Individual ambidexterity: The role of Intellectual Capital,
Entrepreneurial Orientation and the firms’ value proposition in small
and medium sized Dutch professional service firms
Final version Master thesis
MSc. in Business Administration – Strategy Track Amsterdam Business school
Date of submission 17th of August 2018
Student Marc Hoogvliet / Student number 5673186
University of Amsterdam, Faculty of Economics and Business
Supervisor Dr. M.P. Tempelaar
Statement of originality
This document is written by Student Marc Hoogvliet 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
Introduction 1
The context of the Professional service firm 6 Individual ambidexterity 9
Value propositions 11 Theoretical model 16 Method 17
Data collection 17
Exploratory factor analysis 18
Measures 21 Individual ambidexterity. 21 Entrepreneurial orientation 23 Intellectual capital 25 Value propositions 27 Control variables 28
Further assessment of Measurements 30
Analysis and results 37 Discussion 49
Limitations and future research 49
Conclusions 51 Bibliography 52
ABSTRACT
This thesis responds to the numerous calls for scholars to conduct multi-level studies examining the antecedents and outcomes of ambidextrous behavior. Placed within the context of research on Professional service firms, consultancy and ambidexterity, it adds to extant studies by an analysis of a firm-level value proposition construct and the effect it has on the consultants’ ambidextrous behavior. Fitting the context, it also examines the effect of the consultants’ Intellectual Capital his/ her ambidextrous behavior, mediated by an individual level Entrepreneurial Orientation. Data is gathered through collaboration with a Dutch consultancy trade organization, which resulted in an effective sample size of 102 participants consisting of self-employed consultants and those working for firms or in collective groups. While results are mixed the study finds significant relations between several Intellectual Capital subscales and Entrepreneurial Orientation subscales, detailing how these relations work within the Dutch consultancy firm context. No significant relation was found between Intellectual Capital and levels of ambidexterity, the firms value proposition and levels of ambidexterity. The mediation effect was also not strong.
Introduction
“The test of a first-rate intelligence is the ability to hold two opposed ideas in the mind at the same time, and still retain the ability to function” (Fitzgerald, 1945). This ability to hold two
opposing ideas simultaneously can be applied to the subject of organizational ambidexterity; a subject that has become a central topic within research on organizational theory and related fields (Birkinshaw & Gupta, 2013; O'Reilly & Tushman, 2013; Smith, Binns, & Tushman, 2010). Our conceptualization of organizational ambidexterity stems from March’s seminal article, where he argued that ‘maintaining an appropriate balance between exploration and exploitation is a primary factor in system survival and prosperity’ (1991, p.71). While exploration allows firms to search for new market opportunities and the creation of new knowledge, exploitation is needed to be able to earn rents from the existing market opportunities and capabilities (Levinthal & March, 1993). Authors have shown that by being able to pursue both exploration and exploitation, firms are able to earn higher profits and improve customer satisfaction (e.g. He & Wong, 2004; Sarkees & Hulland, 2009). As a result, the concept of organizational ambidexterity has seen a significant increase in attention in recent years (Stettner & Lavie, 2014) and has been applied to many fields (Simsek, 2009).
Within professional service firms, individuals often have to balance exploring for new knowledge and exploiting existing knowledge; balancing the needs of their current and potential future clients, with the needs of the organization. Finding this balance is hard, as often the time that is not spent on clients is seen as non-productive time, pushing the focus back to routines (Simons, 1995). At the same time, consultants are expected to add to the firms’ knowledge assets, not only for the firm’s success, but also to advance their careers. The
demands from the environment combined with the reward structure within many of these PSF’s make it so that the ability to properly balance these exploration and exploitation tasks are instrumental to the consultants themselves. PSF’s and their consultants need to be able manage both changes in client demands in the short-term and develop their firm’s assets in the long term in order to survive (Anand et al., 2007).
Professional service firms serve as an example of highly knowledge intensive firms (e.g. Von Nordenflycht, 2010). Compared to manufacturing firms, PSF’s have relatively flat structures and the need to both produce and consume their generated knowledge at the forefront, while dealing with a high level of customer involvement into their finalized outputs (Grönroos, 2011). Similar to the role held by strategic account managers in Tempelaar and Rosenkranz’s (2017) article, the need to balance current client needs and business demands provides an interesting context for individual level ambidexterity. The difficulties noted by Levinthal and March (1993) are even greater as here, as there are less tools available to use for balancing exploration and exploitation. Looking at how PSF’s are often organized (Von Nordenflycht, 2010), structural ambidexterity, or the use of physically separate organizational units to focus on the different exploration and exploitation tasks (Gupta et al, 2006), is less likely to occur as knowledge gets created and consumed at the forefront. The same goes for temporal cycling such as described by Burgelman (2002); consultants tend to move from project to project with little time in between in an effort to keep up their billable hours (Simons, 1995). As productivity within these firms is measured predominantly by calculating the ratio of input and output resources, or the total amount of hours divided by the hours charged to the client (Lowendalh, 2000; Maister, 1993), allocating time for the exploitation of knowledge does not receive priority. It is therefore then that the context of PSF’s is one where
individual differences in the ability to act ambidextrous are critical to one’s own career and the firm’s performance.
By examining the role of Intellectual capital, the consultants Entrepreneurial Orientation and the firms value proposition (Treacy and Wiersema, 1993), this thesis will answer to calls by authors in the field to study additional individual- and multi-level antecedents and outcomes of individual level ambidextrous behavior (Birkinshaw & Gupta, 2013; O'Reilly & Tushman, 2013; Tempelaar & Rosenkranz, 2017).
Within PSF’s there are important reasons to focus on building Intellectual Capital. Knowledge (Løwendahl, Revang, & Fosstenløkken, 2001) and human capital are seen as the most important resources. Through human capital, PSF’s are able to develop their Social Capital as they stored over time in the relations they built with clients (Hitt, Bierman, Shimizu, & Kochhar, 2001; Suddaby & Greenwood, 2005). Internally, it allows organizational learning to take place through the sharing of insights (Stata & Almond, 1989). Through organizational capital, PSF’s are able to preserve knowledge and improve over time (Subramaniam & Youndt, 2005). Despite the seeming importance, there is little research on the influence of intellectual capital at the level of the individual in combination with ambidexterity.
Even within the organizational context, individuals still exhibit different levels of ambidexterity (e.g. Mom et al., 2009). As argued by (Kauppila & Tempelaar, 2016) there is a need to look at more individual level factors influencing ambidexterity. In their study they looked at self-efficacy and learning orientation as individual level factors drawn from social-cognitive theory to explain individual differences in ambidexterity. To get to a broader understanding of the concept we need to identify more factors that play a role into the extent that an individual acts ambidextrous. Due to the type of work, the role of a consultant is inherently entrepreneurial; the high level of autonomy and the constant need to create new
opportunities for the business and themselves (Baaij, 2013) require consultants to be pro-active (Anand et al., 2007) and innovation is required to be able to create new knowledge assets (Maister, 1983). While originally a firm level construct (e.g. Covin & Slevin, 1989), Entrepreneurial Orientation has recently started to get recognition as an individual level construct (Bolton, 2012; Langkamp Bolton & Lane, 2012). At the individual level it represents the traits and attitudes of an individual instead of those of the firm they work for. Entrepreneurial Orientation measures the propensity to take risks, be pro-active and be innovative. As a firm level construct Entrepreneurial Orientation has shown to influence the type of learning and levels of exploratory and exploitative learning (Kollmann & Stockmann, 2010). This thesis will look at if a consultants Entrepreneurial Orientation mediates the relation between his Intellectual Capital and level of ambidextrous behavior.
The third and final construct we will look at in relation to ambidexterity is the consultancy firms value proposition. Treacy and Wiersema (1983) empirically observed three value-based strategies used by leading firms to achieve strategic fit with their target markets and customers: customer intimacy, operational excellence and product leadership. Firms that were successful chose on of the three strategies while keeping up with the best standard on the other two dimensions. As we will try to argue, these three value proposition place different requirements on employees with respect to the needed balance of exploration and exploitation, thereby potentially influencing an employees’ ambidextrous behavior.
Our theory and analysis will make contributions to our current understanding of the ambidexterity within the context of professional service firms. First, by examining the relation between ambidexterity and Entrepreneurial Orientation within the consultancy context, we are able to test and introduce concept that has to our knowledge only been used at the firm
level, to measure firm or strategic ambidexterity (e.g. Kollmann & Stockmann, 2010). As recent studies by Bolton (2012) have shown the validity of the construct at the individual level, we answer to a calls made by Tempelaar and Rosenkranz (2017) and others to identify and examine more possible antecedents for individual level ambidextrous behavior. The current context of consultancy is well suited as it is a context where pro-activeness and innovation are required, both for firm survival through the creation of knowledge assets and for career advancements (Maister, 1983; Von Nordenflycht, 2010).
We answer to calls for studies needing to examine multiple levels of analysis. By introducing a new strategy level concept to the ambidexterity literature in the form of Treacy and Wiersema’s (1993) value propositions. By doing so we introduce a contextual factor that can potentially influence the levels of exploration and exploitation as different value propositions require different ratios of exploration and exploitation.
Through extensive hierarchical linear regression modeling provide a fine-grained insight in how within the context of consulting firms different aspects of intellectual capital relate to Entrepreneurial Orientation and its subscales.
While we only find weak levels of evidence for some of our Hypotheses, the lack of stronger results can potentially be traced back to the small sample size and some idiosyncrasies regarding the data set. The results do provide insights into where future research is needed.
The context of the Professional service firm
Professional service firms serve as an example of highly knowledge intensive firms (Von Nordenflycht, 2010) and are seen as important for the development of the knowledge economy (Gardner, Anand, & Morris, 2008; Lorsch & Tierney, 2002). Relying on the expertise and experience of a highly educated and (semi-) professionalized workforce, they create and deliver a set of intangible expert services to their clients (Anand, Gardner, & Morris, 2007; Hitt et al., 2001; Maister, 1983; Suddaby & Greenwood, 2005; Suddaby, Greenwood, & Wilderom, 2008; Teece, 2003). Knowledge (Løwendahl et al., 2001) and human capital are seen as their most important resource, through which these firms develop their social capital solidified in relations with clients (Hitt et al., 2001; Suddaby & Greenwood, 2005). As other types of firms become more knowledge intensive, scholars have taken an interest at looking at professional service firms to understand how to manage human capital and knowledge within such firms (Teece, 2003). Others have stressed the difference with manufacturing firms, stating that PSF’s have relatively flat structures and the need to both produce and consume their generated knowledge, and deal with a high level of customer involvement into their finalized outputs (Grönroos, 2011).
Within the group of PSF’s, Von Nordenflycht (2010) introduced a typology based on characteristics such as capital intensity, professionalization of the workforce and available slack. Relevant to this thesis is what he identified as ‘neo-PSF’s’ which include advertising agencies and consulting firms. These PSF’s, characterized by low slack, a semi-professionalized workforce and low capital intensity operate in a highly dynamic environment due to a lack of protection. The paradoxical nature associated with ambidexterity is inherent to the business and its organizational and economic structure. Productivity within these firms
is measured by calculating the ratio of input and output resources. For these firms this means the total amount of hours divided by the hours that can be charged to the client (Lowendalh, 2000; Maister, 1993). Time that is not spent on clients is then seen as non-productive time, pushing the focus to routines (Simons, 1995). At the same time, consultants are expected to add to the firms’ knowledge assets, not only for the firm but also to advance their careers (e.g. Anand et al., 2007). Crossan, Lane, and White (1999) argue that PSF’s and their consultants need to manage both changes in client demands in the short-term and develop their firm’s assets in the long term. If we look at ambidexterity literature this is similar to the role held by strategic account managers from the article by Tempelaar and Rosenkranz (2017). Strategic account managers also need to be able to balance the demands for both current clients and business and therefore provide an interesting context for individual ambidexterity.
Suddaby and Greenwood (2001) also showed a process of commodification of knowledge in effort to create efficiency gains similar to how Maister (1982) described how PSF’s aim to gain efficiency by increasing leverage ratios. This is done through a process of codifying, abstracting and translating knowledge so that it can get used by lower level consultants on wider scale increasing the efficiency of the firm. They describe this process as if it is a temporal separate process, where knowledge gets created, gets used and spread, till new knowledge comes to replace it.
Extant research on PSF’s has focused on a variety of topics; the role of culture on (role-) identities (e.g. Robertson and Swan (2003(role-)(role-) and performance (Choo et al., 2006(role-); agency issues due to knowledge asymmetry between professionals and clients (Sharma, 1997) and the co-production of knowledge (Bettencourt, Ostrom, Brown, & Roundtree, 2002) and the motivations for clients to hire consultants (Pardos, Gómez-Loscos, & Rubiera-Morollón, 2007). As stated above however, research that focusses on innovation and what causes some
individuals to explore or exploit more than others within these firms has been lacking (Fischer, 2011). When there was research on how new knowledge gets created it is focused on the process in larger firms (Anand et al., 2007).
For our current research context, we are interested in the smaller firms. Within smaller firms the problems associated with the paradoxical nature become even more interesting as the size of the unit determines the extent to which exploration and exploitation can be done simultaneously (Gupta et al., 2006). While in many large conglomerate PSF’s knowledge centers or centers of excellence were created (Moore & Birkinshaw, 1998) which could be seen as a form of structural ambidexterity. Due to the imperfect task programmability and outcome measurability, contextual forms of ambidexterity are needed. Professionals within PSF’s are assume a certain degree of autonomy, while still needing to be part of a system (Von Nordenflycht, 2010). Some authors have pointed to processes such as performance measurement systems and the requirement to create new knowledge to progress in your career have worked (Anand et al., 2007), socialization processes or social identities (Alvesson,1995) and the importance of normative controls (Greenwood et al., 2005) as a way to promote contextual ambidexterity. However, within smaller firms and especially freelancers, these options seem mostly absent. Questions regarding ambidexterity in this context should therefore most likely focus on the individual level.
Individual ambidexterity
The subject of ambidexterity has received considerable attention. Organizational ambidexterity is a term first coined by Duncan (1976). It was given meaning when authors applied it to March’s (1991) discussion on the need for organizations to conduct exploration and exploitation activities to be able to survive. Several challenges have been observed and described by existing literature (Abernathy & Utterback, 1978; Andriopoulos & Lewis, 2009; Benner & Tushman, 2003; Turner, Swart, & Maylor, 2013). Exploration and exploitation are seen as fundamentally different activities that rely on distinctive organizational routines and involve different organizational learning models (Benner & Tushman, 2003). Where exploration relies on search for new knowledge, thus facilitate experimentation, flexibility and risk taking (McGrath, 2001), exploitation is based on routines that leverage firm’s existing knowledge, thus facilitating consistency, stability and control (Benner & Tushman, 2003). These challenges can be solved at different levels (Turner et al., 2013) and further categorized by their reliance of type of intellectual capital (IC). Ambidexterity at the organizational level is thus difficult to achieve (He & Wong, 2004). A firm that engages in both exploration and exploitation is expected to maintain both productivity and innovation, achieving reliability while enabling organizational renewal and thus enjoying enhanced performance (Stettner & Lavie, 2014). Recently some have stated to look at a more fine-grained way towards organizational ambidexterity. Voss and Voss (2013) argued that the relation between ambidexterity and firm performance relies on firm size, firm age and the functional domains (product- or market-domains) within, or across which, firms try to achieve ambidexterity. Functional domain separation such as the one by Voss & Voss (2013) may also be applicable to the individual level. Multiple scholars have argued that achieving ambidexterity at the
individual level is considered be more difficult, if not impossible within the same domain (e.g. Gupta et al., 2006). Over the recent years we therefore saw a rise of focus on the individual level ambidexterity.
The role of individuals within ambidexterity was first studied within structural ambidexterity. As individuals were assumed to work in either structurally or temporal silos, the processes of exploration and exploitation were assumed to be divided in the lower levels, while being orchestrated by top management teams (Smith & Tushman, 2005). However current research on individual level ambidexterity is more focused on contextual ambidexterity introduced by (Gibson & Birkinshaw, 2004). Within contextual ambidexterity firms both trust and encourage employees to exploit, while using systems to make sure they exploit existing value propositions. Individual-level ambidexterity is hypothesized to have positive firm level performance due to their ability to hold contradictory or paradoxical demands (Mom et al., 2009). Extant literature has looked at individual-level ambidexterity in managerial teams (Mom et al., 2007) psychological characteristics and traits (e.g. Good & Michel, 2013; Laureiro-Martinez et al., 2015) and even neuroscience approaches have been looked at (Aston-Jones and Cohen, 2005). More recently authors have looked at behavioral (Keller and Weibler,2014) socio-cognitive antecedents (Kauppila & Tempelaar, 2016). Tempelaar and Rosenkranz (2017) studies individuals’ predisposition for role integration or segmentation. The study by Kauppila & Tempelaar (2016) also found that variance in individual ambidexterity was high in groups within the firm, explicating the need to consider more individual-level predictors of ambidextrous behavior.
Value propositions
To see how value disciplines, help, guide and influence the decisions regarding levels of ambidexterity we need to look at how Treacy and Wiersema (1993) describe the process. Although strategies can be formulated multiple ways (e.g. Porter, 1980), we choose to test Treacy and Wiersema’s (1993, 1995) value disciplines due to their focus on customer relations which matches our service context. In their seminal article, Treacy and Wiersema (1993) empirically observed three value-based strategies used by leading firms to achieve strategic fit with their target markets and customers: customer intimacy, operational excellence and
product leadership. Firms that were successful, chose one of the three strategies while keeping
up with the best standards on the other two dimensions. Compared to Porter’s (1980) generic strategies and the framework by Miles, Snow, Meyer, and Coleman (1978), the value discipline framework introduced the importance of client relationships through its strategy of customer intimacy. With customer intimacy firms are assumed to pursue a strategy aimed to generate a thorough understanding of their customers and tailor their existing products to their (changing) needs. A key focus therefore becomes the relationship with existing clients and the need for
adaptability (Ulaga & Reinartz, 2011) in the form of exploration through constantly making
exploitive changes to existing products based on input from the environment using iterations of their current technologies. Product leadership is heavily dependent on exploration, sometimes to the detriment of existing products, with a focus on constant innovation and development of the firms’ portfolio. This is similar to Atuahene-Gima (2005) definition of exploration as the firms’ ability to develop new products through experimentation. Operational
excellence is heavily dependent on exploitation of the firms’ current resources, looking to
The focus on customer value and the importance these value propositions lend to the relations between the firm and its customers make it suitable for PSF’s (e.g. Hitt et al., 2001; Greenwood et al., 2005) as Hitt, Bierman, Uhlenbruck, and Shimizu (2006) indicate that selection between service providers is done by looking at who provides them with the most value (Sirmon, Hitt, & Ireland, 2007). The need to be responsive to clients and provide services custom to their needs is also pointed out by (Griffith & Harvey, 2004). The model is also grounded in industrial economics theory (Tallon, 2007) and has been used in strategic management literature over the last two decades (e.g. Thornhill & White, 2007). While some studies have looked at linking ambidexterity to strategical choices, none have looked at Treacy and Wiersema’s (1993) value propositions. Magnusson and Martini (2008) stressed the need to balance operational excellence with continuous innovation to be important for all firms. Tushman and O’Reilly (1996) argued that the firms’ strategy needs to support innovation, differentiation and cost simultaneously. A study by Auh and Menguc (2005) has looked at a similar area with respect to the framework by Miles and Snow (1978) and found that the groups in the framework, exploration and exploitation had different relations with firm performance. However, Sethi and Sethi (2009) argue that focusing on one aspect (i.e. quality) negatively influences the firm’s ability to innovate. Within the small consulting firm context of this study we expect these choices to be of a bigger factor. Ebben and Johnson (2005) found that within small firms, those that focused either on efficiency or on flexibility performed better than those that tried to balance both. Coa et al. (2009) however argued that for smaller firms a balanced state between exploration and exploitation is more suitable. The strategic choices made therefore seem to influence the levels of exploration and exploitation and the resulting level of ambidexterity. We argue that within the small consultancies that form the context of our study, these strategic choices are closely related to the value proposition provided to clients. As Treacy and Wiersema (1993) argue that we should increase focus on one while keeping parity on others we expect levels of ambidexterity to vary by the degree to which they focus on a certain value proposition. When placing Product Leadership in the context of the three archetypes of consulting by Maister (1983), it closely resembles Brain-type consulting. The
focus here is exploration in the product domain, which leads us to predict that for firms implementing a Product Leadership value proposition we expect a higher level of exploration while maintaining parity on explorative tasks, resulting in higher levels of ambidexterity.
Intellectual capital
While we argue that the choice of value propositions by Treacy and Wiersema (1993) influences the levels of ambidexterity within these small firms, we also need to examine what enables them to behave ambidextrous. As we have already argued, PSF’s are heavily knowledge-based organizations, where learning takes place by exploiting and refining existing knowledge and generating new knowledge. Similar to Kang and Snell (2009) and Simsek et al. (2009), we therefore argue that within these small firms we need to look at their knowledge and Intellectual Capital (IC) resources as the inputs for enabling ambidexterity. Given the small size of our firms however, the levels of ambidexterity of the firm will often equal the level of individual ambidexterity, allowing it to be studied at a new level compared to existing research. IC is described by Youndt et al. (2005) as all knowledge and knowing capabilities that can be used to create a competitive advantage. Most research seems to separate three aspects of IC; namely human capital (HC), social capital (SC) and organizational capital (OC) (e.g. Swart, 2006). HC has been most commonly referred to as tacit and resting within the individuals of the firm (Simon, 1991), representing the skill and abilities the firms’ employees possess (Flamholtz and Lacey, 1981). Burt (1992) defined SC as the resources embedded within the firm that available and derived from a network of the relationships that exist within the firm or with its customers. OC refers to the firms institutionalized knowledge and its codified experience stored in patents, databases, manuals and routines (Youndt, 2004). Kang and Snell (2009) introduce a model based on the type of HC (generalist vs specialist), SC (entrepreneurial vs cooperative) and OC (organic vs mechanistic) to come to different organizational structural solutions. However, within our small consulting firms comprised often of just a single consultant, these are often not an option. Similar to Youndt et al. (2005), using a scale based on their work, we will therefore look at differences on the overall level of
each type of IC (HC, SC and OC) and their relation to individual level of ambidexterity. We realize that due to their definitions, OC and SC (internal SC) are not suitable for self-employed consultants and will limit self-employed consultants to Human Capital. According to Nahapiet and Ghoshal (1998) SC helps with the creation and exploitation of knowledge, while Subramanian and Youndt (2005) found that a firm’s SC enables it capability to develop incremental and radical innovation. We therefore expect HC and SC to have a positive influence on individual level ambidexterity by influencing both exploration and exploration. OC is expected to help with the codification of knowledge within these small firms therefore helping exploitation without decreasing the amount of exploitation.
Entrepreneurial Orientation
According to Lumpkin and Dess (1996) entrepreneurial orientation (EO) is about the processes that explain how firms deal with the creation of new ventures. The most prevalent definition is based on a definition for entrepreneurial firms by Miller (1983) which focusses on the tendency for firms to innovate, partake in risky ventures and be pro-active. Multiple studies have linked EO to performance at the firm level in terms of profitability, growth and product innovations (Avlonitis & Salavou, 2007; Wiklund & Shepherd, 2003). Wiklund and Shepard (2003) argue that this effect is even greater in small firms. While there has been this focus on the firm level, recently scholars have looked at applying the concept at the individual level where it showcased the same three dimensions, this time representing personal tendencies to be more or less innovative, proactive or willing to take risk. To our knowledge studies on EO and ambidexterity have been low in number and limited to the firm level, this while entrepreneurship is often studied at the individual level. Ahuja and Lamper (2001) described the goal of EO to use existing knowledge and new knowledge to bring something new to the market. Kollman and Stockman (2014) test the effect of EO on exploration and exploitation individually and find that innovativeness, pro-activeness and willingness to take risk all stimulate exploration. They only find pro-activeness to have a positive influence on exploitation, while neither innovativeness and willingness had a negative relation to
exploitation. As March (1991) and others have argued, exploration activities are inherently riskier as they involve more uncertainty. Individuals that are less risk-averse and more pro-active for instance are therefore more likely to explore even within a given value discipline. This could help to explain why some individual consultants focus more on exploitation of their existing knowledge and customers. In larger consultancies it could help to address the notion of agency identified by Anand et al. (2007) as a requisite for individuals to create new knowledge structures along the different paths. At the individual level EO is transformed to a measure of the individuals preferred extent to which they act entrepreneurial (Langkamp Bolton & Lane, 2012), acting it to become a more state-like behavioral orientation that is somewhat stable over time. As we will study EO as a mediator, we look at Miller (2011) who argues that IC and specifically SC can be seen as an incentive to increase innovation and risk-taking by improving relations with suppliers. As there has been a lot of discussion lately on whether EO should be looked at as a unidimensional construct by testing each of the subscales individually, we will look at both in the current theses (for a full discussion see Covin, 2008) with respect to what extent the relation between IC and IA is mediated by EO.
We also argue that the value proposition chosen by the firm influences level of EO. The reason for this is found in Baaij (2013) and Maister (1982) among others who describe a consulting firms’ value proposition as one that is hard to change. Maister (1982) and Hudson, Smart, and Bourne (2001) argue that the firms value proposition is heavily linked to the firms economic and organizational structure, which in turn determines both the consultants it attracts (and thus its IC) and the type of clients a firm attracts. The type of clients the firm attracts influences the type of knowledge that is needed and the balance between exploration and exploitation (or codification)(Baaij, 2013). By influencing the type of clients the firm attracts, we argue that the value proposition also influences the level of EO as it influences the type of demands that are placed on the firm with respect to the need to innovate, be pro-active and take risk. As we focus on the Product Leadership value proposition we argue that the mediating role of EO between IC and ambidexterity is positively moderated by the extent to which the consultant or his firm implements the Product leadership value proposition.
Theoretical model
Hypothesis 1 Intellectual capital as represented by (a) Human Capital, (b) Organizational Capital and (c) Social capital is positively related to individual level ambidexterity
Hypothesis 2 Entrepreneurial Orientation and each of its components mediates the positive relationship between Intellectual capital and ambidexterity. Hypotheses 3 The extent to which a consultancy firm or self-employed
consultant implements a Product Leadership value proposition towards his customers is positively related to the consultants’ ambidextrous behavior.
Hypothesis 4 The extent to which a consultant firm or self-employed consultant implements a Product Leadership value proposition moderates the positive relationship between the consultants Entrepreneurial
Orientation and ambidextrous behavior such that the relation is more positive when product leadership is high.
Method
Data collection
We collected data through surveys that were spread among members of the ‘Orde van organisatie adviseurs’ (Ooa); a Dutch trade organization for management consultants which consists of more than 1.200 members. Ooa board members indicate that current member demographics show that the average age of its members is 47 years and consists mostly of three different categories of consultants; (1) small consultancy firms, and consultants that either (2) work within networks in collective groups or (3) are self-employed and offer their services on their own. Within the Netherlands a recent study found that over 90% of the total registered management consultants were self-owned and/or operated within a network (CBS, 2016), with numbers almost doubling over the last decade (Consultancy.nl, 2016).
Members were informed on the survey first through a newsletter which was spread through email, after which another email was sent containing the cover letter with a link to the online survey. To help increase participation we offered the incentive of having the ability to receive a personalized report based on the results from the survey, promising to provide some feedback on how they compared to others within the sample while making sure confidentiality was promised. In total, we send out x invitations to emails from the Ooa’s database of members. The members of the Ooa completed measures of Value Propositions, Intellectual Capital, Entrepreneurial Orientation, Exploration and Exploitation activities, as well as Performance management and support context and on project selection preferences. The survey also contained a set of questions that were part of an index developed to gain insight in the current consultancy market. This part contained questions on e.g. the percentage of billable hours,
yearly revenue asked respondents to answer how much of their time they spend on clients in different industries. Unfortunately, due to timing restrictions all data was collected in a single survey, as there was no room for a two-wave design separating independent and dependent variables.
The total number of respondents was 174, corresponding to a response rate of 15%. The relatively low response rate can be at least partially attributed to unfortunate timing. To increase response rate, three reminders were sent out through email and the total running time for the survey was extended by one week for a total of 28 days. The total number of completed responses was lower and the decision was made to exclude responses that answered less than 40% of all questions. Furthermore, two responses were excluded due to a lack of variation in their answers as well as unrealistic answers in their demographics. This left us with 113 valid responses, of which 103 fully completed the questionnaire. The average age in the sample is 54 years and 63 percent was male. Almost half of the sample (47 percent) indicated to operate as freelancers, while the rest was acting as part of an organization (38 percent) or as a freelancer is a permanent collaborative partnership with others (14 percent). Government and healthcare were indicated to be the most prevalent industries.
Exploratory factor analysis
For the extraction of factors in our exploratory factor analysis we use principal axis factoring (PAF). Gorsuch (1997) and McArdle (1990) argue that compared to principal components analysis (PCA), PAF is less likely to inflate the variance accounted for by the components. For each factor analysis we do, we also run an additional analysis using principal component analysis to see if significant differences between the two extraction methods would show. Result show that the number of extracted factors stayed the same and all items were loaded
to the same factors with only slight differences in the size of the loadings. This seems to complement arguments by Arrindell & van der Ende (1985) and Schoenmann (1990), who argue that there is little difference between the outcomes of the two extraction methods. As others argue to limit the use of components analysis in favor of true factor analysis (e.g. Benterler & Kano, 1990), and to improve the readability of our tables, factor loadings will be interpreted and presented using the loadings of the extraction done with principal factor analysis. To ensure that correlations between items are sufficiently large for PAF we will also conduct Bartlett’s test of sphericity. To further provide support for the suitability of our sample size for factor analysis on the items we use, we will conduct and interpret a KMO-test based on work by Kaiser (1970) and Kaiser & Rice (1974), which provides guidelines to determine if our sample is suitable for the factor analysis.
To determine how many factors to accept for our variables we could use the scree plots as a first indicator (Cattel, 1966). However, Stevens (2002) argues that the scree plot only provides a fairly reliable indicator for factor selection when the sample size is over 200 cases. Since our sample size is smaller, we will look instead at Kaisers’ criterion (1960, 1970), which posits that we should pertain factors with a value greater than 1. Other guidelines exist, such as Jollife’s (1972, 1986) accepting factors with eigenvalues greater than 0.7, however Stevens (2002) found that this guideline often overestimates the number of factors. While Kaiser’s (1960, 1970) criterion has received some criticism recently (e.g. Hayton, Allen & Scarpello (2004), it still provides a clear guide that is prevalent within the current body of research on ambidexterity.
For individual items we look at those that load high on a single factor. As our sample size is just over a 100 individuals, Comrey and Lee (1992) would classify it as a poor sample size to conduct factor analysis on. Stevens (2002) argued that for a sample size of 100,
loadings for each item should be greater than 0.512. Guadagnoli and Velicer (1988) argue that we need to find factors with loading greater than 0.6, while MaCallum, Widaman, Zhang & Hong (1999) stress that with our sample size of 100 cases, we should have communalities between 0.5 and 0.6 to get reliable measures. As a result, will select items based on loadings greater than .512 following Stevens (2012) and for unclear cases will also look at their communalities.
To determine internal consistency, we will use Cronbach’s alpha (Cronbach, 1951). Although there has been some discussion on what alpha constitutes as a sufficient value, with values ranging as low as 0.5 for early stage research (Nunnaly, 1978) to as high as 0.8 (Kline, 1999), for psychological constructs such as our Entrepreneurial Orientation and Ambidexterity variables, we will consider our scales reliable if Cronbach’s alpha is 0.7 or higher. In accordance with Cronbach (1958) we will test for reliability on individual subscales and the overall scale where applicable. As more researchers have been pointing out the limitations of using Cronbach’s alpha as the single measure for reliability (e.g. Shook, Ketchen, Hult, Kacmar, 2004; Cortina, 1993) we will also present each measure’s composite reliability (CR) as it considers the varying factor loadings of each item on the scale.
Measures
Individual ambidexterity.
We measured individual level ambidexterity using scales developed by Mom et al. (2007, 2009), which have been validated by multiple authors since (e.g. Kauppila and Tempelaar, 2017). According to Mom et al. (2007) individual level ambidexterity should be measured in two steps. The first consists of capturing the separate Exploration and Exploitation scales, followed by a second step that computes the multiplication of the mean scores on these scales to get the score for individual ambidexterity. Participants were asked to respond to items belonging to both the Exploration and Exploitation scales, which asked participants to evaluate the extent to which they had pursued activities related to explorative and exploitative tasks on a five-point Likert scale (1 = ‘To a small extent’, 5 = ‘To a large extent). Examples for items for the Explorative scale are ‘Activiteiten waarvoor u nieuwe vaardigheden
of kennis moet opdoen’ and ‘Activiteiten die (nog) niet duidelijk bij het bestaande
bedrijfsbeleid horen’. For the exploitative scale items asked participants ‘Activiteiten waar u al veel ervaring mee had’ and ‘Activiteiten die u uitvoert alsof het routine is.’
To confirm the two scales, we conducted an exploratory factor analysis using the principal axis factoring extraction method, using both Direct Oblimin and Varimax rotation methods. Direct Oblimin was initially used to check for possible correlation between the two factors that would be left unexplained when using only Varimax rotation. An initial factor analysis and correlation matrix caused us to remove two items from both scales. Bartlett’s test of sphericity, Xˆ2 (28) = 191.363, p = .000, indicated that the use of PAF was indeed justified. Kaiser-Meyer-Olkin measure verified the sampling adequacy for the analysis, KMO = .690 (just short of ‘middling’ according to Kaisers and Rice (1974)), and all KMO values for
individual items were greater than .604, exceeding the lower boundary of .5 (Kiaser and Rice, 1974). A decision was made to remove another item both the Exploration and Exploitation scales. This was the result of item loadings that were below the threshold (.512) based on Stevens (2002) discussed above, combined with a low communality (<.3). We found 8 (28%) non-redundant residuals between the observed and reproduced correlations, which is below the threshold of 50% and thus no cause for concern. The determinant of the resulting correlation matrix was .140.
After removing the items, factor analysis resulted in two factors representing Exploration and Exploitation activities, cumulatively explaining 42.658% of total variance. Eigenvalues after extraction were respectively 1.993 for the Exploitation scale and 1.419 for the Exploitation scales. Communalities for items on these scales were moderate after extraction (Velicer and Fava, 1998), providing further support for our choice for principal axis factoring (Gorsuch, 1997; McArdle, 1990). The resulting Exploration scale had a questionable reliability (Cronbach’s a = 0.671) which could be attributed to the sample size, convergent reliability (CR = .693, AVE = .364), loadings above .533 and cross-loadings below .122, and further deleting items would not improve the scale further. The four item Exploitation scale had an acceptable reliability (Cronbach’s a. =.773), convergent reliability (CR = .779, AVE= .472), with corrected item-total correlations above 0.521 and no shown improvement when deleting any of the remaining items. Item loadings were above .625 and cross-loadings below .126.
As Mom et al (2007, 2009) have theorized Exploration and Exploitation as two orthogonal constructs, principal factor analysis was also conducted using a Varimax rotation. This created nearly identical factors with loadings on individual items differing no more than 0.05 on an item’s loading, which provides further support that Exploration and Exploitation
can indeed be considered orthogonal constructs when using the current scales (Coa et al., 2009, Gibson and Birkinshaw, 2004; Mom et al., 2007). To complete the first step, we computed the scores for the Exploration and Exploitation scales by calculating the averages of the items that we have retained.
The second step for Mom et al.’s (2007) ambidexterity scale is completed by computing the multiplication of the values for exploration and exploitation to create the variable for individual ambidexterity that we will use in our model. This method is in line with previous studies (e.g. Tempelaar and Rozenkranz, 2017, Mom et al., 2007). It complies to the view of ambidexterity as a multi-dimensional construct, consisting of the interdependent dimensions of exploration and exploitation (e.g. Coa et al., 2009). As individual ambidexterity is a dependent variable in our model we calculated its value using the multiplication of the non-mean centered values from Exploration and Exploitation.
Entrepreneurial orientation
Entrepreneurial Orientation was measured using a 9-item measurement scale adopted from Covin and Slevin (1989) after advice from the thesis supervisor. While the original scale was used for the firm level construct, all items were translated to represent respondents’ own positions. The scale was also adjusted to the service context on the items where it was necessary. We measured all items using a 7-point semantic different scale, used in the original scale as Covin and Slevin (1986, 1989) argue that organizations could either be administrative or entrepreneurial.
To test the measurement scales we conducted an exploratory factor analysis using the principal axis factoring extraction method, using a Varimax rotation method which was shown
to be used primarily in extant research. Bartlett’s test of sphericity, X2(28) = 336.000, p = .000, indicated that our use of PAF was justified. Kaiser-Meyer-Olkin measure verified the sampling adequacy for the analysis, KMO = .784 (‘middling’ according to Kaisers and Rice (1974)), and all KMO values for individual items were greater than .722, exceeding the lower boundary of .5 (Kaiser and Rice, 1974). An initial analysis showed no items sharing correlations above 0.8 or below 0.3, and the determinant of .045 indicated no multicollinearity between items. Factor analysis did show that one item loaded high on both Proactiveness (λ = 0.528) as on Innovativeness (λ = .510). This is consistent with prior research that suggests that proactiveness can be quite ambiguous and is sometimes left out (Lumpkin and Dess, 1996). We decided to remove the item from the scale, leaving our Proactiveness factor with just two items. While a factor with fewer than three items is generally considered weak and unstable (Osborne & Costello, 2009), recent evidence suggests that even single-item measures may suffice (Bergkvist and Rossiter 2007; Wanous, Reichers, and Hudy 1997). Proactiveness had a high reliability (Cronbach’s α = .812), convergent reliability (CR = .755, AVE = .406), with corrected item-total correlations above .684, loadings above .692 and cross-loadings below .271. For the Innovativeness scale we were left with three items, with a high reliability (Cronbach’s α = .783), convergent reliability (CR = .732, AVE = .477) with factor loadings above .602, cross-loadings below .329. Finally, for willingness to take risk three items were retained and the scale had a high reliability (Cronbach’s α = .791, CR = .758, AVE = .512), with loadings above .666 and cross-loadings below .312. Communalities were moderate to high for the remaining eight items and were all above .532, which together with the corrected-Item Total correlations between items (above .611), indicated that no more items needed to be dropped. We only found two (7.0%) non-redundant residuals between the observed and reproduced correlations, which is far below the threshold of 50%.
To be able to test for individual dimensions as well as the Entrepreneurial Orientation variable, we first created variables for each dimension by calculating the mean scores from their respective items. We then calculated the Entrepreneurial Orientation variable by taking the mean scores for all the remaining eight items as was done in extant research (examples). Higher overall scores on the EO scale indicate a more entrepreneurial orientation, while lower scores can be interpreted as representing a more conservative orientation (Covin and Wales, 2011).
Intellectual capital
Intellectual capital was measured using a scale adapted from Reed, Lubatkin and Srinivasan (2006). The scale was already adapted to the service industry and only small adjustments needed to be made for the consulting industry context. Intellectual capital is built from taking the means of the items for the dimensions of human capital (12 items), internal social capital
(7 items), external social capital (had to remove 1 item as it was found not suitable for the
research context) and organizational capital (5 items). All items were adapted to the individual level where appropriate and possible in consultation with the thesis supervisor.
Human capital was adapted from Reed et al (2006), who adapted it from Huselid et al. (1997) and Youndt et al., (2004). Although Reed et al. (2006) have already adapted wording for individual items to the service industry for their research, some wordings for items were adjusted to better fit the consulting industry. Initial analysis showed issues with the correlations between different items on the Human capital dimensions, at least six of the items showed low correlations with multiple other items. After an extensive and careful factor analysis using principle axis factoring and Direct Oblimin rotation, we found that we had to
delete multiple items from the total scale. For Human Capital we were only able to retain four out of the original twelve items. Adding any of the items we removed would increase the number of factors, without being able to find a shared concept between the questions of the items. Items that were removed had low factor loadings and high cross-loadings with the internal social capital factor. After ended up with a six item Human Capital factor with a low reliability (a = 0.535, CR = .79, AVE = .48), no items to be deleted and correct item-total correlation above .312. All the mentioned steps were traced and redone using the principle component analysis extraction method and Varimax rotation, which led to the same factors and items being retained and removed. We therefore believe that the current factors provide the best fit considering our sample. However, the analysis did reveal a sort of dual nature to the scale as we adapted it. This might be due to our translations not perfectly fitting on the consultancy context, but it is something that deserves further attention.
Organizational Capital proved to be much less of an issue, and we ended up with all five original items from Youndt et al. (2004) for a scale with a high reliability (a = .875, CR = 0.9, AVE = .64). Internal Social capital ended up with four out of the seven items (a = .764, CR
= .79, AVE = .49). We do have concerns about the current set of items we have left from the
total scale and the extent to which it still represent the full Intellectual Capital scale. However, after having retraced each step multiple times, there seems to be no possibility to fix this with the current data set.
Value propositions
Despite popularity of Treacy and Morgan’s (1993,1995) value discipline framework, it is hard to find actual measures facilitating empirical research that are properly validated and used in multiple studies. Measures that do exist are often a collection of items from multiple articles, often originally used for other strategic orientation frameworks such as that by Porter (1980). A recent example of one that does exist is the one developed and used by Reimann et al. (2010), which includes seven items for the product leadership dimension (5-point Likert scale), three items for operational excellence and seven items for customer intimacy. As a firm level variable, it measures the extent to which the firm places focus on the different value propositions as identified by Treacy and Wiersema (1993). Respondents were asked to indicate to what degree they agreed to a certain aspect having been a focal point of their attention over the last 12 months.
With a high CR (CR = .88) for Product Leadership, CR = .71 for Customer Intimacy and CR = .86 for the Operational Excellence scales in Reimann’s (2010) study, we had hoped that we would get clear defined factors. However, especially for the Customer Intimacy scale this wasn’t the case. We had to remove 5 from the 7 items to get to a single factor, at which point we could choose whether the items we would use were about purely a focus on customer relationships or on the use of advertising. This might indicate either issues with our translation, between the scale and the current research context or the scale on its own. For our current study we ended up with a four-item Product leadership study (CR =0.65, AVE = 0.49). We took out the items for Customer Intimacy and Operational Excellence as they loaded over multiple factors in the integrated factor analysis and gave trouble in the confirmatory factor analysis discussed below.
Control variables
We controlled for age as Mom et al. (2009) among others have found that age is negatively related to ambidexterity. In prior research we saw age was taken as the logarithmic, however for the our sample we did not see a problematic skewness or kurtosis. Using a logarithmic translation on the variable resulted in a variable with both moderately substantial skewness and kurtosis. We therefore decided not to transform age. Following Kauppila and Tempelaar (2016) gender was also controlled for due to possible discriminatory effects. Gender was transformed from using 1 for Males and ‘2’ for Female, to 0 and 1 to prepare them for linear regression. Following results by Smith et al. (2005) we controlled for education level. Since we had answers only indicating HBO, WO and PHD, we created two dummy variables where the baseline was HBO. We also controlled for the number of clients as Floyd and Lane (200) found that the number of accounts may influence the extent to which consultants need to divide attention between different contexts. We transformed the values using a natural logarithm which reduced the skewness and kurtosis to close to zero.
Firm age was included as incumbent firms are more inclined toward exploitative efforts (Gilbert, 2004). As the data set had both a very strong skewness (2.975) and Kurtosis (9.034). We therefore transformed it using natural logarithms. This reduced skewness and kurtosis to be within |.3| and therefore acceptable. While the Shapiro-Wilk test was significant (p = .015) the Kolmogorov-Smirnov test was not. As both are strong tests and both the kurtosis and skewness were severely reduced as a result of the translation, we accepted the transformed version of firm age as a control variable. Similar to Jansen et al (2009) and others we include firm size as a control variable. This is important as these smaller businesses vary in size and their availability of slack resources (Simsek, Veiga, & Lubatkin, 2007). It is
measured by taking the logarithmic translation of the number of employees. Additionally, outliers within the data set were checked either online or through the Dutch KvK registers to make sure answer were valid. The logarithmic translation was required as the data set had a very strong skewness. Removing outliers would have removed valuable information and further reduced our sample size to below our threshold of N = 100 and was therefore not desired. Firm size and age were also controlled for as they are seen as affecting survival (e.g. Hannan and Freeman, 1984). Tenure is added as Beckman (2006) showed that tenure can shape managerial knowledge and understanding of firm routines, and that either a very high or low tenure will influence managers ability to combine strategic orientations for exploitation and exploration.
We also included a control variable for the industries the consultant primarily conducts work in. This due to the fact that industries differ in their environmental stability and require different levels of exploration and exploitation (e.g. Jansen, 2009). As a result, the consultants working for those specific industries may feel more or less pressure to explore and exploit as well. Participants were asked to allocate a percentage over multiple segments to indicate in which of the market segments they were most active. They were able to allocate over the segments of financial services, industrial, retail, government, utilities, healthcare and
diverse. After we analyzed the results we found that many of the respondents allocated the
same percentage to multiple industries. In order to not lose valuable information by selecting a single segment for dummy coding or leaving them out completely, we made effort to mitigate the potential loss of information for cases where equal time was allocated to multiple segments. To create useful dummy variables we first conducted an analysis to see how many participants allocated duplicate percentages over multiple segments and created dummy variables for these cases based on the following conditions; 1) respondents had to have
indicated the same percentage to two or more of the categories (this was the case in 33 of the respondents); 2) these shared percentages had to be the highest of the percentages allocated and 3) had to be above 30% and thus account for at least the majority of their allocated time. As a result, we ended up 9 additional combinations. We then created dummy variables setting diverse as the base segment and allocating each case to one of the dummy variables.
Further assessment of Measurements
Next we ran additional tests on our measures. First, we ran an integrated exploratory factor analysis on all the retained items for the independent and dependent variables in our model, using a principal component analysis with Verimax rotation. Unfortunately, this did not give us the results we had expected. While we did extract 9 factors, only 7 of them had all the items loading to the appropriate factor. The items for the scales of Proactiveness and Exploration shared some of their items with Product Leadership, which itself was split over two separate factors. The seven factors that did load on their respective factors, did so with high loadings (>0.5) and low cross-loadings. Through an extensive process and analysis, we have tried to redo our initial factor analysis on all of the scales multiple times, starting with the Exploration and the Value Proposition scales, and then later involving the scales for Entrepreneurial Orientation and Intellectual Capital, just to see if there were other combinations that would lead to a clear separation of factors. It seems this might be possible with the current sample however, only by limiting the number of items per scale significantly to three items or less. Doing this results in scales with lower reliability and less coverage of the original scales. We therefore decided not to do this. While the overall results may be a
cause of some concern and will need further inspection, they may also have been the result of the relatively small sample size judging by the KMO test value (KMO = .463).
Table containing the Integrated factor analysis
Having completed the exploratory factor analysis, we conducted a confirmatory factor analysis adding all retained items as observed variables linked to their latent variables. From this analysis we found that the model fits relatively well (Chi-square = 596.428, p = .000), a root mean square error of approximation (RMSEA = .055) which is very close to a good fit according to Steiger and Lind (1980), an incremental fit index (IFI = .867) and a comparative fit index (Bentler, 1990) (CFI = .846). We found that IFI and CFI were not quite high enough,
as a value of greater than .9 is generally considered as a cut-off criterion for the CFI (Bentler, 1990). In more recent studies it has been suggested that this should be adjusted to .95 (Hu and Bentler, 1999). However, there are some possible explanations for our lower value. We found that in the confirmatory factor analysis model all items did load on their respective latent variables (>.5) except for one Human Capital item (0.43). Deleting this item from the scale would however reduce its reliability and coverage of the original scale even further and therefore is not ideal. A second possible factor influencing our fit is our sample size; According to Meyers, Ahn and Jin (2011) we need a sample size of 200 (N>200) for properly testing our theoretical model with CFA. While these results can be interpreted and used as a critique of the model and its variables, it should be done with this in mind. The full model used for the CFA with can be seen in Appendix including the standardized regression weights.
We also used confirmatory factor analysis with different possible models to test the distinctiveness of our latent variables. We did this by juxtaposing models that differed in the number of factors to which items belonged to. For instance, we used our base model (Model 1) and compared this to a model that had all items combined into one factor (Model 5). For each of the models we saw that the fit became worse by looking the measures. As a result, we have some evidence that our proposed factors provide the best fit of the current data set. Taking the results from our CFA we believe that these findings provide further evidence of convergent and discriminant validity of our measures (Bagozzi and Yi, 2012), despite some of the issues we encountered in the Human Capital scale.
Due to time and logistic constraints we were not able to separate our independent and dependent variables in two separate surveys by time. This might be a cause of common method variance (Podsakoff, MacKenzie, & Podsakoff, 2012; Podsakoff, MacKenzie, Lee, & Podsakoff, 2003). Our survey participants did however have expertise and experience, which should enable them to fully comprehend our questions. We could not assure anonymity to our respondents, as respondents were promised a personalized report for which we would need to link their responses to their personal details. However, we did assure them confidentiality and the fact that their personalized rapport would be available only to them. It was made clear that the content of the report will contain their responses being analyzed in comparison to the full anonymized data set. We also repeatedly asked participants to be conscientious about the questions throughout the survey and provided an introduction when questions changed measurement levels and topics to help them better interpret the questions.
Similar to Tempelaar and Rozenkranz (2017) and following Podsakoff et al. (2003) we run additional tests to make sure common method bias was not an issue. First, we performed a Harman’s one factor tests using the items from our independent and dependent variables. The Harman’s one factor test dictates that if a single factor emerges from factor analysis, or
if one general factor will account for a majority of the covariance among the variables, we need to assume that there’s significant common method variance in our data set. To check for this, we ran PAF and PCA using Varimax rotation and one unrotated PCA forcing everything on to a single factor. The unrotated version led to a single factor explaining only 16.9% of total variance. The PAF and PCA using Varimax rotation were similar to our integrated factor analysis discussed previously and let to 9 factors, of which 7 clearly defined. The first factor explained 10.26% percent of total variance.
We then introduced a common latent factor into our CFA model to test for common method variance. According to Podsakoff et al. (2013) we should use the new model and compare it with our original model to find common variance by completing two steps; The first is to study the differences between the loadings on latent variables for all the items and compare them to the original model. After adding a common latent factor to each of the items, we find that the largest difference is on an item from the Organizational Capital scale (Δ! = 0.09). We then completed the second step by squaring the estimate made for our common latent factor (0.11). By doing this we get the common variance that can be attributed to this common latent factor (1.21%). These results let us to believe that our items are not subjected to a common method bias.
Even though our total sample size should be enough for up to 6 predictors if we are looking for a medium size effect Cohen (1988) and Field (2015), it is important to test all our independent variables for multivariate normality before we start our analysis. It may also provide more support for our use of the principal factor methods we used in our explanatory factor analysis, as principal factor analysis is required if the assumption of multivariate
normality is violated (Fabrigar et al., 1999). To test multivariate skewness and kurtosis we tested all our independent variables using Mardia’s test (1970,1974) for multivariate skewness (normalized = 61.8, p = .70) and Small’s test of multivariate normality (VQ3 = 22.38, p = .07) and found that they were not significant (p < .05). Then we also determined the critical values for the Mahalobis distances (F(7,45) = 20.59) using methods described by Mahalobis (1936) and found no single case to have a greater distance than the upper limit. We therefore have found no evidence that the assumption of multivariate normality is violated (DeCarlo, 1997).
Finally, to account for possible response bias due to the long running time of our survey, we measured early and late response bias. We do this by running t-tests on all of our independent and dependent variables as well as some of our control variables. By dividing the sample into two groups by order of responses and testing for differences, we found no significant differences on any of the variables. The only variable that came close was Exploration (p = 0.08). To conclude, there are some issues with our current sample and model if we look at the outcome of the integrated factor analysis. However, we have found no evidence of common method and response bias and no evidence of common method variance through our analysis. Besides that, we have tried to provide evidence of the reliability and validity of our model and have found no evidence that the multivariate normality assumption has been broken. We believe that our confirmatory factor analysis also showed an adequate fit if we consider our sample size and the requirements for conducting a CFA on a theoretical model.
