Master Thesis
The influence of big data analytics capability on firm performance: the
mediating role of innovation ambidexterity
MSc. Business Administration Digital Business Track
Name Sofia de Haan Student number 10597093
Date 22-06-2018
2
Statement of originality
This document is written by Sofia de Haan 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.
3
Abstract
Big data has become a popular topic among scholars and practitioners. In particular, the ability to analyze and extract insights from big data is currently considered an important competitive weapon. Recent literature has identified a number of resources that combined build a big data analytics (BDA) capability, which has been tested in relation to firm performance. Building on this stream of research, this study examines the relationship between BDA capability and firm performance, highlighting the role of innovation ambidexterity as a mediator. Innovation ambidexterity refers to a firm’s ability to exploit current resources while simultaneously exploring new opportunities for both incremental and radical innovation, and has been subject to numerous studies in the management field. Data was collected by distributing an online survey among BDA managers, which provided a sample of 62 respondents. The results show that BDA capability positively influences firm performance, however, innovation ambidexterity is not found to be a mediator of this relationship. The findings of this study imply that having a BDA capability can be of great importance for firms to achieve better performance. Furthermore, this study contributes to a better understanding of how BDA capability can be turned into a competitive advantage.
4
Table of contents
1. Introduction………. 5
2. Literature review………. 8
2.1 Resource-based view and IT capability……….. 8
2.2 Big data……… 12
2.3 Big data analytics capability……… 14
2.4 Innovation ambidexterity ……….... 19
3. Data and method………. 25
3.1 Research design………... 25
3.2 Conceptual model……….... 25
3.3 Sample………. 26
3.4 Data collection………. 27
3.5 Measures……….. 28
3.6 Common method bias……….. 30
4. Analysis and results……….. 31
4.1 Descriptive statistics………. 32
4.2 The BDA capability model assessment……… 33
4.3 Hypotheses testing………... 37
5. Discussion and conclusion……….. 41
5.1 Theoretical contributions………. 42
5.2 Managerial implications……….. 44
5.3 Limitations and future research……… 45
6. References………. 47
5
1. Introduction
Investment in information technology (IT) to support strategic decision making has become a crucial asset for firms to enhance performance. Organizations often invest millions of dollars in IT to increase productivity, enhance customer satisfaction, improve product and service quality or gain new customers (Wang & Alam, 2007). Big data analytics is now emerging as a hot topic among scholars and is becoming a priority for organizations to implement. The term “big data” describes massive, complex, and real-time streamed data that require sophisticated management, processing and analytical techniques to gain insights from (Beyer & Laney, 2012). As a result of the worldwide diffusion and adoption of big data-enabling tools like mobile devices, social networks and identification technologies enabling the ‘Internet of Things’, unthinkable amounts of data are being distributed and collected every day. Since more data means more knowledge, organizations are increasingly making use of these tools to achieve and sustain competitive advantage (Fosso Wamba et al., 2017). According to academics, big data has the potential to completely transform the way organizations do business by allowing, among other things, enhanced transparency and improved performance measures (McAfee & Brynjolfsson, 2012). In recent literature big data analytics has already been identified as “the next big thing for innovation” (Gobble, 2013, p.64), “the next management revolution” (McAfee & Brynjolfsson, 2012) and “the fourth paradigm of science” (Strawn, 2012, p.34).
Yet, previous research on the business value derived from information systems (IS) investments has reported mixed results, resulting in the so-called ‘IT productivity paradox’: IT investments do not always seem to lead to better firm performance. Scholars have therefore identified an IT capability, which refers to the ability of an organization to efficiently collect, integrate and deploy its IT resources to create competitive advantage (Bharadwaj, 2000). Years
6
of research on IT capabilities and firm performance has proven that firms with superior IT capability generally achieve superior firm performance (Zhang, Zhao & Kumar, 2016). Following this stream of research, scholars are now working towards a big data analytics capability, as empirical evidence for a link between big data analytics investments and firm performance is still limited (Akter, Wamba, Gunasekaran, Dubey & Childe, 2016).
The concept of a of big data analytics capability highlights the importance of creating business insights using data management, technology and human capability to transform business into a competitive force (Kiron, Prentice & Ferguson, 2014). Several scholars have stressed the importance of these capabilities in order to enhance firm performance (McAfee & Brynjolfsson, 2012; Barton & Court, 2012; Kiron et al., 2014) and emerging empirical studies on this topic increasingly confirm a positive relationship between big data analytics capability and firm performance (Gupta & George, 2016; Akter et al., 2016; Wamba et al., 2017). However, little research has been done on firm-specific or situational factors that could influence this relationship. Additional research on this topic is therefore important for managers to realize the opportunities of turning a big data analytics capability into competitive advantage.
Prior research has shown that the relationship between big data analytics capabilities and firm performance is influenced by dynamic capabilities (Fosso Wamba et al., 2017) and organizational agility (Côrte-Real et al., 2017). Both terms refer to the capability of firms to respond to rapidly changing markets. Over the past two decades these topics have gained attention from scholars in explaining firm performance, because they address the uncertain and ever-changing business environment firms nowadays operate in (Tseng & Lin, 2011). A theory that also addresses this ability is innovation ambidexterity, which can be seen as a form of dynamic capability (O’Reilly & Tushman, 2007). Innovation ambidexterity highlights the importance of both exploratory and exploitative innovation to an organization’s success. It is the capability of an organization to exploit existing knowledge to satisfy current demands while
7
simultaneously exploring unknown areas to adapt to changes in the environment (Gibson & Birkinshaw, 2004; Tushman & O’Reilly, 1996). While empiric research on this topic has showed mixed results, researchers generally agree that that pursuing an ambidextrous orientation is highly desirable as it helps balance the short and long-term needs of the company (O’Reilly & Tushman, 2007).
In recent years the role of ambidexterity in relation to IT capabilities has also been gaining attention from scholars. Several studies have pointed out that IT capabilities provide the foundation for firms to explore new knowledge and exploit existing knowledge, which in turn may lead to higher innovation performance (Benitez, Castillo, Llorens & Braojos, 2018) or higher firm performance (Benitez, Llorens, Braojos, 2018). The existing literature on big data analytics capability builds on IT capability theories. However, the relationship between big data analytics capability, innovation ambidexterity and firm performance has not yet been investigated. To address this gap in the literature, the following research question is examined:
RQ: To what extent does innovation ambidexterity affect the relationship between big data analytics capability and firm performance?
The study is structured as follows: The second chapter contains the theoretical foundation for the study, where hypotheses are formed based on the evaluation of existing relevant literature. The third chapter consists of the methodology part, where the research design, sample, data collection method and measurements of the constructs are discussed. The fourth chapter provides the analysis of the data and the corresponding results, as well as answers to the hypotheses. The fifth chapter contains the conclusion and discussion of the results of the study. The sixth chapter consists of the references, and chapter seven presents the appendices.
8
2. Literature review
This section provides the theoretical background for the study, leading to the development of hypotheses related to the research question.
2.1 Resource-based view and IT capability
The resource-based view (RBV) considers an organization as a collection of resources, and it has been the primary theoretical foundation for explaining how the resources of a firm can create a sustained competitive advantage (Barney, 1991; 2001). Several scholars have argued the relevance of RBV in explaining the effect of IT resources on firm performance (Bharadwaj, 2000; Melville, Kraemer, & Gurbaxani, 2004; Doherty & Terry, 2009; Lioukas, Reuer & Zollo, 2016) The RBV logic is based on two assumptions: that of resource heterogeneity and resource immobility. Resource heterogeneity indicates that all firms possess resources which differ from other firms, and according to the assumption of resource immobility those resources cannot move from one firm to another. In addition to these assumptions, the RBV assesses competitive advantage based on the VRIN criteria: it argues that a firm can achieve competitive advantage by possessing tangible and intangible resources that are valuable (V), rare (R), inimitable (I) and non-substitutable (N). These criteria later became the VRIO criteria, where ‘non-substitutable resources’ was replaced by ‘organization’, referring to the way the resources are organized. The first three aspects apply to the characteristics of the resources, while the organization dimension focuses on the management of these valuable, rare and imperfectly imitable resources to leverage their full potential (Barney & Clark, 2007). Thus, the two most important components of RBV are resources and capabilities. RBV argues that a competent firm is one that has the capabilities to effectively manage its critical resources to achieve firm performance (Grant, 2002).
9
Building on the RBV, various studies in the information systems (IS) field have argued that different types of IT resources (e.g. physical, technical or human) can add value to firms’ operations. However, RBV is often criticized because of its static nature and its lack of explanatory power on how certain resources are organized, how they can create unique and distinctive capabilities and how they ultimately lead to competitive advantage. It is therefore that several scholars have argued to take a broader view of IT resources by integrating IT capabilities (Bharadwaj, 2000; Santhanam & Hartono, 2003; Bhatt & Grover, 2005; Doherty & Terry, 2009; Lioukas et al., 2016). IT capability is defined as “a firm’s ability to mobilize and deploy IT-based resources in combination or co-present with other resources and capabilities” (Bharadwaj, 2000). While the RBV focuses on both resources and capabilities, these two elements differ from each other in the sense that capabilities are more difficult to replicate than resources, as they are more deeply rooted in the culture of the firm (Bharadwaj et al., 1999). Hence, not the technology and IT resources themselves, but the associated capabilities to mobilize and deploy them provide firms with a sustainable competitive advantage. It is therefore important for firms to possess an IT capability, rather than just owning IT resources which might be easily imitable by competitors.
Bharadwaj (2000) employed the RBV to develop the concept of IT as an organizational capability that is created by the interaction of IT infrastructure (tangible IT resources), human IT resources (technical and managerial IT skills) and IT-enabled resources (intangibles such as knowledge assets, customer orientation and synergy). The IT infrastructure comprises physical assets like computer and communication technologies and technical platforms and databases. From an RBV point of view it is unlikely that physical resources will be sources of competitive advantage, due to the fact that they can be duplicated relatively easy by competitors (Mata, Fuerst & Barney, 1995). However, RBV does not take into account the synergistic benefits of integrated systems. In addition, such infrastructures evolve over time and in a way that make
10
their value difficult to imitate. The human IT resources comprise the assets employees possess related to training, experience, relationships and insights. These competencies lead to either technical or managerial IT skills, which also evolve over time due to experience. Specifically managerial skills are often tacit as they depend on personal relationships which may take years to develop, and are therefore difficult to acquire or imitate. IT-enabled intangibles comprise tacit, deeply-rooted assets like know-how (Teece, 1998), corporate culture (Barney, 1991) and environmental orientation (Russo & Fouts, 1997). But also intangible benefits of IT such as improved customer service, enhanced product quality and faster responsiveness to markets. The RBV already recognized the importance of intangible organizational resources, which are key to achieving competitive advantage due to their inimitability and non-substitutability.
The IT productivity paradox
Several studies have reported a positive relationship between a firm’s IT capabilities and firm performance, both directly and indirectly. For example Bharadwaj (2000) and Santhanam and Hartono (2003) found a direct effect of IT capability on financial firm performance. Lioukas et al. (2016) found that firms with better IT capabilities can derive greater value from alliances. Tippins and Sohi (2003) found an indirect effect of IT capability on firm performance through organizational learning, and Kim et al. (2011) found an indirect effect of IT capability on firm performance through process-oriented dynamic capabilities. Yet, while some existing studies have established positive effects of IT capability on firm performance, more recent studies that have investigated the impact of IT capability on firms’ competitive advantage report mixed results. Wang and Alam (2007), for example, found that IT capability provides value to firms beyond traditional accounting information. Muhanna and Stoel (2010) studied two archival data sets representing the pre-internet (1992-1994) and post-internet (1999-2006) eras to examine the effects of IT capability on market value and accounting performance. They found support for both value creation and positive future earnings. Masli, Richardson, Sanchez and Smith
11
(2011), however, examined the structural shifts in the returns generated from IT capability over a time period of 1988-2007, and found that firms with superior IT capability reported higher firm performance only until 1999. After 1999, that advantage was no longer observed. This inconsistency in research results relates to the IT productivity paradox, which refers to the failure of organizations to prove a significant positive relationship between IT investments and firm productivity (Bhatt & Grover, 2005). Due to the missing transparency of the costs and the difficulty of linking firm performance back to activities, IT value has been characterized as unquantifiable and uncertain (Ross, Beath & Goodhue, 1996). The conflicting findings in IT research suggest that there may be several factors that influence the relationship between IT capabilities and firm performance.
Due to the rise of new kinds of information technologies, such as big data analytics, cloud computing and mobile device technology, IT investments are regaining a prominent position on the agenda of managers. IT capability is becoming critical for firms to have, in order to achieve positive firm outcomes such as product innovation (Wang et al., 2013), enhanced financial performance (Kim et al., 2011; Masli et al., 2011) and sustainable competitive advantage (Mata et al., 1995; Ross et al., 1996). Where IT was considered an important competitive weapon in the 1980s, it is now big data that is expected to be the next big thing for organizations to achieve a competitive position (Gupta & George, 2016). Given the speed at which organizations in all industries are jumping on the big data bandwagon, Gupta & George (2016) argue that it is likely that we are going towards a “big data productivity paradox” as well. According to Ross, Beath, & Quaadgras (2013), the majority of the big data investments fail to pay off because most companies are either not ready or do not make decisions based on the intelligence extracted from data. It is therefore necessary to create a big data analytics capability, as previous research has shown that firms achieve competitive advantage by building capabilities.
12 2.2 Big data
The term “big data” was initially created to describe the enormous size of data generated as a result of using new forms of technology like social media, radio-frequency identification (RFID) tags, smart phones, and sensors (Gupta & George, 2016). Over the years, scholars have come up with more dimensions of big data.The term is now mainly defined by the notion of the 4 Vs: ‘Volume’ referring to the massive amounts of data that are being generated; ‘Velocity’, which is about the frequency or the speed at which data are created; ‘Variety’, which relates to the fact that data are generated from a wide variety of sources which contain multidimensional data fields with structured and unstructured data; and ‘Value’, which highlights the importance of deriving economic benefits from the available big data (Fosso Wamba et al, 2015). Later, a fifth dimension – ‘veracity’ – was added (White, 2012), which refers to the importance of good quality data and trust in several data sources. While these characteristics have contributed to our understanding of big data, the Vs mostly emphasize the technical aspect and leave out the importance of several other resources, like e.g. human skills and organizational culture, that are equally important to realize the benefits of big data (Gupta & George, 2016).
In their definitions several scholars focus on the analysis aspect of big data, instead of the data itself. Fosso Wamba et al. (2015) found in their study on the impact of big data that the ability to analyze big data enables organizations to create business value in several ways: by increasing transparency, enabling experimentation, segmenting and adapting to populations, innovating new business models and supporting or even replacing human decision making. Many other studies on big data also emphasize the importance of storage and analysis to deal with big data (Jacobs, 2009; Manyika et al., 2011, Havens et al., 2012). Manyika et al. (2011) define big data as “datasets with a size that is beyond the ability of typical database software tools to capture, store, manage, and analyze.” The International Data Corporation (IDC)
13
identifies three characteristics of big data: the data itself, the analytics of the data, and the presentation of the results of analytics that allow business value creation. Fosso Wamba et al. (2015) argue that big data needs to be viewed more in terms of developing crucial skills to be able to store, analyze and generate valuable insights from big data to realize competitive advantage through co-creation and realization. They define big data as “managing, processing and analyzing the 5 V’s in order to create actionable insights for sustained value delivery, measuring performance and establishing competitive advantages”. From a literature review and in-depth case study, they concluded that the big data revolution is evolving and organizations should embrace it in order to build critical capabilities for realizing competitive advantage.
Big data analytics (hereafter referred to as BDA), is about technologies (e.g. database and data mining tools) and techniques that an organization can exploit to analyze large scale, complex data (Kwon, Lee & Shin, 2014) to extract insights which are not obtainable using past data technologies (Garmaki et al., 2016). It is considered a game changer for firms, as it has high strategic and operational potential due to its ability to improve business efficiency and effectiveness (Fosso Wamba et al., 2017). While BDA is a relatively new form of IT, is expected to have massive impacts within several industries. For example, in healthcare BDA is expected to reduce operational costs and improve the quality of life. In retail, BDA can improve customer service, reduce fraud, and make personalized recommendations. In manufacturing and operations management, BDA can enable supply chain visibility, enhance manufacturing and industrial automation and enhance business transformation (Fosso Wamba et al., 2017). Some emerging literature on this topic has already provided empirical evidence for a link between BDA and firm performance. Germann, Lilien, Fiedler and Kraus (2014) found a positive relationship between the deployment of customer analytics and firm performance. Gunasekaran et al. (2017) also found that big data and predictive analysis is positively related to supply chain and organizational performance. So as BDA is increasingly becoming a crucial component of
14
decision-making processes in organizations (Hagel, 2015), several scholars have aimed to create a BDA capability to explain firm performance (Gupta & George, 2016; Fosso Wamba et al., 2017; Wang & Hajli, 2017; Côrte-Real, Oliveira & Ruivo, 2017).
2.3 Big data analytics capability
Big data analytics capability (hereafter referred to as BDA capability) is defined as the competence to provide business insights using data management, infrastructure (technology) and talent (employees) capability to transform business into a competitive force (Kiron et al., 2014). In recent years, several researchers have focused their attention on investigating and understanding these capabilities. Wamba et al. (2017) conducted a research on BDA capabilities similar to the IT capabilities literature, from which they derived three dominant dimensions: management, infrastructure and personnel capabilities. They point out that management capability is important for optimizing decision models; infrastructure (technology) capability is essential to explore and manage a variety of data; and data science capability (personnel) is important to understand, develop and apply analytics models. These three components are in turn explained by eleven constructs: BDA planning, investment, coordination, control, connectivity, compatibility, modularity, technical knowledge, technology management knowledge, business knowledge and relational knowledge. They thus propose BDAC as a third-order, hierarchical model, where infrastructure, personnel and management capability are key components.
Similar to the model BDA capability model created by Fosso Wamba et al. (2017) and the IT capability model created by Bharadwaj (2000), Gupta and George (2016) draw on the RBV to propose a higher-order formative construct to conceptualize BDA capability. They propose three types of resources, which are comprised of seven more specific resources that enable firms to create a BDA capability. Following the resource classification of Grant (2010), they distinguish between tangible resources (data, technology and other basic resources),
15
human resources (managerial and technical big data skills) and intangible resources (data-driven culture and organizational learning). In this study the BDA capability construct by Gupta and George (2016) is adopted to test the research question and hypotheses. The choice for their model over that of Fosso Wamba et al. (2017) is driven by its inclusion of intangible resources, which have been emphasized as important assets by the RBV (Barney, 1991) and IT capability literature (Bharadwaj, 2000; Santhanam & Hartono, 2003). All resources forming the construct are hereafter discussed in more detail.
Tangible resources
Referring to the RBV, Gupta and George (2016) describe tangible resources are resources that can be sold or bought in a market, for example financial resources such as equity or physical resources such as equipment. As RBV argues that resources need to be inimitable and non-substitutable, these resources on their own are unlikely to provide a competitive advantage. Yet these resources are required to create capabilities. The tangible resources comprise data, technology and basic resources.
A firm’s data can be categorized into internal data and external data. Internal data are firm-specific and are created by the internal operations of a firm. Examples are inventory updates, accounting transaction sales and human resource management. External data are data collected from sources outside of the organization. Examples are the internet, e-commerce communities, mobile phones and sensors. As both types of data create little value by themselves, for firms to create a BDA capability they must integrate their internal and external data. Furthermore Gupta and George (2016) identify five sources of data: public data (government-owned), private data (firm-owned), data exhaust (data with no direct value), community data (generated by social networks) and self-quantification data (personal data from wearable technologies).
16
Advanced technologies are needed to handle these gigantic amounts of diverse and fast-moving data. Organizations often use relational database management systems (RDBMS) to store and analyze structured data such as employee records, customer orders and financial transactions. However, this method is only applies to data that is structured. Gupta and George (2016) state that according to estimates, about 80% of an organization’s data is unstructured. Therefore organizations are adopting new, more advanced data storage and analysis methods such as Hadoop or SQL.
Basic resources refer to the investments made in big data and the time given to achieve its objectives. Given that big data and its related technologies and tasks are quite new, investments may not immediately generate desired results. It is therefore important that firms devote enough time and money to their BDA initiatives to achieve their objectives.
Human resources
The human resources of a firm comprise the experience, knowledge, abilities, business insights, leadership qualities and relationships of its employees. The two corresponding aspects are technical and managerial skills, corresponding with prior IT capability research.
Technical skills, or “big data” skills, refer to the ability to extract intelligence from big data by using new technologies. These skills include competencies in machine learning, data extraction, statistical analysis and understanding programming paradigms. Technical IT skills in general are not considered rare as they can be translated into procedures or manuals. However, given the newness of big data and the associated skills, organizations with data-skilled employees are likely to have some advantage over competitors.
Managerial skills on the other hand cannot be imitated but are developed within a firm. It is the result of strong interpersonal bonds between members of an organization, which are created over time. These skills are tacit and deeply rooted into an organization culture, and therefore hard to imitate or substitute.
17 Intangible resources
The RBV already emphasized the importance of intangible resources, as they mostly meet the VRIN criteria earlier discussed (Barney, 1991). Intangible resources do not have clear boundaries and their value is context-dependent, making them highly heterogenous across firms. Especially in dynamic markets these resources are important. Gupta and George (2016) suggest two intangible resources that are likely to be valuable for firms investing in BDA: data-driven decision making culture and intensity of organizational learning.
Data-driven culture is defined as “the extent to which organizational members (including top-level executives, middle managers, and lower-level employees) make decisions based on the insights extracted from data”. Important here is that employees at all levels in an organization are involved in data-related decision-making. Previous studies have shown that organizational culture is a source of sustained firm performance, and also recent studies on big data suggest that organizational culture is critical for the success of big data initiatives (Ross, Beath & Quaadgras, 2013; LaValle, Lesser, Shockley, Hopkins & Kruschwitz, 2014).
Organizational learning, as defined by Grant (1996), is a process through which firms explore, store, share and apply knowledge. Extending the RBV, he proposed the knowledge-based view, which considers the specialized knowledge of individuals the most important strategic resource of a firm. However, as knowledge might become outdated, firms need to make efforts in exploiting their existing knowledge and also exploring new knowledge to deal with uncertain market conditions. Gupta and George (2016) argue that firms with high levels of organizational learning will likely have an advantage of combining their knowledge with BDA-generated insights to make informed decisions.
BDA capability and firm performance
Gupta and George (2016) tested their model in relation to firm performance, and found that BDA capability positively influences both market performance and operational performance.
18
Fosso Wamba et al. (2017) also examined the relationship between their BDA capability model and firm performance, and found both direct and indirect effects. They found that BDA capability positively influences firm performance, and that process-oriented dynamic capabilities play a strong mediating role in this relationship. The same model was used by Akter et al. (2016), who found a positive relationship between BDA capability and firm performance, with a significantly moderating impact of analytics-capability-business strategy alignment. Furthermore, Côrte-Real et al. (2017) proposed a conceptual model to assess BDA value, based on a knowledge-based view and dynamic capability theories. They also found a positive effect of BDA on firm performance, with knowledge assets being antecedents and organizational agility a mediator of the relationship. In addition, Wang and Hajli (2017) examined the influence of BDA capability in healthcare organizations and also found that it leads to several paths of value creation. In line with these findings the following hypothesis is tested:
H1: BDA capability has a positive effect on firm performance.
There have thus been various studies proving a positive relationship between BDA capability and firm performance, with several mediating and moderating variables. Many of these studies address terms like dynamic capabilities (Fosso Wamba et al., 2017; Côrte-Real, 2017) and agility (Côrte-Real, 2017), which both refer to the ability of firms to identify and respond to environmental threats and opportunities and quickly adjust their behavior. These terms are adopted into BDA capability research, as prior research on IT capability and firm performance has already highlighted the important role of several types of dynamic capabilities. One of these types of dynamic capabilities is discussed in the next chapter.
2.4 Innovation ambidexterity
Organizations often operate in uncertain, complex, and ambiguous environments, where managers have to deal with contradictory decision choices. They have to aim for operational effectiveness to deliver growth, while simultaneously focus on innovation to adapt to the rapidly
19
changing business environment (March, 1991). In order to resolve this issue, an organization is required to become “ambidextrous”, that is, be able to simultaneously pursue exploration and exploitation, efficiency and flexibility or alignment and adaptability (De Clercq, Thongpapanl & Dimov, 2013). Innovation is described as a complex and dynamic process through which firms develop capabilities by exploring new resources or exploiting new combinations of resources (Zhang, Edgar, Geare & O’Kane, 2016). Building on these perspectives, innovation ambidexterity is characterized as the ability of a firm to develop explorative and exploitative capabilities for both radical and incremental innovation (Tushman & O’Reilly, 1996; He & Wong, 2004; Lin et al., 2013). Exploratory innovation aims at pursuing new knowledge, breakthrough ideas and potential opportunities satisfying emerging customers and markets. Exploitative innovation involves activities aimed at extending or improving existing knowledge and skills, focusing on current customers and markets (Zang & Li, 2017). The simultaneous investment in exploration and exploitation is supported by various scholars as exploration and exploitation are self-reinforcing (Levinthal & March, 1993), pursuing them simultaneously yields synergies (He & Wong, 2004), and it leads to long-term adaption to new developments (exploration) and short-term alignment with existing markets (exploitation) (Gibson & Birkinshaw, 2004)
He and Wong (2004) argue that exploration and exploitation create tensions as they are fundamentally different orientations that compete for a firms’ scarce resources. Both orientations require substantially different structures, processes, strategies, capabilities, and cultures and may have different impacts on firm adaptation and performance. Firms therefore need to know how to manage the balance between the two. They test the ambidexterity hypothesis in the context of technological innovation. They extend the exploration versus exploitation construct and define a new classification of technological innovation strategy along two generic dimensions: (1) an explorative innovation dimension which refers to technological
20
innovation activities aimed at entering new product-markets and (2) an exploitative innovation dimension which refers to technological activities aimed at improving existing product-market positions. Considering that their construct of innovation ambidexterity relates specifically to technological innovation and has been adopted by several academics in ambidexterity research (Morgan & Berthon, 2008; Álvarez Santos, Miguel-Dávila & Nieto Antolín, 2016), their measures are used in this study. The construct will be discussed in more detail in the methodology section of this study.
Innovation ambidexterity and firm performance
Ambidexterity theory suggests that organizations capable of exploiting existing businesses while simultaneously exploring new opportunities may achieve superior performance compared to firms emphasizing one at the expense of the other (Bøe-Lillegraven, 2014). Too much focus on exploitation of current competencies will lead to a “success trap” – organizational inertia that reduces firms’ ability to adapt to changing environments – which will in the long run cause poor performance outcomes. On the other hand, too much focus on exploration will lead to a “failure trap”, where innovations are replaced by new ideas too quickly to contribute to a firm’s revenue (Levinthal & March, 1993). Various empirical studies have provided evidence for a positive association between ambidexterity and different types of firm performance (Gibson & Birkinshaw, 2004; He & Wong, 2004; Lubatkin et al., 2006; Junni, Sarala, Taras & Tarba, 2013, De Clercq et al., 2013; Soto-Acosta & Martinez-Conesa, 2018). He and Wong (2004), for example, showed that the interaction between explorative and exploitative innovation strategies is positively related to sales growth rate, and that the relative imbalance between the two types of strategies is negatively related to sales growth rate – both supporting the assumption that firms with high levels of innovation ambidexterity perform better than firms that emphasize one over the other. Gibson and Birkinshaw (2004) identified a positive relationship between ambidexterity and business unit performance as perceived by senior and middle management,
21
and Lubatkin et al. (2006) found that firms with higher levels of ambidexterity achieve higher levels of profitability and growth, relative to that of industry competitors.
While the general agreement in literature is that concurrently exploiting existing competencies and exploring new opportunities leads to ambidexterity, and ambidexterity enables a firm to enhance its performance and competitiveness, studies on the ambidexterity-performance relationship have generated mixed results. Some studies have found no or a negative association (Athuahene-Gima, 2005), while others have found a contingent effect (Lin, Yang & Demirkan, 2007). Junni et al. (2013) systematically examined the ambidexterity-performance relationship through a meta-analysis of prior studies, and concluded that ambidexterity was overall positively and significantly related with performance. However, the positive and significant organizational ambidexterity-performance relationships were to a large extent moderated by contextual factors or methodological settings (Junni et al., 2013). For instance, they found that studies based on cross-sectional surveys reported stronger results than archival studies, and that subjective performance measures provide stronger results than objective performance measures. A possible explanation for this could be that cross-sectional studies are affected by common method bias. Also, they observed that exploration was mainly related to increased growth, whereas exploitation was more related to firm profits. Furthermore, regarding the industry context, they found that organizational ambidexterity is less important in manufacturing firms than in service and high-technology firms. They argue that this might be due to the higher level of environmental dynamism in knowledge-intensive service firms and high-technology industries. To further examine this relationship, the following hypothesis is proposed:
22 Innovation ambidexterity as dynamic capability
Academics have also characterized ambidexterity as a form of dynamic capability (Eisenhardt & Martin, 2000; He & Wong, 2004; O’Reilly & Tushman, 2007; Jansen, Tempelaar & Van den Bosch, 2009; Kriz et al., 2014; Zimmermann, Raisch & Birkinshaw, 2015). Dynamic capabilities refer to a firm’s ability to “integrate, build, and reconfigure internal and external competences to address rapidly changing environments” (Teece et al., 1997, p.516). Researchers on innovation ambidexterity have pointed out the similarities with dynamic capability theory. Eisenhardt and Martin (2000) state that overall, dynamic capabilities require a mix of the logic of exploration and the logic of exploitation. O’Reilly and Tushman (2007) argue a set of senior team actions, processes and design choices that comprise a set of dynamic capabilities, which enable firms to reconfigure existing resources and learn new capabilities to both explore and exploit. They use the taxonomy of sensing (sensing opportunities and threats) seizing (respond to what is being observed) and reconfiguring (need for leaders who reallocate resources toward emerging growth opportunities) to explain innovation ambidexterity. Zimmermann et al. (2015) state that innovation ambidexterity is a mechanism to effectively manage the shift between explorative and exploitative activities, consistent with the dynamic capability viewpoint of integrating and reconfiguring competences. In line with this, Jansen et al. (2009) conceptualize innovation ambidexterity as an organizational-level dynamic capability which emphasizes the importance of structural differentiation and integration.
Innovation ambidexterity and IT capability
As mentioned before, dynamic capabilities have been subject of several studies on IT capabilities and firm performance. For example, Chen et al. (2014) showed that the impact of IT capability on firm performance is fully mediated by business process agility, Lu and Ramamurthy (2011) found significant positive effects of IT capability on market capitalizing agility and operational adjustment agility, and Kim et al (2011) found a mediating role of
23
process-oriented dynamic capabilities on the relationship between IT capability and financial firm performance. Hence, given that dynamic capabilities have been proven to mediate the relationship of IT capabilities on firm performance, as well as the relationship of BDA capability on firm performance (Fosso Wamba, 2017; Côrte-Real, 2017), and given that innovation ambidexterity is a form of dynamic capability, it is plausible to assume that innovation ambidexterity might also play a mediating role in the relationship between BDA capability and firm performance.
Supporting this assumption, recent research has also identified a positive relationship between IT capability, innovation ambidexterity and firm performance. For example, Zang and Li (2017) showed that technology capabilities and marketing capabilities can complementarily improve innovation ambidexterity, which in turn further enhances organizational performance. Soto-Acosta and Martinez-Conesa (2018) found that IT capability, knowledge management capability and environmental dynamism are antecedents of innovation ambidexterity, which further leads to better firm performance. Benitez et al. (2018a) argue that IT plays a key role in both a firms’ opportunity exploration and opportunity exploitation. They specified an IT infrastructure capability as a second-order composite construct determined by technological, managerial and technical capabilities. Their results showed that IT infrastructure capability influences opportunity exploration through business experimentation and business flexibility, and IT-enabled business flexibility helps firm to develop the operational proficiency to exploit opportunities and enhance their performance. In addition, IT infrastructure capability enables a firm to both explore new knowledge and exploit existing/new knowledge (knowledge ambidexterity), which enhances innovation performance (Benitez et al., 2018b). These results were further supported by Soto-Acosta and Martinez-Conesa (2018), who investigated the effect of technological, organizational and environmental factors on innovation ambidexterity, and its subsequent influence on firm performance. The results showed that IT capability is
24
positively associated with innovation ambidexterity, which in turn is positively associated with firm performance. In line with this, the following relationship is expected:
H2b: BDA capability has a positive effect on innovation ambidexterity.
In short, various studies have shown empirical support for the mediating role of dynamic capabilities, including that of innovation ambidexterity, in the relationship between IT capability and firm performance. Given that BDA capability literature builds on IT capability literature and the two types of capabilities are strongly connected, it can be expected that this mediating role also applies to BDA research. Hence, the following hypothesis is proposed:
H3: Innovation ambidexterity mediates the relationship between BDA capability and firm performance.
25
3. Data and method
In this section the methodology of the study is explained. The research design, conceptual model, sample, data collection process and the individual measurements of all variables are discussed.
3.1 Research design
For this explanatory study a quantitative, cross-sectional survey design was used to collect the required data. A quantitative approach was chosen because it is best suited to answer research questions that deal with dependencies and causal relations, as is the case in this study. The questionnaire was prepared and distributed in collaboration with four fellow students investigating the same topic. Online survey software tool Qualtrics was used to compose the questionnaire, which ultimately consisted of a total of 70 questions. The questionnaire as constituted by Gupta and George (2016) was adopted to measure BDA capability and firm performance, and the personal constructs along with some demographic and control questions relevant for this research were added. The questionnaire was subsequently distributed among big data analytics managers by e-mail. All questions are presented in appendix 1.
3.2 Conceptual model
The conceptual model, depicted in figure 1, shows the hypothesized connections between the independent variable (BDA capability), the dependent variable (firm performance) and the mediating variable (innovation ambidexterity). As the model shows, both a direct effect and indirect effect through innovation ambidexterity of BDA capability on firm performance are expected.
26
Figure 1. Conceptual model
3.3 Sample
The population relevant for this study were big data analytics managers. As that population was too large to include every individual, a non-probability purposive sampling technique was used to target respondents. With this type of sampling respondents are selected based on the judgement of the researcher. To reduce the size of the population, only firms located in Europe were targeted. The participants were thus BDA managers from European firms in a variety of industries. In total 1483 e-mails were sent with a request to participate in the survey. Out of this sample, 167 responses were collected, representing a response rate of 11%. This data was cleaned and prepared using the Statistical Package for Social Sciences Program (SPSS). After deleting all irrelevant or incomplete responses and excluding missing values, a total of 62 useful responses remained. The characteristics of the survey sample are presented in table 1.
Innovation Ambidexterity H2b +a H2a H1 Firm Performance Big Data Analytics Capability + + + H3
27 3.4 Data collection
Collecting the data was done over a period of three weeks, where respondents were targeted through LinkedIn and with help of online data-search tool RocketReach. BDA managers were searched by using the queries ‘Big Data Analytics Manager’ and ‘Chief Technology Officer’ on LinkedIn, and filtering on location to reach our specific target group. The countries that were included in the sample were The Netherlands, Belgium, France, Spain, Germany, Ireland,
Table 1
Sample characteristics
Frequency Percent Years invested in BDA
Less than 3 years 17 27,4
3-6 years 26 41,9
More than 6 years 19 30,6
Number of employees Fewer than 100 16 25,8 Between 101 and 250 5 8,1 Between 251 and 500 7 11,3 Between 501 and 1000 3 4,8 Between 1001 and 2500 4 6,5 Between 2501 and 5000 1 1,6 More than 5000 26 41,9 Industry Computer/Software 15 24,2 Manufacturing 6 9,7
Finance, Insurance, Real Estate 11 17,7
Retail, Wholesale 1 1,6
Services 7 11,3
Healthcare 4 6,5
28
Switzerland, Norway, Denmark and Finland. Furthermore, London was targeted separately, since the United Kingdom generated too many results (> 1000) to analyze. The BDA managers were then manually selected based on the relevance of their function, and imported into RocketReach. In RocketReach the business e-mail addresses of the big data analytics managers were collected, which were then merged using an e-mail merging tool. Subsequently, an e-mail with the request to participate in the survey was sent to all collected addresses. Access to the results of the research project was offered as an incentive to participate. To increase the response rate, a maximum of three reminders were sent to the respondents who had not completed the survey.
3.5 Measures
All core constructs of this study were measured using multi-item scales that were validated in prior studies, increasing reliability and validity of the measures. They are presented in more detail in appendix 1.
Big data analytics capability (BDA capability) is a multidimensional third-order formative construct adopted from Gupta and George (2016), who theoretically substantiated and validated this model in their study. It is formed by three second-order resource categories (tangible, human and intangible resources), which are in turn formed by seven specific first-order resource types (data, technology, basic resources, managerial skills, technical skills, data-driven culture and organizational learning) measured by 32 underlying indicators. The construct thus in total consisted of 32 survey questions, formulated as “to what extend do you agree or disagree with the following statements”. The questions all referred to big data-related activities and were answered on a 7-point Likert scale (1 = strongly disagree to 7 = strongly agree).
Innovation ambidexterity was measured by adopting the scales that He and Wong (2004) used in their study. The construct consisted of eight 5-point Likert scale items (1 = not important
29
to 5 = very important) to measure how firms divide attention and resources between innovation activities with explorative versus exploitative objectives in the last three years. This was done by means of the question ‘what were the firms’ objectives for undertaking innovation projects in the last 3 years?’. Four statements comprised the explorative innovation strategy (e.g. ‘introduce new generation of products’ or ‘enter new technology fields’) and four statements comprised the exploitative innovation strategy (e.g. ‘improve existing product quality’ or ‘improve production flexibility’). In their study, He and Wong (2004) showed sufficient reliability for the scales: the scale for explorative innovation strategy had a Cronbach’s alpha of 0.75 and the scale of exploitative innovation strategy of 0.80. These two separate scales of exploration and exploitation strategy together form the construct of innovation ambidexterity. In accordance with this, a reliability analysis was performed in SPSS. The exploration strategy scale had sufficient reliability with a Cronbach’s Alpha of 0.76, however, the scale for exploitation strategy was less reliable with a Cronbach’s Alpha of 0.58. This value would not sufficiently increase if an item were deleted, so the scales were not changed. To include this variable into further analysis, the two separate scales had to be merged into one measure of innovation ambidexterity. While the operationalization of ambidexterity varies among existing literature, more support is found for combined measures than for balanced measures (Junni et al., 2013). Since there seems to be most support for multiplying the scales of exploration and exploitation (He & Wong, 2004; Gibson and Birkinshaw, 2004), this method is applied in this study.
Firm performance was divided into two dimensions: market performance and operational performance, using the measures proposed by Gupta and George (2016). They define firm performance as the extent to which a firm generates superior performance relative to its competitors. It is operationalized as an 8-item construct, divided into four measures representing market performance (e.g. ‘we have entered new markets more quickly than our
30
competitors’) and four measures representing operational performance (e.g. ‘our productivity has exceeded that of our competitors’). They were included in the analysis as first-order reflective constructs. Market performance and operational performance were assessed separately as they measure different aspects of the construct.
Control variables were added to avoid any bias due to demographics. The variables firm size, industry and BDA age were recoded into dummy variables. Firm size was divided into small and medium sized enterprises (1-249 persons employed) and large enterprises (250+ persons employed) according to EU standards. As previous research has found that firms operating in an information-technology industry are able to extract substantially more value from BDA assets than firms in other industries (Müller et al., 2018), the variable industry was divided into the computer/software industry vs other industries, where all other industry options were merged into one category. Finally, BDA age was divided into less than 6 years of BDA investment and more than 6 years of BDA investment.
3.6 Common method bias
In research, respondents often tend to answer in a socially desirable way, which can result in a measurement error and generate distorted results. This is referred to as common method bias (Podsakoff, Mackenzie, Lee & Podsakoff, 2003). To prevent the occurrence of this, some measures were taken. First, the respondent was given the option to answer anonymously, which takes away the feeling of being judged. Second, to increase reliability, only constructs were used that were already tested and validated in prior research. Third, the items of the dependent and independent variables were randomly positioned in the questionnaire, which makes it more difficult for respondents to sense an underlying link between items.
31
4. Analysis and results
In this chapter the hypotheses are tested by analysing the data using variance-based structural equation modelling, more specifically the partial least squares path modelling method (PLS-SEM). PLS-SEM is suitable for research with a prediction-oriented objective, and under circumstances where theory is less developed and the structural model is complex (Hair Jr et al., 2013, pp.14,19). In addition, this method is suited for small sample sizes, as is the case in this study. To perform the analyses, SmartPLS 3.0 was used.
4.1 Descriptive statistics
Table 2 presents the descriptive statistics of the variables innovation ambidexterity and firm performance. The lowest recorded value, highest recorded value, mean, standard deviation of the constructs are included in the table. The results report higher average response values for exploration than for exploitation, and slightly higher average response values for market performance than for operational performance.
Table 2
Descriptive statistics of innovation ambidexterity and firm performance
N Min Max Mean SD
Innovation ambidexterity Exploration 62 1.00 5.00 3.64 0.92 Exploitation 62 1.00 5.00 3.44 0.76 Firm performance Market performance 62 2.00 7.00 4.61 1.14 Operational performance 62 2.00 7.00 4.52 1.13
32 4.2 The BDA capability model assessment
The BDA capability model was created in SmartPLS. Following the guidelines of hierarchical modelling (Becker et al., 2010; Chin, 2010), the same number of indicators were used to estimate the scores of the first-, second- and third-order constructs. First of all the 32 indicators were connected to their corresponding first-order latent variables (data, technology, basic resources, technical skills, managerial skills, data-driven culture or intensity of organizational learning). Then all indicators were again linked to the corresponding second-order construct (tangibles, human skills of intangibles) and after that they were linked to the third-order construct (big data analytics capability). All latent variables were subsequently connected with each other to create the full model. The model is depicted in figure 2.
Reliability and validity
To measure the validity of the BDA capability construct, first a confirmatory factor analysis was performed on the 32 items that measure the construct. The analysis provides outer loadings values for all the reflective constructs in the model, which represent the correlations of each indicator variable with its corresponding construct. The outer loadings are preferred to be at least 0.7, but in studies that are exploratory in nature, a score of 0.4 or higher is also acceptable (Hulland, 1999). In addition, the model has been validated in previous research by Gupta & George (2016) and due to some limitations of this study, which will be discussed in chapter 5, the factor analysis results are interpreted less strict than usual. The factor analysis showed that all outer loadings were above 0.4 and therefore all indicators of the reflective constructs were kept in the model. After this, non-parametric bootstrapping was applied to test the significance of the outer loadings. All values were significant with a p value of < 0.000.
33
Figure 2. Hierarchical Model Specification using Repeated Indicators Approach
Hereafter, the reliability and convergent and discriminant validity of the reflective constructs (managerial skills, technical skills, data-driven culture and organization learning) were examined. Reliability of the constructs was assessed using the composite reliability (CR) measure and Cronbach’s Alpha, which both measure internal consistency. The convergent validity was assessed by means of Average Variance Extracted (AVE). EVA measures the
34
amount of variance that a construct captures from its indicators, in relation to the amount of variance due to measurement error (Akter et al., 2016). Discriminant validity ensures that a reflective construct has the strongest relationships with its own indicators (Hair Jr et al., 2013), and was assessed using Fornell and Larcker’s criterion and the Heterotrait-Monotrait Ratio (HTMT). A correlation matrix of the latent variables for the first-order constructs including reliability and validity is depicted in table 3.
As the results from the correlation matrix show, both CR and Cronbach’s Alpha are above 0.8 for all constructs, indicating good reliability of the measures. The recommended threshold for the AVE value is 0.5. Only the technical skills variable has an AVE value just below 0.5. All other variables exceed 0.5 and therefore convergent validity for those variables is proven. Fornell and Larcker’s criterion for measuring discriminant validity calculates the square root of the AVE’s of each latent variable. If the number is greater than its correlation with any other construct, then the variable can be perceived as a valid measure. As this is the case for all reflective constructs, discriminant validity is also proven. A new measure of assessing discriminant validity is the Heterotrait-Monotrait Ratio (HTMT), which is based on the average of the correlations of indicators across constructs relative to the average of the correlations of indicators within the same construct. A HTMT ratio below 0.85 represents sufficient discriminant validity (Henseler et al., 2014). After running the HTMT test on the reflective constructs in SmartPLS, all values were found to be below 0.85, providing further support for discriminate validity of the constructs.
35 Table 3
Inter-correlations, reliability and validity of the first-order constructs*
Construct CR α AVE 1 2 3 4 5 6 7 1. Data NA NA NA NA 2. Basic Resources NA NA NA 0.47 NA 3. Technology NA NA NA 0.58 0.48 NA 4. Managerial Skills 0.94 0.94 0.72 0.42 0.46 0.26 0.85 5. Technical Skills 0.84 0.85 0.48 0.37 0.38 0.50 0.57 0.70 6. Data-driven Culture 0.83 0.83 0.50 0.45 0.41 0.49 0.58 0.65 0.71 7. Organization Learning 0.92 0.92 0.69 0.59 0.47 0.64 0.49 0.54 0.63 0.83 *Square root of the AVEs on the diagonal
For the formative constructs in the model, the outer weights were checked after the bootstrapping test. Quite some outer weights turned out to be not significant (p > 0.05). However, as Gupta and George (2016) argue in their study, a nonsignificant indicator of a formative construct can be kept in a model as long as the researchers can justify the contribution of it. Since this study is built around the BDA capability construct as validated created by Gupta and George (2016), removing all nonsignificant indicators would change the nature of the construct and might rule out important underlying explanatory factors. Therefore it was decided to only remove the indicators which have p values over 0.85. Removing those indicators can be justified by the fact that p values that high indicate that the indicator statistically has no effect on the construct. This resulted in a total of four indicators being removed: indicator T1, TS3, MS1 and DD4. Then the formative constructs were checked for their multicollinearity, as high multicollinearity among the indicators of formative constructs is undesirable. A way to rule this out is to calculate the variance inflation factor (VIF). Following Gupta and George (2016), 3.3 is the cut-off value for the VIF. To test validity, Edwards’ adequacy coefficient (R2a) was
36
manually calculated (Edwards, 2001). Values above 0.50 represent a valid construct. Table 4 shows the formative construct validation.
Table 4
Formative construct validation
Construct Measures Weight Significance VIF R2a*
Data D1 0.50 p < 0.01 1.196 0.75 D2 0.50 p < 0.01 1.287 D3 0.31 ns 1.350 Technology T2 0.23 ns 1.462 0.49 T3 0.27 ns 1.623 T4 0.78 p < 0.001 1.358 T5 -0.05 ns 1.757 Basic Resources BR1 0.35 ns 1.644 0.34 BR2 0.74 p < 0.01 1.644 Tangibles Data 0.38 p < 0.001 1.637 0.66 Technology 0.45 p < 0.001 1.650 Basic Resources 0.37 p < 0.001 1.397
Human Managerial Skills 0.43 p < 0.01 1.422 0.78 Technical Skills 0.72 p < 0.001 1.422
Intangibles Data-driven culture 0.60 p < 0.01 1.750 0.90 Organization Learning 0.55 p < 0.01 1.750
BDA capability Tangibles 0.42 p < 0.001 2.086 0.79 Human Skills 0.33 p < 0.001 2.237
Intangibles 0.37 p < 0.001 2.690 * Edwards adequacy coefficient
The results show that most of the indicators have significant weights on their respective higher-order constructs. Five indicators of the first-order constructs have nonsignificant weights, but as argued before it was decided to keep these in the model. All VIF values were below 3.3, representing low multicollinearity among the indicators. Furthermore, most R2a
37
values were greater than the recommended values of 0.50 and therefore represent valid constructs. The technology and basic resources construct have an R2a value of below 0.50, and thus questionable validity. However, these low values can be explained by the nonsignificant indicators that were kept in the model in the interest of the overall technology construct, which likely generated a decrease in the validity values. Another way to assess discriminant validity is by checking if the formative constructs correlate less than perfectly (< 0.71) on other constructs, according to MacKenzie et al. (2011). Table 3 shows that this is the case for all formative constructs.
4.3 Hypothesis testing
After validating the BDA capability construct, the variables firm performance and innovation ambidexterity were added to the model and ran through the PLS algorithm and bootstrapping, this time using path analysis. Again the outer loadings and weights of the constructs were evaluated. The indicators of managerial skills and data-driven culture had outer loading values below 0.4, indicating low correlations with the corresponding constructs. All other indicators had outer loading values above 0.4, indicating sufficient correlations with the corresponding constructs. The outer weight values for the formative constructs were for a large part nonsignificant. However, given that in the previous part the model has already been validated and adapted after assessing the outer weights and loadings, the constructs were kept in the model. The inter-correlations of the first-order constructs of BDA capability, innovation ambidexterity, market performance and firm performance are depicted in table 5. All constructs indicate sufficient reliability with CR and Cronbach’s Alpha values above 0.75, and sufficient convergent validity with AVE values above 0.50. Furthermore, according to both Fornell and Larcker’s criterion and the Heterotrait-Monotrait Ratio (HTMT), discriminant validity was also proved for all constructs (square root of AVE is greater than its correlation with any other construct and HTMT are values below 0.85).
38 Table 5
Inter-correlations, reliability and validity of first-order constructs*
Construct CR α AVE 1 2 3 4 5 6 7 8 9 1. Data NA NA NA NA 2. BR NA NA NA 0.49 NA 3. Tech NA NA NA 0.58 0.48 NA 4. MS 0.95 0.93 0.79 0.40 0.45 0.24 0.89 5. TS 0.86 0.79 0.55 0.32 0.35 0.48 0.48 0.74 6. DD 0.86 0.78 0.60 0.44 0.38 0.46 0.48 0.57 0.78 7. OL 0.94 0.92 0.75 0.55 0.45 0.60 0.45 0.50 0.56 0.87 8. OP NA NA NA 0.19 0.17 0.09 0.37 0.19 0.45 0.32 NA 9. MP NA NA NA 0.30 0.63 0.29 0.47 0.35 0.44 0.51 0.64 NA 10. IA NA NA NA 0.27 0.17 0.20 0.00 0.05 0.26 0.21 0.28 0.25 *Square root of the AVEs on the diagonal
BR = Basic resources, Tech = Technology, MS = Managerial skills, TS = Technical skills, DD = Data-driven culture, OL = Organizational learning, OP = Operational performance, MP = Market performance, IA = Innovation ambidexterity
Next, the hierarchical model was evaluated by assessing the indicators’ path coefficients (weights) of the second- and third-order constructs. The path coefficients in the structural model are interpreted as standardized beta coefficients of an ordinary least squares (OLS) regression analysis (Hair et al., 2011): they express the average changes in the standard deviation of the dependent variable due to a one-unit change in the standard deviation of the explanatory variable. According to Hair et al. (2011), path coefficient weights below 0.10 indicate a small effect, values around 0.30 indicate a medium effect, and values above 0.50 indicate a large effect. Of the indicators for the tangibles construct, only basic resources had a significant weight (β = 0.59, p < 0.05), indicating that if the basic resources variable increases by one standard deviation, the tangibles variable increases by 0.59. The indicator managerial skills had a positive significant weight on the human skills construct (β = 0.76, p < 0.001), whereas the
39
effect of the technical skills indicator had a nonsignificant weight. On the intangibles construct, both the indicators organizational learning (β = 0.52, p < 0.01) and data-driven culture (β = 0.61, p < 0.001) had significant weights.
After this, the relationship between BDA capability and the two dimensions of firm performance (market performance and operational performance) was examined. A significant, positive effect was found of BDA capability on both market performance (β = 0.65, p < 0.001) and on operational performance (β = 0.52, p < 0.001). In other words, the BDA capability variable increases by one standard deviation, the market and operational performance variables increase by 0.65 and 0.52 respectively, indicating a slightly larger effect on market performance than on operational performance. Hypothesis 1 is therefore fully supported. The model explained 44.6% of the variance in market performance and 35.4% of the variance in operational performance. The relationships are depicted in figure 3.
Figure 3. Analysis results
Subsequently, the mediating effect of innovation ambidexterity on the relationship between BDA capability and firm performance was examined. No significant effects were found of BDA capability on innovation ambidexterity, nor of innovation ambidexterity on
BDAC Operational performance R² = 35.4% Market performance R² = 44.6%
