Big Data Analytics Capabilities and Firm Performance: The
moderating effect of innovative organizational culture and
decentralization of decision-making.
Master thesis
Author: Moritz Steinert / Student N. 11802456 MSc. Business Administration: Strategy Track
Institution: University of Amsterdam, Faculty Economics and Business
Supervisor: Andreas Alexiou, University of Amsterdam
Statement of originality
This document is written by Moritz Steinert who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Table of contents
ABSTRACT ... 4
1. INTRODUCTION ... 5
2. LITERATURE REVIEW ... 8
2.1RESOURCES-BASED THEORY (RBT) ...8
2.2BIG DATA ... 10
2.3BIG DATA ANALYTICS CAPABILITIES (BDAC) ... 12
2.3.1 Tangibles ... 13
2.3.2 Human resources ... 14
2.3.3 Intangibles ... 15
2.4ORGANIZATIONAL CULTURE ... 16
2.4.1 Innovative organizational culture... 18
2.5ORGANIZATIONAL STRUCTURE ... 19
3. THEORETICAL FRAMEWORK & HYPOTHESES ... 21
3.1BIG DATA ANALYTICS CAPABILITIES AND FIRM PERFORMANCE ... 21
3.2INNOVATIVE ORGANIZATIONAL CULTURE AND FIRM PERFORMANCE ... 22
3.3DECENTRALIZATION OF DECISION-MAKING AND FIRM PERFORMANCE ... 23
4. METHODOLOGY ... 24
4.1DATA COLLECTION ... 24
4.2MEASURES ... 25
4.2.1 Dependent variable ... 26
4.2.2 Independent variable ... 26
4.2.3 Innovative organizational culture... 27
4.2.4 Decentralization of decision-making ... 27
4.2.5 Control variables ... 28
4.3ANALYSIS METHOD ... 29
4.4BDAC MODEL SPECIFICATION ... 30
5. ANALYSIS & RESULTS ... 30
5.1EVALUATION OF THE MEASUREMENT MODEL... 31
5.2EVALUATION OF THE STRUCTURAL MODEL ... 35
6. DISCUSSION ... 38
6.1MAJOR FINDINGS ... 38
6.2MODERATING EFFECT OF INNOVATIVE ORGANIZATIONAL CULTURE ... 39
6.3MODERATING EFFECT OF DECENTRALIZATION OF DECISION-MAKING ... 40
7. CONTRIBUTION ... 41
7.1THEORETICAL IMPLICATIONS ... 41
7.2MANAGERIAL IMPLICATIONS ... 42
8. LIMITATIONS AND FUTURE RESEARCH ... 42
9. CONCLUSION ... 44
10. REFERENCES ... 45
Abstract
Few other technologies are as prominent in our society as big data. Unsurprisingly, it has become a key topic of practitioners and researchers alike. Despite the broad consensus about the benefits of big data analytics for organizations, there remains a lack of research on the factors enabling firms to leverage big data analytics. This study aims to contribute to the emerging field of big data analytics capabilities (BDAC) research by examining the effects of innovative organizational cultures and decentralization of decision-making on BDAC. In order to collect the data necessary for examining the proposed research model, a survey was distributed to big data analytics managers, senior managers, and C-suite individuals who were perceived to have an overview of the relevant organizational factors. The study uses a partial least square path modeling (PLS-SEM) to estimate the model and to test the structural relationships between the variables. As a result, the study yields further support to prior BDAC literature, indicating a strong positive relationship between BDAC and both market performance (MP) and operational performance (OP). Furthermore, the study identifies a negative relationship between decentralization of decision-making and operational performance. No significant results could be found for the relationship between decentralization and MP, as well as innovative organizational culture and MP or OP. These findings advance the emerging field of BDAC research by providing further support for the positive effects of BDAC on FPER. Furthermore, this study connects the fields of BDAC literature to both organizational culture and organizational structure literature. By connecting these fields, the study provides opportunities for further research. Last, the study provides managers with a roadmap for understanding and developing BDAC within their organizations.
1. Introduction
Big Data has become one of the most pervasive concepts in today's economy. Technological advancements in hand-held devises or sensors, the rise of the mobile internet and social networks, and the emergence of technologies such as the Internet of Things (IoT) have created a continuously increasing flow of unstructured data, also known as ‘Big Data’. This plethora of data available to organizations is now considered one of the most important, if not the most important, resource in the 21st century (Garmaki, Boughzala, & Wamba, 2016). What is more, advancements in artificial intelligence, quantum computing, and cloud computing allow firms to organize and analyze big data in order to extract useful insights. There is already broad agreement on the impact big data analytics will have on both economy and society. It promises to revolutionize subjects such as disease prevention, energy management, or infrastructure utilization (Kayyali, Knott, & Kuiken, 2013; Lerner, Viatchaninova, Gee, & Plagge, 2017; Neumann, 2015). In the US alone, big data is expected to add roughly $300 billion to the healthcare market annually (Manyika et al., 2011). What is more, Big Data could help to save an annual $400 billion globally by reducing congestion and increasing efficiency of infrastructure usage (Neumann, 2015). As a result, organizations around the world are investing heavily in big data analytics. Between 2013 and 2014 global investments in BDA increased from $2.1 trillion to roughly $3,8 trillion (Akter, Wamba, Gunasekaran, Dubey, & Childe, 2016). However, financial investments in big data analytics do not guarantee large financial profits. That is, not all organizations can profit to the same extent from their investments as some of their competitors. Developing the necessary organizational capabilities to analyze and leverage big data successfully is now more than ever an imperative for organizations seeking to gain a competitive advantage in the digital economy (Constantiou & Kallinikos, 2015).
To understand what exactly contributes to an organization’s ability to use big data analytics successfully, researchers have increasingly shifted their focus towards the concept of BDAC. However, BDAC research is still at a very early stage. Although some studies have first developed conceptual frameworks, it is important for scholars and practitioners to advance the general understanding about BDAC (Akter et al., 2016; Gupta & George, 2016). Specifically, advancing an understanding about the contingencies between BDAC and other organizational factors is critical for advising managers and organizations on how to implement and leverage big data analytics successfully. A few studies have already explored organizational factors that may influence the implementation of BDAC. (Akter et al., 2016) examined the effect of aligning organizational strategy with BDAC to increase value creation through BDA. Likewise, (Wamba et al., 2017) examined how process-orientated dynamic capabilities mediate the relationship between BDAC and firm performance. Others have examined the general nature of value creation from big data. In this vein, (Wamba, Akter, Edwards, Chopin, & Gnanzou, 2015) have identified five avenues for value-creation from big data, namely improved transparency, increased operational efficiency, improved customization of products/services, data-driven decision-making, and increased business model, products or services innovation. Nonetheless, there remains a substantial amount of empirical research that needs to be done to establish a general understanding of BDAC.
Two aspects that have not yet been examined are the effects of organizational culture and organizational structure on an organization’s ability to leverage BDAC (Gupta & George, 2016; Wamba et al., 2017). Given the tremendous success of today’s data-driven tech companies, many of which are famous for their innovative organizational cultures, it stands to reason that organizational culture has important implications for BDAC. Moreover, findings of prior research on the effects of innovative organizational cultures provide several arguments for a positive impact of culture on the value creation through BDAC. For example, firms with
innovative organizational cultures tend to emphasize innovativeness and strategic renewal (J. B. Barney, 1986; Slater & Narver, 1995). Accordingly, these organizations are likely to analyze available data more thoroughly and are expected to embrace innovative ideas generated from the derived insights. Therefore, such organizations may be able to derive more value from the same data as other, less innovative firms. Similarly, decision-making processes are another interesting point that ought to be empirically examined (Gupta & George, 2016). The majority of research on making has identified positive effects of decentralization of decision-making on organizational performance (Ruekert, Walker, & Roering, 1985). That is, decentralized decision-making enhances knowledge creation and knowledge management within the organization (H. Lee & Choi, 2003). As a result, these organizations tend to have larger knowledge stocks which may positively influence an organization’s ability to integrate insights derived from data. For these reasons, the study aims to answer the question:
‘How do an innovative organizational culture and decentralization of decision-making moderate the relationship between big data analytics capabilities and firm performance?’
This study contributes to the emerging debate on BDAC in several ways. First, the results yield further support for the positive relationship between BDAC and FPER. Furthermore, the study connects the field of BDAC research to both organizational culture and organizational structure literature. Thereby, it advances the understanding about BDAC and its contingencies with other organizational factors, as suggested by (Akter et al., 2016; Wamba et al., 2015). Lastly, the study provides managers with a roadmap for building BDAC.
The thesis is structured in the following way. First, the relevant literature will be reviewed to create a theoretical foundation for this study. Next, based on the findings of the literature review, the hypotheses will be formulated. Following the hypothesis development, the methodology section introduces the sampling strategy, variables, and method of analysis. After the methodology follows the analysis of the measurement model and the structural model.
Next, the study discusses the findings and implications, followed by both limitations and future research. Finally, the findings are summarized in the conclusion section.
2. Literature Review
2.1 Resources-based theory (RBT)
The resource-based theory has emerged as one of the most crucial and robust theories of firm performance and competitive advantage within the field of management. The RBT is an efficiency-based theory of performance differences. That is, it does not explain performance differences purely by using market power, collusion or strategizing, but instead, it looks inside the firm and focuses on firms’ resources (Peteraf & Barney, 2003). Resources are the building blocks for final products and services produced by an organization (Akter et al., 2016; Amit & Schoemaker, 1993). Following (R. M. Grant, 2016) there are three kinds of resources, namely tangible (e.g., equipment, cash, property), intangible (e.g., organizational culture, organizational learning) and human resources (e.g., managerial skills).
Underlying the RBT are two main assumptions. First, not all firms own the same resources, that is, resources are distributed heterogeneously across firms. Resource heterogeneity implies that some firms have the capacity to accomplish certain tasks by deploying superior resources. Thus, they can do so more efficiently compared to their competitors (J. Barney, 1991). Second, the RBT assumes that resources are imperfectly mobile. That is, resources cannot easily be exchanged between firms. This suggests that advantages resulting from superior resources can be sustained as these resources are bound to the firm (Peteraf, 1993). However, not all resources can be a source of competitive advantage. To be a source of competitive advantage, resources need to be valuable (V), rare (R), imperfectly (I) imitable and properly organized (O) (J. B. Barney, 1995).
As described by (Peteraf & Barney, 2003), the RBT ‘provides a resource-level and
enterprise-level explanation of sustained performance differences among firms’ (p.312). It
differentiates between resources and capabilities controlled by an organization. While resources refer to resource-level differences, capabilities explain why some firms achieve superior performance on an enterprise-level (Peteraf & Barney, 2003). Following (Amit & Schoemaker, 1993, p. 35), capabilities resemble ‘‘intermediate goods’ generated by the firm
to provide enhanced productivity of its resources, as well as strategic flexibility’’.
Organizational capabilities are firm-specific, knowledge-based processes that are developed over time. They are the result of complex interactions between human capital, or human capital and other resources. Therefore, they can be a source of sustained competitive advantage as they are valuable, rare, immobile and properly organized (Akter et al., 2016; R. Grant, 1991). As a result, the RBT provides an interesting point of view for analysing and exploring BDAC for several reasons. Firstly, current big data research has highlighted the importance of building BDAC for achieving competitive advantage in the emerging data economy. Thus, organizations rely on internal resources and capabilities to gain a competitive advantage. As an efficiency-based theory, the RBT allows to closely examine which internal factors contribute to the emergence of superior big data analytics capabilities (J. B. Barney, 1995; Peteraf, 1993). Secondly, the RBT differentiates between three types of resources, which is in line with the emerging debate on BDAC (R. Grant, 1991). Several researchers have argued that organizations need a mix of tangible, intangible, and human resources for building advanced BDAC (Davenport, Barth, & Bean, 2012; Gupta & George, 2016; McAfee & Brynjolfsson, 2012). Therefore, the RBT provides suitable point-of-view for further analysing the specific resources contributing to superior big data analytics capabilities.
2.2 Big Data
Big Data refers to a continuous stream of vast amounts of data that are continuously being generated in our modern society. Most definitions within the emerging body of literature on big data build upon the three V’s framework which defines big data along three key dimensions (Germann, Lilien, Fiedler, & Kraus, 2014; Gupta & George, 2016; Jacobs, 2009; Russom, 2011; Wamba et al., 2015). The first dimension, velocity, refers to the real-time speed at which data is being generated. With the advent of hand-held devices, mobile internet, and social media a continuous flow of information and data has emerged that generates data in real-time. Secondly, Volume refers to the sheer amount of data that is being generated. According to (IBM, 2017), 2,5 quintillion bytes of data are generated every day. As a result, 90% of all data available today has been created in the last two years alone. Lastly, variety highlights the numerous different sources and formats of big data, e.g., visual, clickstream, structured, or unstructured data that result from the abundance of devices generating data (Russom, 2011). Following this framework, IBM defines the concept of big data as ‘[data that] comes from
sensors, devices, video/audio, networks, log files, transactional applications, web, and social media - much of it generated in real time and in a very large scale.’ (IBM, 2018).
Others, however, have highlighted the importance of specific dimensions. (Jacobs, 2009) describes big data as data too large to be processed or stored by traditional technologies, emphasizing volume as the key dimension, while (Schroeck, Shockley, Smart, Romero-Morales, & Tufano, 2012) focus on the broader scope of data, including real-time data, social media data or new-technology-driven data. Although velocity, volume, and variety have received the majority of attention in the literature, some scholars have extended the V framework by introducing other dimensions. (Davenport, 2014) highlights value as an additional dimension, that is, the business value that is embedded within big data. (White,
2012), on the other hand, suggests veracity, that is the inherent unpredictability and messiness of raw data, as a fifth dimension to big data.
For most organizations, the power of big data lies within predictive insights that can be derived by using advanced analytics and statistics (Dubey et al., 2017). Indeed, managing and analysing the overabundance of data has become a critical objective for firms in today’s data economy (Mazzei & Noble, 2017; McAfee & Brynjolfsson, 2012). BDA is expected to have a profound impact on a large number of industries. In manufacturing, firms can leverage BDA for improved business transformation, enhanced automation and improved process and asset monitoring (Davenport et al., 2012; J. Lee, Lapira, Bagheri, & Kao, 2013). In healthcare, BDA is expected to improve operational efficiency, while simultaneously improving overall life quality (Y. Wang, Kung, & Byrd, 2018). Furthermore, research in the fields of big data, IT, and business intelligence identified a positive link between customer insights gained from data analytics and firm performance (Germann et al., 2014). Amazon.com, one of the world’s largest retailers is already leveraging BDA to improve its retail business, where personalized customer recommendations based on BDA account for roughly 35% of all sales (Wills, 2014). What is more, big data analytics is becoming an essential part of the decision-making processes within organizations, allowing firms to “analyze and manage strategy through a data lens” (Brands, 2014, p. 65). As shown by (McAfee & Brynjolfsson, 2012) companies that ranked among the top third of their industry in the use of data-driven decision-making were found to be on average 5% more productive and 6% more profitable than their competitors. However, in order to fully leverage big data analytics, organizations need to build the relevant big data analytics capabilities.
2.3 Big Data Analytics Capabilities (BDAC)
Drawing upon the RBT, several scholars have conceptualized BDAC as higher-order capabilities. According to (Garmaki et al., 2016), BDAC reflects a firm’s capacity to leverage BDA resources and align BDA with the firm strategy to enhance firm performance and gain competitive advantage. Research has identified three critical aspects contributing to strong BDAC, namely organizational factors, human factors and physical factors (Akter et al., 2016). In this vein, (McAfee & Brynjolfsson, 2012) suggested talent management, decision-making capabilities, and IT infrastructure to be critical elements of BDA capabilities, while (Davenport et al., 2012) identified data-driven decision-making, data scientists, and advanced IT infrastructure capabilities (e.g., cloud-based computing, open-source platforms) as critical aspects of BDAC. However, given the infant stage of research, there is still a paucity of conceptual frameworks describing BDAC.
(Akter et al., 2016) introduced a third-order, hierarchical BDAC framework consisting of three second-order constructs, namely BDA infrastructure flexibility, BDA management capability and BDA talent capability. These three second-order constructs are composed of a total of eleven first-order constructs, namely BDA planning, investment, coordination, control, connectivity, compatibility, modularity, technical knowledge, technology management knowledge, business knowledge and relational knowledge (Akter et al., 2016; Wamba et al., 2017). Adding to this model, (Garmaki et al., 2016) identified relational BDA capabilities as a fourth dimension, suggesting firms’ abilities to access, share, coordinate, and reconfigure data within interfirm networks as a critical dimension for advanced BDAC.
Similarly, (Gupta & George, 2016) proposed a third-order, hierarchical framework for BDAC. However, following (R. M. Grant, 2016), they suggest a distinction between tangible, human and intangible aspects as constituents of a higher-order BDAC. While tangible
and human aspects include technical, managerial and talent-related aspects, the intangible dimension adds two important concepts that are critical for advanced BDAC, namely data-driven culture and intensity of organizational learning. As both of these intangible factors are important for firm performance, the framework by (Gupta & George, 2016) was chosen as the basis for this study.
2.3.1 Tangibles
Tangibles are resources that can either be sold or bought in markets. (Gupta & George, 2016) suggest three kinds of tangible resources that are critical for strong BDAC, namely basic resources, data, and technology. Basic resources refer to financial and non-financial investments made into big data by the focal firm. That is, organizations need to commit enough time, and other financial or non-financial resources to the creation of BDAC. (Akter et al., 2016, 2016; Gupta & George, 2016).
A second integral part of BDAC is the access to vast amounts of data. In order to fully leverage the potential of data analytics, organizations need to integrate both internal and external data into their operations. However, given the immense value embedded within big data, organizations and other players are reluctant to share data with other organizations. Following (Galbraith, 2014), leadership and organization design need to facilitate data sharing within organizations and within networks of organizations in order to unleash the full power of data analytics.
Lastly, organizations need to adopt technologies that can handle the processes of analysing vast amounts of diverse and unstructured data (Gupta & George, 2016). Without the right technologies, organizations will not be able to derive any valuable insights from data.
2.3.2 Human resources
Human resources are an essential factor for organizational success. These resources come in the form of employees’ knowledge, skills, problem-solving capabilities, relationships, or business acumen. That is, they are embedded in the members of an organization. Prior research on business intelligence, IT, and big data has identified technical and managerial skills as critical features pertaining to strong BDAC (Gupta & George, 2016; Wamba et al., 2017).
First, organizations need to ensure that employees have the relevant technical capabilities. That is, in order to analyze and extract insights from the big data, organizational members need to develop skills such as coding, working with AI, advanced analytics or cloud computing (Akter et al., 2016). Redeveloping current employees and hiring of new talent are critical for organizations to ensure the relevant knowledge and skills are accumulated within the organization.
However, insights extracted from data are of little value if there is a lack of suitable management. Big data managers need to understand and foresee where to apply new insights. They need to ensure technical staff understands business-related goals in order to align analytics function with business goals (Wamba et al., 2017). Furthermore, maximizing returns from big data analytics requires communication and collaboration throughout the entire organization, thus big data management needs to emphasize and facilitate information sharing and communication (Akter et al., 2016). What is more, achieving mutual trust and strong relationships between data managers is an important objective for organizations as it enables the formation of superior, hard-to-copy BDAC (Gupta & George, 2016). As opposed to technical skills, however, managerial skills cannot be acquired through hiring. They develop over time and are highly firm-specific. Managerial skills are embedded in the organizational DNA and can be described as norms through which managers full-fil their work (Gupta & George, 2016). They are a result of the context or setting within which employees and managers
perform their work. As a result, superior managerial skills may allow organizations to gain a competitive advantage (Gupta & George, 2016; Wamba et al., 2017).
2.3.3 Intangibles
While tangible resources can be assigned an explicit value, intangible resources are more ambiguous in nature. Their value is highly context-specific, and they do not have clear boundaries. Nonetheless, intangible resources have received much attention from management scholars and are thought to be critical factors contributing to firm performance in dynamic markets (Gupta & George, 2016). Most intangible resources, with the exception of trademarks, patents, or other intellectual property, cannot be bought and sold through markets. Moreover, most intangible resources conform to the VRIN criteria, thereby making them a potential source of sustained competitive advantage (Teece, 2014). Following (Gupta & George, 2016), two types of intangible resources critically affect a firm’s ability to derive value from big data, namely data-driven culture and intensity of organizational learning.
The first concept, data-driven culture refers to the extent to which an organization’s decision-processes rely on insights extracted from data. This is critical for organizations as data-driven decisions tend to be better than decisions based on experience and intuition (McAfee & Brynjolfsson, 2012). However, many organizations struggle to fully embrace data-driven decision-making and still rely on past paradigms such as the highest paid person opinion (HiPPO) or other decision-structures. Furthermore, it is critical for organizations to ensure the dissemination of data-driven decision-making throughout the entire organization, ensuring that members at all levels have the ability to make informed decisions based on evidence and insights derived from data (Gupta & George, 2016; McAfee & Brynjolfsson, 2012; Ross, Beath, & Quaadgras, 2013).
Another intangible resource that has been linked to BDAC is the intensity of organizational learning (Gupta & George, 2016). Adding to the static view of the BRT, (Teece, Pisano, & Shuen, 1997) proposed that firms need to reconfigure their resource-bases in order to remain competitive in changing environments. The ability to do so is reflected in a firm’s intensity of organizational learning which describes an organization’s ability to explore, store and apply new knowledge (R. M. Grant, 1996). Organizational learning has two important implications for understanding the formation of BDAC. First, organizations with a higher intensity of organizational learning are able to leverage larger knowledge stocks for creating BDAC. Second, such organizations may also leverage their extensive knowledge base to better understand and incorporate new insights into ongoing business processes. Thus, they can extract more value from available data (Gupta & George, 2016).
2.4 Organizational culture
Organizational culture is an important aspect that has long been linked to performance and other favourable organizational outcomes. Since 1980 more than 4600 articles on organizational culture have been published in an attempt to create a pervasive understanding of the topic (Hartnell, Ou, & Kinicki, 2011). Despite an ongoing debate concerning an exact definition of the term, and the many connotations associated to the concept, scholars have adopted the view that organizational culture is a manifestation of a shared set of norms, values, and beliefs within the firm (J. B. Barney, 1986; Hartnell et al., 2011; Schein, 2010; Smircich, 1983). According to (Ribière & Sitar, 2003), organizational culture critically influences informal parts of an organization. Thereby, organizational culture affects employees’ behaviour, ways of communicating, and relationships with colleagues (Denison & Mishra, 1995).
Organizational cultures, that is, norms, values, and beliefs, are shaped by an interplay of both, leadership and managerial systems (Yukl, 2008). Leadership influences culture primarily through behaviour, for example by articulating a compelling mission, leading example or explaining why new initiatives are needed. The importance of leadership and leadership style for organizational performance has been subject for a large body of research (Podsakoff, Mackenzie, Moorman, & Fetter, 1990; Sarros, Cooper, & Santora, 2008). Managerial systems on the other hand further enhance the impact of leadership behaviour on corporate culture by facilitating employee empowerment. Intellectual stimulation through transformational leadership may only significantly enhance innovation output if managerial systems encourage innovative behaviour through well-aligned reward systems or by providing a climate of psychological safety for risk-taking (Yukl, 2008).
While there is agreement on the mechanisms that give rise to organizational culture, there is no consensus on a uniform framework for the phenomenon. A multitude of frameworks for organizational culture can be identified in the literature (Smircich, 1983). One of the most commonly cited frameworks is the Competing Values Framework (CVF) introduced by (Quinn & Rohrbaugh, 1983). It conceptualizes culture along two dimensions, namely focus and structure. The focus dimension emphasizes the extent to which an organization focuses on its internal capabilities and employees, as opposed to an external focus emphasizing differentiation. The structure dimension, on the other hand, differentiates organizations that accentuate flexibility from those emphasizing stability and control (Hartnell et al., 2011). The CVF thereby differentiates between flexibility orientations and control orientations. In a similar fashion, (Harrison, 1972) described organizational culture along the dimensions of centralization and formalization, while (Deal & Kennedy, 1983) conceptualized culture as a two-dimensional model consisting of feedback speed and risk propensity. Other scholars have analyzed specific types of organizational cultures such as bureaucratic, innovative, and
supportive organizational culture support (Groysberg, Lee, Price, & Cheng, 2018; J. Wallach, 1983).
Another important aspect of analyzing organizational culture is the level of analysis (Hofstede, 1998). Some authors argue that distinct cultures may exist at different levels of an organization (Groysberg et al., 2018). That is, organizations may develop individual subcultures based on factors such as geographical location of business units, nature of tasks within departments, or regional leadership. While some scholars analyzed cultural on an organization-wide level, others have examined culture at the business unit or even the department level. For this study, culture is analyzed at an organizational level, where it is considered an important factor contributing to overall firm performance (J. B. Barney, 1986; Groysberg et al., 2018).
2.4.1 Innovative organizational culture
In recent years, technology companies such as Google or Microsoft have redefined the rules of competition by outperforming competitors both financially and in terms of innovation. A unique feature of many of these tech companies is their innovative organizational cultures (Groysberg et al., 2018). Consistent with culture research, an innovation culture is a manifestation of norms, values, and beliefs shared by members of a firm. It is characterized by the extent to which an organization values change and encourages innovation (Schleimer & Pedersen, 2013). Several scholars have emphasized the importance of innovative organizational cultures for firms’ long-term success (Menon, Bharadwaj, Adidam, & Edison, 1999). Following (Merrifield, 2000), establishing an organizational culture that continuously fosters innovation has become a key objective for firms in dynamic markets. Firms with innovative cultures aim for constant renewal of their organization and strategy, thereby leveraging their resources in new, superior ways (Slater & Narver, 1995). These organizations
emphasize openness to new ideas and value creativity (Schleimer & Pedersen, 2013). An innovative culture increases the proclivity for information and knowledge sharing within the organization, thereby positively affecting knowledge creation. Furthermore, management in organizations with innovative organizational cultures actively nurtures diverse views and encourages innovative behavior, thereby creating wider mental model roadmaps within the organizations (Cohen & Levinthal, 1990). Wider mental roadmaps allow organizations to identify emerging opportunities more easily, as these opportunities are more likely to relate to existing knowledge stocks (Schleimer & Pedersen, 2013).
2.5 Organizational structure
Organizational structure has become an important part of management studies over the past decade. Designing organizational structures that facilitate knowledge creation, knowledge sharing, and knowledge integration is a key objective for firms operating in a dynamic environment (Chen & Huang, 2007). The most frequently discussed concepts within organization design literature are formalization, integration, and centralization. Formalization refers to the extent to which jobs within an organization are guided by formal rules, policies, and regulations (H. Lee & Choi, 2003). Some scholars have associated higher levels of formalization with increased efficiency and knowledge performance (Pertusa-Ortega, Zaragoza-Sáez, & Claver-Cortés, 2010a; Ruekert et al., 1985). While others have suggested that less formalization facilitates social interaction and innovative behavior (Chen & Huang, 2007). Integration, on the other hand, refers to the extent to which organizational subdivisions or business units operate interrelatedly (Germain, 1996). Integration increases intra-organizational communication, reconciles diverging goals, or increase assertiveness of managers by providing them with combined bargaining power (Miller, 1987). What is more, integrated structures facilitate the formation of more advanced communication channels within
the organization allowing employees to more efficiently share knowledge and expertise (Janz & Prasarnphanich, 2003).
Centralization, on the other hand, refers to the concentration of authority and power in a firm (Chen & Huang, 2007; Zheng, Yang, & McLean, 2010). Several scholars have examined the relationship between centralization and firm performance (Rangus & Slavec, 2017; Robert Baum & Wally, 2003; Zheng et al., 2010). Despite a small number of studies indicating a positive relationship between centralization of decision-making and firm performance, the vast majority of studies suggest a positive relationship between decentralization of decision-making and firm performance (Ruekert et al., 1985; Zheng et al., 2010). Decentralization creates an environment of participation which increases employee communication, involvement, and commitment (Damanpour, 1991). Higher levels of involvement have been found to increase job satisfaction and intrinsic motivation, as employees perceive their work as meaningful (Dewar & Werbel, 1979). What is more, higher employee involvement in decision-making processes fosters knowledge creation by incorporating a greater variety of knowledge and information (Pertusa-Ortega, Zaragoza-Sáez, & Claver-Cortés, 2010b). Further, decentralization increases social interaction between members of an organization, thereby, increasing information sharing within an organization (Chen & Huang, 2007). As a result, decentralized structures facility the creation of novel ideas by enhancing creative processes and knowledge-creation (Pertusa-Ortega et al., 2010). Moreover, decentralized structures facilitate faster decision-making within organizations and, consequently, higher responsiveness to external events (Jones, 1993; Robert Baum & Wally, 2003).
Centralized structures, on the other hand, inhibit the creation of creative ideas. Interdepartmental communication and sharing of novel ideas are less frequent in centralized organizations (Pertusa-Ortega et al., 2010). Furthermore, time-consuming communication channels in such organizations create discontinuities and distortions in creative processes (H.
Lee & Choi, 2003). What is more, centralization of operational tasks significantly decreases decision-speed within organizations (Robert Baum & Wally, 2003).
3. Theoretical Framework & Hypotheses
3.1 Big Data Analytics Capabilities and firm performance
While BDA research is still in very early stages, there is already a broad agreement on the positive relationship between advanced analytics and firm performance. The vast amounts of data that are continuously being generated greatly surpass human capabilities for structuring and analyzing. However, using advanced analytics can enable firms to extract insights and ideas from this abundance of unstructured data (Garmaki et al., 2016). Thus, big data analytics enables firms to extract useful insights from data, thereby increasing organizational performance. (Wamba et al., 2017) showed that BDAC significantly increases an organization’s process-orientated capabilities. Furthermore, (Gunasekaran et al., 2017) showed that big data analytics and predictive analytics positively influence supply-chain performance and consequently firm performance. In addition, BDAC literature has identified several avenues through which organizations derive value from big data analytics (Wamba et al., 2015). BDA can help organizations innovate their business model, products or services. It allows firms to increase operational efficiency in supply-chains, operations, or other organizational processes. Therefore, BDA influences both, market performance as well as operational performance of the focal firm. It is important, however, to understand that investments in big data analytics do not guarantee an increase in performance. In order to fully leverage the data available to an organization, it is critical for the firm to build and develop advanced BDAC (Akter et al., 2016; Gupta & George, 2016; Wamba et al., 2017). That is, they need to develop the right managerial and technical skills while building stocks of tangible and
intangible resources required for big data analytics. Thus, the study postulates the following hypothesis:
H1a (+): There is a positive relationship between BDAC and market performance (MP). H1b (+): There is a positive relationship between BDAC and operational performance
(OP).
3.2 Innovative organizational culture and firm performance
In order to maximize the value derived from big data analytics, it is important that organizations nurture a culture that facilitates data-driven decision-making. Additionally, organizations should nurture a culture that facilitates innovative behavior and emphasizes exploration. An innovative organizational culture increases the propensity of an organization to analyze and incorporate new information and to engage in an in-depth examination of strategic alternatives. Thus, such organizations will engage in more exploratory analysis of the vast amounts of data available aiming to identify new products, markets or business models (Mazzei & Noble, 2017). What is more, (Cohen & Levinthal, 1990) argue that an innovative culture leads to an increased knowledge diversity within organizations. Therefore, firms with an innovative culture can easier relate information and insights gained from analytics to existing internal knowledge. In addition, the resulting wider mental maps may enable firms to recognize opportunities for creating novel products and services before their competitors, thus increasing market performance of the focal firm relative to its competitors (Wamba et al., 2015). Equally important, organizations with an innovative organizational culture are likely to embrace innovative cost-saving measures or technologies based on insights derived from BDAC (Wamba et al., 2015). As a result, an innovative organizational culture will increase the
amount of valuable information an organization can extract from a given amount of data, thereby further increasing the firm’s performance. Hence, it is proposed:
H2a (+): Innovative organizational culture positively moderates the relationship between BDAC and MP so that the effect is stronger when innovation culture increases.
H2b (+): Innovative organizational culture positively moderates the relationship between BDAC and OP so that the effect is stronger when innovation culture increases.
3.3 Decentralization of decision-making and firm performance
As introduced by (Galbraith, 2014), the continuous stream of data that is generated at real-time will require organizations to increase their decision-speed to be able to gain a competitive advantage in today’s data-driven economy. Allocating more decision-rights to the lower levels of the organization positively influences the decision-speed, as organizational members may act upon insights extracted from data without using time-consuming vertical communication channels (Robert Baum & Wally, 2003). That is particularly relevant for introducing new products or services in times of short-product life-cycles and fast-changing demand patterns. Furthermore, organizations with decentralized decision-making are likely to be more efficient in developing tailored solutions and offerings to customers (Galbraith, 2014). What is more, decentralization of decision-making increases knowledge diversity and information sharing within the organization (Chen & Huang, 2007; Pertusa-Ortega et al., 2010). As a result, insights gained from BDA a more likely to relate to existing knowledge, which allows organizations to better integrate new information and insights into products and services (Cohen & Levinthal, 1990; Galbraith, 2014). Moreover, increased communication and information sharing, resulting from decentralization of decision-making is likely to increase the flow of information throughout the organization. Thus, business units and other
organizational divisions can leverage a larger pool of information and data to generate useful insights (Galbraith, 2014). Thus, the study postulates:
H3a (+): Decentralization of decision-making positively moderates the relationship between BDAC and MP so that the effect is stronger when decentralization of
decision-making increases.
H3b (+): Decentralization of decision-making positively moderates the relationship between BDAC and OP so that the effect is stronger when decentralization of
decision-making increases.
The aforementioned hypotheses are illustrated in the research model, as shown in Appendix A.
4. Methodology
4.1 Data Collection
The study used data that was collected through a survey targeting organizations that are utilizing big data analytics. The study identified big data analytics managers and senior managers as primary respondents. These positions were perceived to have enough relevant knowledge regarding organizational factors relevant to this study. Top-level managers (CIO, CTO, CDO) were also contacted as respondents, however, obtaining contact information and answers of said individuals proved highly difficult. A possible reason for this may be the setting of a master thesis, as C-Level individuals may deem such research not worthy of their limited time. Nonetheless, to ensure the respondents have the relevant knowledge for this study, respondents were asked whether they directly or through subordinates manage big data analytics within the firm. The survey was distributed using random convenience sampling.
Respondents were approached via LinkedIn by searching for the term ‘Big Data Analytics Manager’ in Denmark, Netherlands, Belgium, France, Finland, Spain, Norway, Switzerland, Ireland, Sweden, and Germany. In order to get contact information of the relevant individuals, a software called ‘Rocketreach’ was purchased. The software allows users to search and find business e-mails of individuals on LinkedIn or other social media sites. A total of 1050 relevant individuals were identified during the search period which lasted one and half weeks. During the search, it was ensured that no company was added to the mailing list twice, further when identifying several individuals from the same organizations a preference was given based on the current position held by the individual. That is, CIO or CDO positions were preferred over VP positions or big data analytics managers.
After concluding the data collection, 169 responses were registered. The data screen was done using SPSS. During the data screening, 95 responses had to be deleted because of too missing data. Furthermore, another six responses were deleted because of careless answers, that is said individuals either filled in the survey within just a few minutes or showed continuous response patterns (Meade & Craig, 2012). After screening the data, a total of 68 useful responses remained for further analysis. The characteristics of the final sample are illustrated in Appendix B. Following (Gupta & George, 2016), firm size and years of BDA experience were used to classify the individual responses. As indicated, the sample showed a relatively even distribution with respect to BDA experience. Furthermore, all organization sizes were represented sufficiently.
4.2 Measures
All measures used in the survey have been validated in previous studies and were adopted without any changes. Since the data collection was shared with three other students, the entire survey comprised a total of 70 questions including population questions and filter
questions to ensure the fit of the respondents. All variables were placed in individual, randomly ordered sections to reduce common method bias. Further, respondents were ensured that their answers were treated confidentially and that there are no right or wrong answers (MacKenzie, Podsakoff, & Podsakoff, 2011).
4.2.1 Dependent variable
In order to measure firm performance, a scale introduced by (Gupta & George, 2016) was used. This scale consists of two individual constructs, namely, market performance and operational performance (Gupta & George, 2016; Ravichandran, Lertwongsatien, & Lertwongsatien, 2005; N. Wang, Liang, Zhong, Xue, & Xiao, 2012). Each construct consists of four items measuring the focal firm’s performance relative to its competitors using a seven-point Likert scale (Appendix C). Self-reported data was used as it is difficult to find reliable accounting data on firm performance, especially market performance (N. Wang et al., 2012). Furthermore, prior studies have found self-reported data to be highly correlated with objective data (Noel Capon, John U. Farley, Donald R. Lehmann, & James M. Hulbert, 1992; Ravichandran et al., 2005). What is more, using subjective data is a useful alternative to objective data for measuring multidimensional performance data (Gregory G. Dess, Richard B. Robinson, & Jr., 1984). Finally, by measuring performance relative to competitors, strategic group effects and industry related effects can be mitigated to allow for a more holistic comparison (N. Wang et al., 2012).
4.2.2 Independent variable
As stated in chapter 2.3, the measurement scale introduced by (Gupta & George, 2016) was used to operationalize BDAC. The scale consists of a total of 32 individual items measured using a seven-point Likert scale (Appendix C). The hierarchical model represents the formal
relationships between the individual indicators, sub-dimensions, and higher-order constructs. The scale asked respondents to what extent their organizations possess the necessary resources contributing to BDAC. Following (Gupta & George, 2016), the scale differentiated seven individual first-order variables, namely data, technology, basic resources, managerial skills, technical skills, data-driven culture and intensity of organizational learning. Given the nature of several first-order variables, obtaining objective data appeared to be highly complicated. That is, ambiguous concepts like managerial skills and data-driven culture are hard to measure by means of financial data or annual reports. Thus, self-reported data was chosen for the data collection.
4.2.3 Innovative organizational culture
Innovative organizational culture was measured using a four-item scale introduced by (Schleimer & Pedersen, 2013). The scale is adopted from prior studies by (Menon et al., 1999). Like the other primary constructs, innovative organization culture was measured using a seven-point Likert scale. The questions, as shown in Appendix C, asked respondents to rate to what extent their organization values innovation and change.
4.2.4 Decentralization of decision-making
To operationalize decentralization of decision-making the scale utilized by (Giampaoli, Ciambotti, & Bontis, 2017) was adopted. This scale is a combination of several items from two individual prior studies. The first three items were initially introduced by (H. Lee & Choi, 2003), while the fourth item was adopted from (Kamhawi, 2012). All the items were measured using seven-point Likert scales and self-reported data. The respondents were asked to what extent making is decentralized in their organizations, that is, how much
decision-4.2.5 Control variables
To account for external differences between organizations, two control variables (firm size and years of BDA experience) were included. These controls have been utilized in prior studies in the field of big data analytics capabilities (Garmaki et al., 2016; Gunasekaran et al., 2017).
Firm size. Following several prior studies on BDAC, firms size was chosen as a control
variable to mitigate any effects related to organizational size (Garmaki et al., 2016; Gunasekaran et al., 2017). Firms size was measured by the number of employees, which is in line with prior studies that have identified the number of employees is an appropriate measure for firm size in any given industry (Audia & Greve, 2006). Respondents were asked to indicate the size of their organizations by choosing one of five predefined categories. The scale for firm size was adopted from (Gupta & George, 2016) and had been used in prior BDA research. Controlling for firm size is important, as large organizations tend to have higher stocks of resources available for use. Furthermore, large organizations are likely to have higher revenues, profits, and opportunities to compete than smaller organizations. Hence, controlling for organization size allows mitigating the effects of size on the relationship between BDAC and FPER (MP, OP)
Years of BDA experience. Developing strong BDAC is a process that requires
organizations to devote the necessary resources and time to build these capabilities. Specifically, human resources and intangible resources cannot readily be bought in markets. Thus, organizations need to build these resources over time. Organizations with more experience in BDA are therefore expected to leverage their BDAC better than other organizations, that have only recently made investments into BDA. To account for such age effects, we controlled for the years of BDA use. Using the scale introduced by (Gupta &
George, 2016) the respondents were asked to indicate the time the organizations have been using BDA.
4.3 Analysis Method
To assess the research model, structural equation modeling (SEM) was identified as the method of choice. There are two different kinds of structural equation modeling, namely covariance-based SEM (SEM) and partial least squared path modeling (PLS-SEM). CB-SEM is most suited for confirming or rejecting theories. PLS-CB-SEM, on the other hand, is primarily used for developing and testing theory through explanation and prediction (Akter et al., 2016). That is, PLS-SEM aims to explain the variance in the independent variable when examining the model (Hair, Hult, Ringle, & Sarstedt, 2017). Given the research design, there are several reasons why PLS-SEM is particularly well suited for this study. First, as argued by (Joe F. Hair, Ringle, & Sarstedt, 2011), PLS-SEM is an appropriate approach for assessing hierarchical composite models, specifically constructs containing formative measures. Given the complex design of the hierarchical BDAC construct which is involving several formative indicators, SEM provided an adequate method for the analysis. Another benefit of PLS-SEM lies in its ability to avert the constraints of model complexity, model identification, sample size, factor intermediary, and distributional properties (Hair et al., 2017). Since the sample size (N=68) of the study was relatively small, PLS-SEM appeared to be the best choice for the analysis as it reduced the effects of the sample size.
The study used SmartPLS 3.0 to estimate the model. As suggested by prior PLS-SEM literature, the analysis was split into two parts (Chin, Esposito Vinzi, Henseler, & Wang, 2010). First, the measurement model, containing the indicators and their relationships with the constructs, was assessed. To ensure the validity of the BDAC construct, a confirmatory factor
analysis was carried out. Subsequently, the structural model which includes the latent variables and their relationships was examined to test the hypotheses.
4.4 BDAC model specification
Following (Gupta & George, 2016), the hierarchical construct BDAC was specified, as shown in Appendix D. The repeated indicator approach was used to estimate all lower- and higher-order constructs simultaneously. This is in line with the guidelines of hierarchical modeling as introduced by (Becker, Beverungen, & Knackstedt, 2010; Chin et al., 2010). The first-order latent variables Data, Technology, and Basic resources were modeled as Mode B ‘formative’, while the other four first-order latent variables were modeled as Mode A ‘reflective’. All first-order latent variables were subsequently connected to their corresponding indicators. The second- and third-order variables were then constructed by reiterating the indicators of their underlying latent variables and the lower-level latent variables themselves. Thus, the ‘intangible’ construct was connected to the indicators of organizational learning and data-driven culture, while the ‘tangible’ construct was made up of the indicators for basic resources, and technology. The indicators for managerial and technical skills were connected to the ‘human resources’ construct. Likewise, the third-order construct BDAC was connected to all the indicators of the second-order latent variables. All second- and third-order variables were modeled as mode B ‘formative’.
5. Analysis & Results
The following section explains the steps of the data analysis and includes the results that were obtained. First, the measurement model was examined to assess the validity and
reliability of all individual constructs. Subsequently, the structural model was assessed to understand the relationships between the constructs.
5.1 Evaluation of the measurement model
The study used the regular PLS algorithm, to obtain all required measures. SmartPLS 3.0 also offers the possibility of opting for the newer ‘consistent PLS Algorithm’, however, the consistent algorithm is still lacking adequate empirical validation (Joe F. Hair et al., 2011). To obtain both t-values and p-values, non-parametric bootstrapping with 1000 sub-samples was applied.
First, the psychometric properties of the reflective first-order constructs were examined. All reflective indicators had outer loadings exceeding the cut-off value of 0.7, except TS1, TS2, MP4, and CEN3 (Chin et al., 2010). The aforementioned indicators were deleted individually to assess the model without each indicator. Following (Hair et al., 2017), indicators with loadings between 0.4 and 0.7 should be dropped, if dropping them increases composite reliability (CR) and average variance extracted (AVE) of the relevant first-order construct. As a result, the decision was made to drop all four indicators before continuing the analysis. The final loadings of the reflective indicators are shown in Appendix E. To confirm the convergent validity of the measurement scales, the AVE of the first-order reflective constructs were calculated, as shown in Table 1. The AVE indicates the amount of variance a construct captures from its indicators in relation to the variance that can be attributed to measurement error (Chin et al., 2010). The AVE of all first-order reflective constructs exceeded the cut-off value of 0.5, indicating adequate convergent validity (Hair et al., 2017). Furthermore, the study calculated the composite reliability (CR) and Cronbach’s of the first-order reflective constructs. All measures exceeded the threshold of 0.8 recommended for CR.
of 0.7, as shown in Table 1 (Fornell & Larcker, 1981). To ensure the discriminant validity of the first-order reflective constructs, the Fornell-Larcker criterion was assessed. Following (Fornell & Larcker, 1981), the square root of the AVE must exceed any of the intercorrelations between the constructs to ensure discriminant validity. The square root of the AVE of each reflective construct was calculated, as shown on the diagonal in Table 1. All values were found to be larger than any of the intercorrelation between the constructs. Hence, the discriminant validity of all first-order reflective constructs is assured according to the Fornell-Larcker criterion. Furthermore, the cross-loadings of the first order constructs revealed that indicators loaded heavily on their corresponding construct in comparison to any other constructs, yielding further support for the validity of measurement model.
More recently, (Henseler, Ringle, & Sarstedt, 2015) suggested a more sophisticated approach for assessing the discriminant validity, namely the heterotrait-monotrait ratio (HTMT). The HTMT ratio “is based on the average of the correlations of indicators across
constructs measuring different phenomena relative to the average of the correlations of indicators within the same construct” (Gupta & George, 2016, p. 1057) Following (Henseler
et al., 2015), HTMT values below 0.85 provide sufficient support for discriminant validity.
Table 1
Intercorrelations of the latent variables for first-order constructsª
Number Construct Alpha CR AVE 1 2 3 4 5 6 7 8 9 10 11
1 T N/A N/A N/A N/A
2 D N/A N/A N/A 0.597 N/A
3 BR N/A N/A N/A 0.452 0.459 N/A
4 TS 0.887 0.922 0.746 0.413 0.302 0.319 0.864 5 MS 0.942 0.954 0.777 0.250 0.420 0.455 0.557 0.881 6 DDC 0.840 0.887 0.612 0.411 0.409 0.371 0.546 0.537 0.782 7 OL 0.915 0.937 0.749 0.585 0.544 0.419 0.469 0.478 0.575 0.865 8 Inn. Culture 0.904 0.933 0.778 0.360 0.214 0.421 0.357 0.472 0.527 0.556 0.882 9 Decentr. 0.927 0.954 0.873 0.365 0.206 0.383 0.289 0.373 0.598 0.444 0.710 0.934 10 MP 0.827 0.885 0.661 0.225 0.273 0.494 0.285 0.476 0.343 0.496 0.670 0.400 0.873 11 OP 0.868 0.911 0.720 0.039 0.145 0.145 0.147 0.344 0.425 0.296 0.408 0.304 0.570 0.848
As shown in Table 2, the results gave clear confirmation of the discriminant validity of the reflective constructs, as all values were below the cut-off value of 0.85.
In order to assess the formative first-order constructs, the outer weights of the indicators were examined. Most indicators had significant weights, except D3, T1, T2, T3, T5, and BR1 which had non-significant weights. However, it is not recommended to cull indicators of formative constructs simply because of non-significant weights (Petter, Straub, & Rai, 2007). Since none of the aforementioned indicators behaved this way in prior studies that have deployed the BDAC construct the decision was made to keep D3, T1, T2, T3, and BR1 for further analysis (Gupta & George, 2016). This is also in line with the approach for assessing formative indicators, as introduced by (Hair et al., 2017) who argued that formative indicators with insignificant weights should be kept for analysis if their outer loadings exceeded the recommended cut-off value of 0.5. This was the case for all of the above mentioned formative indicators, as shown in Appendix E. Furthermore, T5 was found to have a negative weight. (Cenfetelli & Bassellier, 2009) suggest that formative indicators with negative weights may also be kept for analysis if they do not show any multicollinearity problems. While multicollinearity is not an issue with reflective constructs, it is highly problematic with formative constructs. To ensure multicollinearity was not a problem, the VIF values were calculated using SmartPLS 3.0. According to (Petter et al., 2007), VIF values of less than 3.3 for formative constructs are a sign of low multicollinearity. As shown in Table 3, all formative
Table 2
HTMT of first-order reflective constructs
Number Consruct 1 2 3 4 5 6 7 8 1 TS 2 MS 0.604 3 DDC 0.639 0.594 4 OL 0.508 0.514 0.647 5 Inn. Culture 0.388 0.509 0.611 0.614 6 DEC 0.305 0.398 0.678 0.478 0.781 7 MP 0.326 0.529 0.394 0.557 0.755 0.443 8 OP 0.182 0.368 0.490 0.324 0.452 0.334 0.674
constructs showed VIF values of less than 3.3 suggesting low multicollinearity. As a result, T5 was kept for the analysis of the structural model. Having tested for multicollinearity, the study then assessed the discriminant validity. The discriminant validity of the formative constructs was assessed by looking at the correlations between the first-order formative constructs and other constructs. All of the correlations were found to be below the cut-off value of 0.71 (MacKenzie et al., 2011). What is more, the study also calculated Edwards Adequacy coefficient R2
a assess the validity of formative constructs.R2a was calculated by summing the
squared correlation between a formative construct and its dimensions and dividing the result by the number of dimensions. All values except tangible resources exceeded the recommended threshold of 0.5 (Edwards, 2001). However, as noted by (MacKenzie et al., 2011), Edwards Adequacy coefficient is still relatively new measure that is lacking substantial empirical validation. As a result, the decision was made to keep the technology construct for further analysis as there were no multicollinearity issues related to the technology construct. Similar to the first-order formative constructs, the higher-order formative constructs were assessed by examining multicollinearity, path coefficients, t-values, and R2
a.
Table 3
Higher-order Construct Validation
Construct Measures Weight T-statistics p-value VIF R2aª
Technology T1 0.127 0.486 0.627 2.098 0.47 T2 0.180 0.862 0.389 1.487 T3 0.244 1.282 0.200 1.907 T4 0.740 4.740 0.000 1.373 T5 -0.018 0.084 0.933 2.174 Data D1 0.508 2.908 0.004 1.210 0.57 D2 0.559 3.160 0.002 1.358 D3 0.220 1.019 0.309 1.301 Basic ResourcesBR1 0.393 1.380 0.168 1.573 0.79 BR2 0.713 2.888 0.004 1.573 Tangibles Technology 0.411 1.753 0.080 1.676 0.66 Data 0.416 1.826 0.068 1.636 Basic Resources 0.389 2.357 0.019 1.350
Human Technical Skills 0.513 2.531 0.012 1.448 0.76 Managerial Skills 0.606 2.817 0.005 1.448
Intangibles Data-driven culture 0.511 4.350 0.000 1.486 0.78 Org. Learning 0.613 4.991 0.000 1.486
BDAC Tangibles 0.347 10.064 0.000 1.878 0.77 Human 0.427 10.561 0.000 2.056
Intangibles 0.368 8.522 0.000 2.462 ª Edwards adequacy coefficient
5.2 Evaluation of the structural model
The structural model contained the relationship between the latent variables, that is BDAC, Decentralization of decision-making, Innovation culture, MP, and OP, as well as the interaction terms. Following (Ravichandran, 2017), a two-stage hierarchical component model analysis is recommended for models involving reflective-formative constructs. In the first stage, the main model without any interaction terms was assessed (Appendix F). Following this, the interaction terms were included in the model to examine the interaction effects (Appendix G). This is in line with prior BDA and IT studies that have deployed PLS-SEM (Akter et al., 2016; Ravichandran et al., 2005).
As suggested by (Hair et al., 2017), the main model was first tested for any collinearity issues. Similar to the measurement model, VIF values smaller than 3.3 indicated significantly low levels of multicollinearity. All values were found to be below the recommended cut-off value, indicating that there are no problems with multicollinearity. Next, the predictive power of the structural model was assessed by examining the coefficient of determination (R2) of the
dependent variable. R2 indicates the total effects of the exogenous variables on the endogenous
variable in the model. Thevalues for MP and OP were found to be 0.565 and 0.301 respectively. This indicates that the model has weak to moderate predictive power for the dependent variables (Hair et al., 2017). Furthermore, the path coefficients of the base model were found to be significant and the p=0.001 level. As hypothesized, BDAC has a strong positive relationship with both MP (ß= 0.469; t= 3.593) and OP (ß= 0.540; t= 4.003), thus supporting
hypotheses H1a and H1b.
Following prior research, the study also examined the effect of the control variables years-of-BDA usage and firm size (Akter et al., 2016; Gupta & George, 2016). A control model was tested involving the two variables. However, as shown in Table 4, neither of the control variables showed a significant relationship with FPER.
After assessing the main model, the interaction model was examined. The four interaction terms were specified using SmartPLS 3.0. Following (Chin et al., 2010; Hair et al., 2017), the interaction terms were included into the model using the two-stage approach provided by SmartPLS 3.0. The software first calculated the latent variable scores for the predictor and moderator variables using the main model. The latent variable scores were then saved and used to calculate the product indicator for the interaction analysis. The study followed (Hair et al., 2017) guidelines for assessing structural models in PLS-SEM. Again, the VIF values were examined to ensure there were now multicollinearity issues. The results indicated that there were no collinearity issues as all values were below 3.3. The R2 of the
depended variables MP (R2=0.599) and OP (R2 =0.377) indicate that the main model explains
59.9 and 37.7 percent of the variation of MP and OP, respectively. Thus, the model has weak to moderate prediction power, following (Hair et al., 2011). According to PLS-SEM literature, the small sample sizes may be one of the reasons for lower R2 values (Hair et al., 2017). The
path coefficients indicated a positive relationship between BDAC and both MP and OP. To assess the significance of the interaction effect the path coefficients and t-values were examined (Chin, Marcolin, & Newsted, 2003). Table 4 shows the obtained path coefficients, t-values, and p-values for the interaction effects. As indicated, the interaction effects between innovative organizational culture and both MP and OP were found to have path coefficients of -0.004 (t=0.041) and 0.211 (t=1.289), respectively. Both interactions proved to be insignificant at the 10% level. As a result, H2a and H2b were not supported.
Similarly, the interaction effect between decentralization of decision-making and MP was found to be insignificant with a path coefficient of -0.133 (t = 1.215) at the 10% level.
Thus, H3a was not supported.
However, the interaction effect of between and decentralization of decision-making and OP proved significant with a path coefficient of -0.266 (t= 1.715) at the 10% level. Contrary