematic because of our mixed method research design, we encourage future research to assess and quantitatively test these identified mechanisms underlying the relationship between boundary spanning and SC resilience.

In addition, further research testing our conceptual framework in different research settings is necessary to increase the generalizability of our results. In particular, the organizations involved in managing the disruptions in our study were heavily interdependent and, in some instances, collocated in a central control center. This may have been an important determinant of these organizations’ boundary spanning efforts, irrespective of the disruption context. That is, in general, the degree of boundary spanning with an exchange partner increases as the dependency on that respective partner increases (e.g., Bode et al., 2011; Marrone, 2010). Examining the role of external interdependence in our conceptual framework advances our understanding of the role of boundary spanning in enhancing SC resilience. Also, future research should seek to replicate the present study’s findings in a SC context that is not specifically designed for responding to and resolving SC disruptions.

decentralized boundary spanning for increasing SC resilience (e.g., Marrone, 2010; De Vries et al., 2017). Through, for example, the use of network measures (cf. De Vries et al., 2014), future research may uncover whether a focal team or organization directing the boundary spanning flows is more beneficial for increasing SC resilience than multiple teams or organizations jointly directing these flows. Similarly, future research can reflect on whether and how horizontal and vertical flows of boundary spanning differently affect SC resilience (e.g., Van der Vegt et al., 2015).

In addition, our conceptual framework may be further extended by examining boundary spanning – an outward-facing resilience practice – concurrently alongside inward-facing resilience practices such as creating redundancies in resources and capacities. Specifically, whereas we observed contexts in which boundary spanning does not increase – or even hurt – SC resilience, future research should assess whether instead relying on inward-facing resilience practices in these contexts does increase SC resilience. Such research may advance our understanding of the environmental or situational characteristics under which a resilience practice is preferred over another.

BIBLIOGRAPHY

Aguinis, H., Gottfredson, R.K. and Culpepper, S.A. (2013), Best-Practice Recommendations for Estimating

Cross-Level Interaction Effects Using Multilevel Modeling, Journal of Management, Vol. 39 No. 6,

pp. 1490-1528.

Alcantara, P. and Riglietti, G. (2016), Supply Chain Resilience Report 2016, Business Continuity Institute, Zurich.

Aldrich, H. and Herker, D. (1977), “Boundary spanning roles and organization structure”, Academy of

Management Review, Vol. 2 No. 2, pp. 217–230.

Ancona, D.G. (1990), “Outward Bound: Strategies for team survival in an organization”, Academy of

Management Journal, Vol. 33 No. 2, pp. 334–365.

Ancona, D.G. and Caldwell, D.F. (1992), “Bridging the boundary: External activity and performance in organizational teams”, Administrative Science Quarterly, Vol. 37 No. 4, pp. 634–665.

Becker, T.E. (2005), “Potential Problems in the Statistical Control of Variables in Organizational Research: A Qualitative Analysis With Recommendations”, Organizational Research Methods, Vol. 8 No. 3, pp. 274–289.

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, Elsevier B.V., Vol. 36, pp. 215–228. Bode, C., Wagner, S.M., Petersen, K.J. and Ellram, L.M. (2011), “Understanding responses to supply chain

disruptions: Insights from information processing and resource dependence perspectives”, Academy of

Management Journal, Vol. 54 No. 4, pp. 833–856.

Boer, H., Boer, H.E.E., Demeter, K. and Szasz, L. (2017), “Towards a contingency theory of operations risk management”, EurOMA 2017 Conference Paper, Edinburgh, Scotland.

Boyer, K.K. and Swink, M.L. (2008), “Empirical elephants - Why multiple methods are essential to quality research in operations and supply chain management”, Journal of Operations Management, Vol. 26 No. 3, pp. 337–348.

Brandon-Jones, E., Squire, B., Autry, C.W. and Petersen, K.J. (2014), “A Contingent Resource-Based Perspective of Supply Chain Resilience and Robustness”, Journal of Supply Chain Management, Vol. 50 No. 3, pp. 55–73.

Britt, D.W. (1988), “Analyzing the shape of organizational adaptability in response to environmental jolts”,

Clinical Sociology Review, Vol. 6 No. 1, pp. 59–75.

Burns, T. and Stalker, G.M. (1961), The Management of Innovation, Tavistock, London.