DOES SMART ALLIANCE GOVERNANCE CREATE SMART FIRMS?

(1)

DOES SMART ALLIANCE GOVERNANCE

CREATE SMART FIRMS?

The influence of governance forms on the absorptive capacity in a

buyer-supplier relationship

by

JOHANNES JACOB NIELS SMID

University of Groningen Faculty of Economics and Business

(2)

Abstract

In the increasingly globalizing world of nowadays it is of crucial importance to be ahead of the game. This can be accomplished by the formation of alliances, as they foster knowledge sharing and allow for a focus on core competencies. Yet, the downside of alliances is that they increase risk, that has to be counteracted by the use of the governance modes “control” and “trust”. Opinions in literature on how these governance modes are related are diverse and even contradicting. Therefore the present study aims to enhance the understanding of this relationship between control and trust. The view adopted in this process is the more recently developed view which claims the relationship depends on the situational factors it is used in. The influence of the following factors was analyzed: absorptive capacity, market turbulence, and being a buyer or supplier in a buyer-supplier relationship. This study uses data from 166 dyadic buyer-supplier relationships that was gathered using questionnaires, and which was analyzed with PLS-SEM in order to test the hypotheses that were formed. The results showed a positive effect of trust on absorptive capacity for both the buyer and supplier data, whereas control only had a positive effect on absorptive capacity for the buyer data. Furthermore, no interaction effect was found for the use of control and trust, indicating that support is lacking for both a negative and a positive effect of complementary using control and trust. Lastly, market turbulence did not seem to influence the effect of control and trust on absorptive capacity.

Keywords: control, trust, absorptive capacity, market turbulence, alliance governance, knowledge,

(3)

Preface

Hereby I present to you my thesis “Does smart firm alliance create smart firms?”. The aim of this paper is to develop more inside on how governance modes influence the absorptive capacity. It was written as a final step in obtaining my Master of Science degree in Marketing Management, at the University of Groningen. The period in which this work has been produced is from the end of August 2015 till half way January 2016.

At the heart of this paper lies work previously done by my supervisor, dr. J. Berger. The data he gathered by the use of questionnaires, is the data that is used to test the hypotheses that are formed during the literature review in this paper. This, however, made the process of writing the thesis sometimes a bit of a challenge. I had to adjust my literature review and conceptual model several times, as limitations in the data did not always allow for my (sometimes overcomplicated) research ideas. Adding to the challenge was the fact that I made use of a statistical program that was the most suitable in this situation, but that I was not familiar with. It took a lot of extra research, exploring, and sometimes frustration in order to get the hang of the program, and thereby produce good results. Though, in the end it was an experience that taught me a lot, and I think it resulted in a strong and interesting final thesis.

As some people have been of great help, mainly in the more challenging moments, I would like to use this opportunity to thank them. First of all I would like to thank my supervisor dr. J. Berger for his help and guidance during the whole process of writing this thesis. If I was stuck or had questions, he always took the time to help me out, and try to make me see other (most of the time better) options and solutions. On top of that I would like to thank him for always looking for possible improvements in my work. Even though I did not always like it at that moment, it led to a significant increase in the quality of my thesis.

Apart from my supervisor, I would like to thank the rest of my fellow students in the same thesis group. It was of great help that it was possible to contact them for questions I needed an answer to. We often came to solutions for mutual problems together, by having contact and exchanging ideas, making the process a lot easier sometimes.

I hope you enjoy reading my work, and hopefully find it useful.

(4)

List of Tables

Table 1 - Reliability and convergent validity reflective scales ... 37

Table 2 - Discriminant validity reflective scales ... 38

Table 3 - Assessment formative scales ... 40

Table 4 - Results of the independent sample t-test between buyers and suppliers ... 41

Table 5 - R-squared (R²) and Stone-Geisser's (Q²) of endogenous latent variables ... 42

Table 6 - Direct effects ... 42

Table 8 - Moderation effects ... 42

Table 9 - Differences high and low trust ... 44

Table 10 - Overview of supported and rejected hypotheses ... 45

List of Figures

Figure 1 - Cost of control vs. cost of relational risk (Colletti et al, 2005) ... 10

Figure 2 - Risks and governance style (De Man & Roijakkers, 2009) ... 15

(6)

6

1 Introduction

Globalization has led to a substantial increase in competition for companies, which puts extra emphasis on having to be on top of your game. This is beautifully described by the words of Nooteboom, Berger, and Noorderhaven (1997) whom state that globalization is one of the causes that “made market competition increasingly like a race”. According to them in order to win this race, companies need to make use of alliances, because they allow a company to 1) share fixed cost and the risks of developments, 2) strengthen their core competencies, 3) access complementary competencies, 4) increase the speed of market entry. An alliance is an interfirm cooperation that goes beyond a market transaction and can come in different forms, like long term purchase agreements, licensing, joint ventures and collaboration on R&D (Nooteboom, 1996).

This notion of increased importance of alliances is not limited to the confines of literature, but also visible in practice, as an increasing amount of companies start to see the importance of working together with other parties (Coletti, Sedatole & Towry 2005; Das & Kumar, 2009). This has resulted in a strong increase in the amount of alliances that have been formed over the last decades (Tomkins, 2001). A good example of a successful alliance, is the one formed between Phillips and Douwe Egberts in the development of the Senseo coffee machine (De Man & Roijakkers, 2009).

However, alliances are not always successful, as according to research Gill and Butler (2003) fifty percent of the formed alliances fail because they do not fulfil the expected results and goals set by the parties. What is the reason for this high failure rate and what does it mean in terms of actions that managers can take, to prevent failing of an alliance?

According to literature, the fact that alliances increase the levels of relational risk and performance risk is one of the main reasons of this alliance failure. Relational risk is the risk that the other party does not act in favour of the alliance (Coletti et al 2005), and is often caused by an asymmetry between the goals an interest of the different parties in the alliance (Langfield-Smith 2008). This asymmetry between the goals and interest of the different parties can lead to opportunistic

(7)

7 Nevertheless, managers can counteract this increased level of risk by using two government

instruments: control and trust (Ring and Van de Ven, 1992; Das and Teng, 1996, 1998; Langfield-Smith, 2008). Control is the more formal way of controlling the other party by the formation of contracts and rules, whereas trust is the more social way by forming expectations about how the other party is going to act, and then behave according to those expectations (Bijlsma-Frankema & Costa, 2005).

The relationship between the two governance modes is heavily disputed in literature (De Man and Roijakkers, 2009), with some researchers claiming the two are complementary and others that they are substitutional (Woolthuis, Hillebrand & Nooteboom, 2005; Bachmann, 2001; Das and Teng, 2001). Next to those more traditional views, some of the more recent work on the relationship between the two argues that the relationship is more complicated than described in previous research (Tomkins, 2001; De Man & Roijakkers, 2009).

(8)

8 ACAP. Meaning that if an influence is found, a company can use this information to use control and trust in a way that fosters ACAP, and thereby possibly gain a competitive advantage.

The second situational factor that will be incorporated in the research is market turbulence. In a turbulent market the performance risk is high, as it is hard to predict the future state of the market (Langfield-Smith, 2008). It was indicated already by De Man and Roijakkers (2009) that, in cases of high performance risk, trust is the more important governance mode for an alliance. In this paper however, market turbulence will be linked to the influence of control and trust on ACAP. Thereby testing how market turbulence is affecting the relationship between the two topics.

Lastly, an important contribution of this paper is that the hypotheses formed later in this paper, will be tested using data of both sides of buyer-supplier relationships. By using both sides of the dyad this paper can add to the literature on understanding how buyers and suppliers differ in their perception of the relationship, which is of crucial importance (Oosterhuis, Molleman & van der Vaart, 2013). Brattström and Richtnér (2014) already mentioned that there is a lack of studies which focus on both sides of the buyer-supplier relationship when it comes to control and trust. A crucial reason however for all research to use both sides of the relationship is related to social identification, which is rooted in social capital, and implies that a person can identify itself better with its own firm. Therefore respondents will always answer in a somewhat biased manner (Steinman, Deshpande and Farley, 2000). In order to compensate for this bias it is important to include both the buyer and the supplier. However, this is not the case in most research done so far. This makes that it is unclear how buyers and suppliers differ in their perception of the relationship, although this could have serious

implications (Oosterhuis et al, 2013). Hence, this paper will also contribute to the literature on the differences between buyer’s and supplier’s perception, as both sides of the dyad will be included. The results will be checked for differences in outcome, determining whether control and trust should be used differently for the buyer or supplier.

(9)

9 questions will be answered based on the results, the research implications, the managerial

implications, recommendations for future research, and limitations of the research.

2 Theoretical Framework

In this section of the paper, existing literature on the topics of control, trust and the relationship between the two will be reviewed. Furthermore, literature on absorptive capacity and market turbulence will be analysed.

During the literature review the focus was on finding differences, disagreements and gaps, to form a clear view of where the problems concerning the topics are. Another objective was to find clear theories and definitions in existing literature on the topics, to be able to define the different topics. This way a clear understanding of the topics will be developed, combined with creating a view on how the topics might be related to, and influenced by one another.

Control

Control, is according to literature, one of the two possible instruments management can use to govern an inter-organizational relationship and manage its risk (Ring and Van de Ven, 1992; Das and Teng, 1996, 1998; Langfield-Smith, 2008). It is the more classical way of dealing with risk in an alliance and has mainly been the focus in more dated research. According to this more dated research, control was often seen as the only way managers were able to deal with the risk in an alliance (Bijlsma-Frankema & Costa; 2005). The view of trust also being a possible instrument for this, was developed in more recent research (ibid.). Using control as a way of lowering the risk finds its roots in transaction cost economics, and focuses mainly on lowering the amount of relational risk in an alliance (Williamson; 1975, 1985, 1996). Researchers supporting the control view consider the relational risk that is present in an alliance to be high. As according to them, although the amount varies per situation, every partner wants to act opportunistically. Possible opportunistic behaviour has to be counteracted by the institution of control, so partners will not use the alliance just for their own benefit (De Man & Roijakkers; 2009). Control reduces this risk by making the behaviour of both parties more predictable, through the institution of standards and rules (Leifer & Mills, 1996). These standards and rules are designed to steer the behaviour of the parties in the favourable direction and in that way, make the behaviour of the partners more predictable and reliable (Bijlsma-Frankema & Costa, 2005).

(10)

10 given state of the world has been realized or maintained (feedback, checking) and b) dealing with the possible deviations and unforeseen events in order to positively cope with them (intervention)” (2000: 809). This definition highlights the fact that control is not only about the institution of standards and rules, but also about checking whether those standards and rules are being lived up to, and if not, dealing with it in the predetermined and proper way. So, control is not a passive way of decreasing relational risk by just installing standards and rules; instead, it takes a more active role by checking and if necessary, intervening.

One of the most common ways to set standards and rules is based in transaction cost theory, and is in the form of contracts (Poppo & Zenger, 2002 and Williamson, 1975). This paper will use

contracting as the operationalization of control, because it is the most common way of exercising control (Poppo and Zenger, 2002). A contract forms an incentive to perform future actions in a certain desired way, because they are explicitly defined in the contracts and by signing a contract, cooperation among those defined actions is promised (Woolthuis et al, 2005). Apart from desired actions, a contract also prescribes the connected disciplinary measures that will be taken in case of violation of the contract (Poppo & Zenger, 2002).

The agency theory presents us with useful insight regarding the most ideal level of control, based on cost of control versus potential cost of opportunistic behaviour. The relationship between the cost of control and the cost of opportunistic behaviour is represented in figure 1, which comes from the paper of Coletti et al (2005). The theory behind this figure is that, with an increase in cost (or investments) of control, the cost of opportunistic behaviour will decrease. Where both lines cross each other, the total cost are at their lowest point, and this is the ideal amount of control in an alliance. This means however, that according to this theory there will always be a certain level of

relational risk, because controlling all the risk will be too expensive, as these costs will be higher than the cost of the risk itself (Coletti et al, 2005). Control has been

described in literature as less suitable to use in case of high performance risk. High performance risk is associated with high uncertainty in the market. Dealing with high uncertainty using control is very challenging, as there are a lot of possible turns and changes the environment can take. Controlling for all these possible outcomes is hard and would

(11)

11 make an extremely complex control structure. Contracts for instance (one of the most used forms of control), would have to become very specific and detailed, resulting in an inflexible company which is very unfavourable in an environment with high performance risk (Griffith & Yanhui, 2015). So with respect to performance risk, taking the agency theory into account, extremely high costs are linked to these complete contracts, thereby offsetting the contracts’ reduced costs of risk.

Two assumptions lying at the basis of control make it a less effective way of government in the fast changing environment of nowadays. The first assumption is that all actions that lead to success can be determined beforehand, so standards and rules can be set. The second assumption is that it is possible and necessary for both parties to have close interactions in order to monitor the behaviour of each other. Both assumptions however are becoming less likely to occur in the fast changing and more turbulent environment of today. The more turbulent environment makes it harder to predict successful actions, and the effects of globalization and virtualization for instance make that

interactions become more distant, making monitoring more of a challenge (Bijlsma-Frankema & Costa, 2005).

Therefore, in more recent literature, trust has become more important as a way of dealing with uncertainty in the more turbulent environment that is forming now.

Trust

Trust is the second way to reduce the risk perceived by both parties, and has shown to be a good indicator of the success of an alliance (Zaheer et al. 1998; Zaheer & Venkatraman 1995). As indicated before, alliances can be quite unstable and have a high failure rate, which according to literature is often caused by a lack of trust between partners, making it an important area of research (Coletti et al, 2005).

(12)

12 In contrast to this more conservative view of trust, there is the view that trust is different from control, and this view is gaining support in recent literature (Emsley & Kidon, 2007). This idea of looking at trust in a way that it is different from control, and perceiving it as a reliable way of safeguarding against opportunistic behavior, comes from social sciences (Woolthuis et al, 2005). Supporters of this view state that trust is a way of dealing with uncertainty, without having to use control and therefore, it cannot be a form of control (Tomkins, 2001; Leifer & Mills, 1996). The debate on whether or not control and trust are different is also clearly visible in the article of Groot and Merchant (2000). They initially state that trust is indeed a form of control, but later on in their article they advocate that trust and control are different subjects. This paper adopts the view that control and trust are distinctly different ways of dealing with risk, which builds upon the more recent view on the subject.

Even though a lot of research has been done on trust over the last few decades, there is still no uniform definition. The economic view on trust is based on calculating the probability that your trust is being betrayed by selfish behavior of the partner. The psychological view on trust is based on the characteristics of the trustor and trustee. Finally, the sociological view considers trust as something that is embedded in society. In previous literature, there seems to be a general acceptance of two elements of trust, namely the presence of positive expectations and the willingness to become vulnerable (Bijlsma-Frankema & Costa, 2005). Both of these element are present in the widely used definition of Rousseau et al (1998, p 135) who tried to incorporate the different views on trust and defined it as “a psychological state comprising the intention to accept vulnerability based on positive expectations of the intentions or the behaviour of another”. It implies that trust is based on the fact that the behaviour of others can harm the other party, and that it is related to risk taking. Instead of steering the behaviour of the other party by the institution of rules and procedures, as is being done with control, the company tries to predict the behaviour of the other party. And that enables a company to take a leap of faith which would not be possible by just using reasoning and control (Lewis and Weigert, 1985; Möllering, 2005). So trusting someone, means that you are willing to act on your beliefs that someone else will not behave opportunistically. Consequently, Rousseau’s defenition is adjusted by incorporating the concept of ‘acting on your expectations’, resulting in the following definition: “Trust is behaving in a congruent way with the positive expectations and beliefs one has about the intentions or behaviour of another, and thereby accepting vulnerability to possible opportunistic behaviour”.

(13)

13 between groups is the actual sum of the trust which the individuals have in one another (Zaheer et al., 1998).

Secondly, it is also important to differentiate between the two different dimensions of trust often used in literature, being competency trust and goodwill trust (Ring and Van de Ven, 1992;

Nooteboom, 1996). Although more forms of trust exist, these two forms of trust are the most

studied and used in literature (Dekker, 2004). Competence trust is about the trust that one has in the ability, competencies, and resources of the other party to get the desired results for the alliance. Goodwill trust, conversely, is more emotionally grounded and assumes that the other party will not behave in an opportunistic way, but instead tries to accomplish work to reach the goals of the alliance (Marzagalli, 2005). However, this paper only measures goodwill trust in determining the relationship between control and trust, but this will be further elaborated on in the methodology part. Still it is important to make note of the two kinds of trust, since they could possibly play a role in explaining the results.

Third, it is important to note that trust is not static, in a way that the level of trust can differ over situations and time. This means that if you trust someone at this point in time, the level of trust can be increased or decreased as a consequence of the interaction between two parties. If the trust in the other party turns out to be in place, because the other party behaves according to the

expectations and in favour of the alliance, then trust will increase. The other way around however, when the other party does not live up to the expectations, trust will decrease or the alliance will be stopped (Bijlsma-Frankema and Costa, 2005).

Fourth, trust is the cheaper way of the two ways available to managers in dealing with

perceived risk. So, in a way it is protecting a company against high transaction costs, which are related to control (Marzagalli, 2005). The downside of trust on the other hand, is that if one party is trusting the other party too much, it leaves a company vulnerable to opportunistic behaviour. Hence, it is necessary to make a good indication of the optimal level of trust that is needed in a relationship (Marzagalli, 2005).

The relationship between control and trust

(14)

14 control and trust which are distinctly different, being complementary, substitution and a more complicated view.

First this paper will look into the complementary view, which is grounded in the transaction cost economics theory and contract theory. According to this view, a higher level of control enables for a higher level of trust as well, while it limits the opportunities and incentives for opportunism. Because of this limitation, by the use of sanctions on the possibility to behave opportunistically, it is easier for different parties to trust one another (Woolthuis et al, 2005). This is based on the fact that control delivers some form of basic assurance that the partner will behave as is intended by the alliance (Poppo & Zenger 2002).

It is being argued however, that the trust resulting from this reasoning does not produce real trust, but only forms so called ‘thin’ trust. This form of trust is only based on the compensation of the negative expectations about the behaviour of the other party and does not include the development of positive expectations about the other party’s behaviour, which is necessary for real, or ‘thick’ trust (Vosselman & Meer-Kooistra, 2009). Sitkin (1995) poses another view, on why the trust and control could be enhancing each other. According to him, control provides parties with clear guidelines to judge the extent to which the other party’s behaviour is in favour of the alliance. This makes the judgement, on whether trust in the other party was justified, easier, and thereby it enhances the possibility that the level of trust in the partner will increase in the coming period of time. Evidence for this complementary view, has mainly been found in dynamic markets (De Man & Roijakkers, 2009).

(15)

15 trust reduces the need for control.

What can be concluded from the previous two views, is that there is empirical and theoretical evidence for both ideas of how trust and control are related. This means that both views could be right, and that maybe it is not as easy as them being always substitutes or always complementary. This idea, of the relationship between control and trust being more complicated than indicated in previous research, is what the third view in literature is about. According to this view, the

relationship between the two cannot always be the same, but it depends on situational factors whether they should be used as substitutes or complements. Tomkins (2001) for instance, is a

supporter of this view, and he argues that the relationship between control and trust is depending on the stage of the alliance. According to him, in early stages of the alliance the amount and the form of the information needed differs from the amount and form of information which is needed in later stages of the alliance. As control and trust influence the way how information is distributed and created within the alliance, they should be used to facilitate the specific information need. This means that depending on the stage the alliance is in, control and trust can be used substitutional or complementary (Tomkins, 2001).

Another paper supporting this view, is that of De Man and Roijakkers (2009) who argue that control and trust can both be complementary and substitutes. According to their article, the way control and trust should be used in a relationship depends on the forms and levels of risks that are present in the alliance. They use the same classification of the risks in an alliance as this paper, being performance risk and relational risk. High levels of relational risk combined with low performance risk should, according to them, lead to the use of mainly control as a way of dealing with the risk (substitute view). High levels of performance risk combined with low levels of relational risk on the other hand, will lead to the use of mainly trust (substitute view). However, when both forms of risk are high, control and trust should both be used (complementary view). The full quadrant of levels of relational risks and the way of dealing with this risk is summarized in figure 2.

In this paper, this third and more recent developed view was adopted, which claims the relationship between control and trust being a more complex relationship than anticipated by researchers in the past. More specifically, this paper will look at the relationship between trust and control, in the light of two different possible influences. The first subject that is analysed on how it might determine the way control and

trust should be used, is absorptive

(16)

16 capacity. Absorptive capacity will be further elaborated on in the next paragraph. The second

possible influence on the relationship between control and trust is that of market turbulence, and will be discussed in paragraph 2.5. The possibility that there is an influence of market turbulence on control and trust, is already highlighted in the work of De Man and Roijakkers (2009). But first of all the influence of the absorptive capacity will be assessed, which will be the topic of the next

paragraph.

Absorptive Capacity

In the last twenty years, knowledge has been increasingly accepted as a crucial source of competitive advantage (Berger, 2015). Resources based on knowledge are usually tacit, sticky, and hard to codify, which makes them difficult to copy for opponents (Jap, 1999). An example of a source of knowledge could be the suppliers that the firm is doing business with (Dyer and Sing, 1998). By forming a buyer-supplier relationship, the buyer and the buyer-supplier can share their knowledge with one another. This can result in an increase in available knowledge and therefore creates a possibility to generate extra value with this obtained knowledge (Palmatier, Dant and Grewal, 2007).

But even when this new knowledge is available, it is sometimes hard for companies to implement it into their own business and create a competitive advantage out of the knowledge (Hult et al, 2000). In order to enhance the understanding of why the implementation of external knowledge is

successful for some companies and unsuccessful for others, the concept of absorptive capacity (ACAP) was introduced to the literature on organizational learning (Revilla et al, 2013).

ACAP as it is used now, at an organizational level, found its origin in the 1980s in literature regarding the learning process of organizations (Volberda, Foss, and Lyles 2010). In past research, a lot of different levels of analysis and measures, have led to a high diversity in definitions of ACAP. In all these different views however, there seems to be an agreement on the aspect of ACAP as a set of firm’s abilities to manage knowledge (Zahra & George, 2002). From all these definitions, the one of Cohen and Levinthal (1990, p128) seems to be the most cited and accepted, defining ACAP as “the firm its ability to recognize the value of new, external knowledge, assimilate it, and apply it to commercial ends”.

(17)

17 The four capabilities have been defined by Zahra and George (2002) as follows: the first capability,

acquisition, implicates the firm’s ability to pinpoint and gather knowledge from external sources

which is crucial to its operations. Secondly, assimilation is the firm’s routines and processes, which enable it to evaluate, process and comprehend the externally acquired knowledge. The third capability, transformation, represents the competence of a firm to create and refine the practices that enable it to combine existing knowledge with the new knowledge that has been acquired and assimilated in the previous two capabilities. The last capability, exploitation, is all about the routines which enable a firm to use its existing competencies in new and refined ways, or form new

competencies using the newly acquired and assimilated knowledge.

The first two capabilities, acquisition and assimilation, together form the potential absorptive capacity (PACAP) of a firm, which corresponds to explorative learning. And the last two capabilities, transformation and exploitation, together form the realized absorptive capacity (RACAP) of a firm, which can be compared with exploitative learning. Even though PACAP and RACAP are different by definition, Zahra and George (2002) state that in practice both forms of absorptive capacity are present at all times, with each of them fulfilling a necessary but insufficient condition in improving the firm’s performance.

(18)

18 learning (PACAP) means that no new knowledge is acquired to exploite (Benner & Tushman, 2003, and Levinthal and March 1993).

The second view which is present in literature is punctuated equilibrium, contradicting the idea that PACAP and RACAP both have to be present at all times. According to supporters of the punctuated equilibrium, PACAP and RACAP form iterating sequences of long periods in which firms have to be focussed on maximizing and enhancing PACAP, followed by short periods where the main focus should be on maximizing and enhancing RACAP (Gupta et al, 2006).

This shows that opinions vary on how PACAP and RACAP should exist within a firm, as is the case with control and trust. Even more resemblance can be found in the opinions on how PACAP and RACAP are related and in the different views on the relationship between control and trust. The

ambidexterity theory for instance, argues that PACAP and RACAP should coexist as they enhance each other, where the complementarity view also states that control and trust should be used together as one strengthens the other (Gupta et al, 2006; Sitkin, 1995; Woolthuis et al, 2005). On the other hand, the punctuated equilibrium theory is similar to the substitute view regarding control and trust. In both views, the two concepts (PACAP and RACAP/control and trust) are too different from one another, resulting in the understanding that they should not be present together at the same time (Das and Teng, 200; Gupta et al, 2006; Sitkin and Roth, 1993).

These similarities exist in the theoretical background of the theories, but more practical links can also be found between the concepts. Exploratory learning, which corresponds to PACAP, has been linked to firms being flexible, decentralized, and having loose cultures (Jansen et al, 2009). This makes sense, while in order to acquire and assimilate new knowledge, it is important to have informal networks in place, which are easier established under more flexible conditions (Tortoriello, 2015). Using trust as a governance method makes an alliance more flexible (Griffith & Yanhui, 2015), meaning that an alliance that uses trust as a governance mode will, therefore, enhance its PACAP. However, it is stated in literature that high levels of trust makes that both parties are more reluctant to share negative information (Skinner, Dietz, & Weibel, 2014), are less creative as both parties start thinking too much alike, and have a lower intent to search for new knowledge. This is referred to in literature as ‘the dark side of trust’ (Selnes & Sallis, 2003). Those effects of high levels of trust lead to lower PACAP, as the ability of assimilation and acquiring of new knowledge is what PACAP is all about (Zahra and George, 2002). This leads to the formation of the next hypothesis:

H1: Trust is positively related to the level of PACAP within an alliance, with this relationship becoming

(19)

19 On the other side, the usage of control to govern a relationship makes it less flexible and more formal (Griffith & Yanhui, 2015). As PACAP is enhanced by flexibility and a loose culture in the alliance (Jansen et al, 2009), alliances which use control to govern the relationship, will decrease their ACAP accordingly. Consequently, the next hypothesis can be formulated:

H2: Control is negatively related to the level of PACAP within an alliance.

The opposite can be found for exploitative learning, which corresponds to RACAP, as it is often associated with firms being efficient, centralized, and having a tight culture (Jansen et al, 2009). This results in more formalization within the company, which makes it easier to retrieve information that has already been internalized, and it enhances the understanding of sets of tasks within units (Jansen et al, 2005). As a result, formalization can help with identifying opportunities for the transformation of external knowledge. Moreover, formalization “codifies best practices”, making knowledge “more efficient to exploit, easier to apply and faster to implement (Berger, 2015: 87). A more formal and tight cultured alliance can be created by the use of control in governing the alliance (Griffith & Yanhui, 2015), meaning that an alliance which uses control as a governance method, thereby enhances its RACAP, leading to the formation of the next hypothesis:

H3: Control is positively related to the level of RACAP within an alliance.

Using trust as the government method on the other hand, makes the alliance more flexible and creates a looser culture (Griffith & Yanhui, 2015). As RACAP is associated with more tight and structured cultures (Jansen et al, 2009), alliances in which trust is being used as an instrument to govern the relationship will, as a result of that, negatively influence its RACAP.

However, due to trust, information will be shared more easily and creativity will be enhanced, meaning it can also be argued that trust will enhance RACAP. This while knowledge transformation and exploitation is enhanced by sharing knowledge as it creates understanding between the two parties (Zahra & George, 2002). Furthermore, it can be reasoned that as transformation and exploitation are about finding new ways to use newly acquired and already existing knowledge (Zahra & George, 2002), the increased creativity will lead to create more and easier recombination of resources thereby positively influencing RACAP (Atuahene-Gima, & Murray, 2007). This last line of reasoning is the one that will be adapted in this paper, meaning it is expected that trust will also have a positive result on RACAP, even though there might also be negative influences.

(20)

20 most likely resulting in firms being less capable to use the newly acquired knowledge to create new competencies and practices, which is what RACAP is about (Zahra & George, 2002). Accordingly, the following hypothesis is being formed:

H4: Trust is positively related to the level of RACAP within an alliance, with this positive effect

becoming stronger for higher levels of trust.

Now that the individual effects of trust and control on the PACAP dimensions have been hypothesized, the next question is how control and trust should be used in order to optimally enhance the PACAP and RACAP of the alliance. How do control and trust relate to the enhancement of PACAP and RACAP? Should they be used in a substitutional or complementary way? Finding an answer to this could also shed more light on the theories of ambidexterity and punctuated equilibrium regarding the presence of PACAP and RACAP within an alliance.

As it hypothesized that PACAP is enhanced by the use of trust and weakened by the use of control, while RACAP is being enhanced by both governance methods (Jansen et al, 2009), it becomes clear that they need different governance modes in order to be strengthened. Therefore, using control and trust in a complementary way, would result in the RACAP dimension receiving the positive influences, while PACAP will receive both a positive and a negative influence. This will result in a bottom line effect of about zero for PACAP, thereby only increasing one dimension, while not enhancing the other.

This might be an argument in favour of punctuated equilibrium, as even though there are two distinct theories on how PACAP and RACAP should be present within a firm, there is acceptance in both theories on the fact that both dimensions should be maximized as they build on each other (Burgelman, 2002; Benner and Tushman, 2001; Choi & Lee, 2015). In case of a punctuated equilibrium between PACAP and RACAP, this would mean that in times of focus on PACAP, trust should be used and control should only be incorporated when the focus is on RACAP to get the best results (Gupta et al, 2006). This would result in each dimension only receiving the positive influence as only the governance mode(s) that are positively related to the dimension are used. This effect will be further strengthened by the fact that, as the dimensions build on each other, maximized

enhancement of one dimension leads to the enhancement of the other dimension as well. This implies that using both control and trust at the same time would not lead to the most optimal result, as control lowers the positive effect of trust for the PACAP dimension, thereby leading to suboptimal results for both dimensions as they build on one another.

Architectural ambidexterity seems to dodge this problem by creating sub units, with each sub unit using the governance mode(s) most suitable for one ACAP dimension that thus will only be

(21)

21 firm (Benner & Tushman, 2001). Because alliances already bring increased risk to the relationship, which form governance difficulties (Das and Teng 1996, 1999, 2001), it is most likely unadvisable to use architectural ambidexterity in an alliance, as it creates even more difficulties that have to be overcome.

So, according to this reasoning, it can be stated that using control and trust in a complementary way would not give the optimum bottom line result. However, it has been hypothesized earlier on in this paper that higher levels of trust would lead to less positive effects on PACAP and RACAP, due to the fact of ‘the dark side of trust (Selnes & Sallis, 2003; Skinner, Dietz, & Weibel, 2014). This means that offseting the risk in a relationship, only by high levels of trust would lead to a stagnating positive effect on PACAP and RACAP. On top of that, as both dimension build on one another, the diminishing positive effect of trust would, in the end, also lead to a decrease in growth of PACAP and RACAP (Burgelman, 2002; Benner and Tushman, 2001; Choi & Lee, 2015). But still, as the effects will be stagnating, and not positive or negative as in the case with using control on RACAP, it seems advisable to use control and trust in a substitutional way to enhance both PACAP and RACAP. In accordance to this the following hypotheses were formed:

H5: Using control and trust in a complementary way, will lead to a less positive influence on PACAP

than using control and trust substitutional.

H6: Using control and trust in a complementary way, will lead to a less positive influence on RACAP

It is however questionable whether it is possible in a practical sense to switch between control and trust from one moment to the other. Trust is something that has to develop over time and cannot just be turned on or off, making it hard to ‘choose’ it as a governance mode (Bijlsma-Frankema and Costa, 2005). Though in this paper, for theoretical purposes, the view will be adopted that it would be possible to switch between governance modes. But when concluding on the results it is important to keep in mind that this switching might not be the case.

In the next paragraph, the possible influence of market turbulence on the effect of control and trust on PACAP and RACAP is the subject of analysis.

Market turbulence

Literature on dealing with a turbulent market has always been an important topic of interest for managers. Especially in the increasingly turbulent market of nowadays, it is of great interest to managers that guidance for these turbulent times is developed (Calantone et al, 2003), while

(22)

22 Van der Stede, 2003). This is caused by high variation of needs between customer, the constant and frequent changes in technologies, customer needs, and product demands (Roberts, 2014) but also by possible changes in laws and the political environment for instance (Merchant and Van der Stede, 2003). An important implication of this market turbulence is that current products quickly become obsolete, since new and more innovative products are wanted by consumers (Sorensen & Stuart, 2000).

Linking market turbulence to control and trust is the fact that in turbulent markets the performance risk is high (Langfield-Smith, 2008). As is stated earlier in this paper, performance risk is the

possibility that the goals of the alliance will not be met even though both parties fully cooperate with each other and this performance risk mainly comes from market forces (Das and Teng 1996, 1999, 2001). According to literature, management has two ways to deal with performance risk (and relational risks), namely control and trust (Ring and Van de Ven, 1992; Das and Teng, 1996, 1998; Langfield-Smith, 2008).

A connection between market turbulence and ACAP on the other hand is not that evident in

literature, as in past research the role and impact of the environment on absorptive capacity tends to be discarded or underestimated (D׳Souza, & Kulkarni, 2015). But how does market turbulence influence control and trust, and thereby PACAP and RACAP?

Employing control to deal with the high performance risk that is associated with market turbulence demands very explicit and detailed control system as they have to control for every possible outcome that can result from the turbulent market (Griffith & Yanhui, 2015). Figure 1 coming from the article of Coletti et al (2005) that is depicted earlier in this paper, illustrated the problem with these

complex control structures. The article looks at the effect of control on relational risk, but in line with the article of Griffith and Yanhui (2015) we can expect the same relation to be present for control and performance risk as higher levels of risk need to be counteracted with higher levels of control. As the cost of control grows exponentially and the cost of performance risk decreases inverse

exponentially this means that a trade-off has to be made between the cost of control and the cost of performance risk. This results in total cost composed of cost for the control system and cost of the remaining performance risk which will increase for higher levels of performance risk (Coletti et al., 2005). The use of trust on the other hand, which is a considerably cheaper way of dealing with risk, results in significantly lower total cost for coping with high levels of risk than the use of control (Marzagalli, 2005).

(23)

23 advantages which is done by the creation of resources that are rare, valuable, not substitutable, and hard to imitate (Barney, 1991). As ACAP is defined as a dynamic capability that is indeed rare, valuable, not substitutable, and hard to imitate it can form a competitive advantage (Berger, 2015). This means that the limited resources should be used as efficiently as possible, which is in the way that fosters the highest creation of rare, valuable, not substitutable, and hard to imitate resources which will result in competitive advantages (possible PACAP and RACAP).

As market turbulence makes the use of control more expensive (Coletti et al., 2005) it is unadvisable to allocate resources to the use of control in case of a turbulent market as the increased costs make it a less efficient tool for securing a competitive advantage. Allocation of limited resources to control would lower the resources allocated to other, more efficient means to create PACAP, RACAP and other competitive advantages (like trust). This decrease in resources allocated to other means is needed while, as said before, the resources available to gain competitive advantage are limited so an increase in the allocation to one mean has to be balanced by a decrease in another mean (Choi & Lee, 2015). The result of this less effective use of resources is the decline of PACAP, RACAP and other competitive advantages present in the alliance. This implies that the influences of control on both PACAP and RACAP will be negatively influenced by the level of market turbulence as it makes control a less effective allocation of the limited resources. Therefore the following hypotheses can be formulated:

H7: Market turbulence negatively moderates the relationship between control and PACAP

H8: Market turbulence negatively moderates the relationship between control and RACAP

Though it is important to note that the increased performance risk has to be dealt with somehow, as risk is the main reason for alliances to fail (Das and Teng, 1996, 1999, 2001). Hence, not

counteracting this performance risk using either control or trust would threaten the performance and survival of the alliance leading to the loss of competitive advantages achieved by the alliance (Coletti et al 2005; Langfield-Smith, 2008). This failure of the alliance would lead to an even bigger decrease in competitive advantage than if control is used to deal with the risk, meaning that the use of control despite it being inefficient would be justified. However, as stated before, trust is the alternative and cheaper way of dealing with high levels of risk (Marzagalli, 2005), making it a more efficient way of dealing with the risk than control.

(24)

24 PACAP, RACAP or other competitive advantages. This implicates that trust enhances the competitive advantages of the firms by both assuring the survival of the alliance by dealing with performance risk as well as through the increased availability of limited resources for other means that foster

competitive advantages.

This leads me to form the following hypotheses:

H9: Market turbulence positively moderates the relationship between trust and PACAP

H10: Market turbulence positively moderates the relationship between trust and RACAP

The finding that trust is more suitable than control in case of high performance risk is in accordance with findings in previous research like that of De Man and Roijakkers (2009). Linking this to the theory on PACAP and RACAP means that in periods of high market turbulence trust is more preferable to enhance the ACAP dimensions. As in more turbulent markets the focus is more on explorative learning (PACAP) and in low turbulence on exploitative learning (RACAP) this would indicate that PACAP should be enhanced using trust while RACAP could be enhanced by both options (Gupta et al, 2006; Levinthal & March, 1993; Mudambi & Swift, 2014). However, as RACAP is

negatively influenced by control, it seems that when incorporating market turbulence in the relationships between control and trust, and PACAP and RACAP trust is the preferable governance function in all situations.

Explorative and Exploitative learning performance

In this paragraph the relationship between PACAP and RACAP on learning performance will shortly be evaluated. Even though this relationship has been extensively tested and confirmed in existing work, it might still be important to test it in this paper to highlight the importance of PACAP and RACAP for management. Furthermore, it might also be useful to retest this relationship, as in this paper data of buyers and suppliers is gathered, making it possible to look for differences in this relationship between both sides of the dyad.

In previous literature, researchers already highlighted that ACAP is of crucial importance to the learning process of a firm (Cohen, & Levinthal, 1990; Lane, Koka, & Pathak, 2006; Gutiérrez, Bustinza, & Molina, 2012). In some research, ACAP is even defined as the ability to learn (Tsai, 2001). The ability of a firm to learn depends on the capability to locate important new knowledge, assimilate it, and use the new knowledge in its advantage (Cohen and Levinthal, 1990). As these capabilities are part of ACAP, it becomes clear that learning indeed depends on ACAP. This is in line with Kim (1998) who states that a firm its learning is a function of its ACAP.

(25)

25 2006), which according to the definition of Zahra and George (2002) is part of the PACAP dimension. This implies that if a firm its PACAP increases, its explorative learning performance will increase as well (Lichtenthaler, 2009). Exploitative learning performance on the other hand consists of the exploitation of knowledge (Lane et al, 2006; Lichtenthaler, 2009), thereby linking it to RACAP (Zahra and George, 2002). Accordingly, as a firm enhances its RACAP, the exploitative learning performance of the firm will increase.

In line with these findings the following hypotheses were formed:

H11: PACAP is positively related to explorative learning performance.

H12: RACAP is positively related to exploitative learning performance.

Conceptual Model

The literature review led to the following hypotheses:

H1: (a) Trust is positively related to the level of PACAP in an alliance, (b) with this positive effect

becoming weaker for higher levels of trust.

H2: Control is negatively related to the level of PACAP in an alliance.

H3: Control is positively related to the level of RACAP in an alliance.

H4: (a) Trust is positively related to the level of RACAP in an alliance, (b) with this positive effect

becoming weaker for higher levels of trust.

H5: Using control and trust in a complementary way will lead to a less positive influence on PACAP

H6: Using control and trust in a complementary way will lead to a less positive influence on RACAP

H7: Market turbulence negatively moderates the relationship between control and PACAP.

H8: Market turbulence negatively moderates the relationship between control and RACAP.

H9: Market turbulence positively moderates the relationship between trust and PACAP.

H10: Market turbulence positively moderates the relationship between trust and RACAP.

H11: PACAP is positively related to explorative learning performance.

(26)

26 Putting these hypotheses together, will form the following conceptual model:

Figure 3 - Conceptual model

(27)

27

3 Research Design

The goal of the analysis was to test the relationships described in the conceptual model, and test whether or not there is a difference between the buyer and supplier for this model. This because as mentioned in the introduction one of the aims of this paper is to test for differences in the

perceptions between the buyers and the suppliers while previous literature indicated that there might be a difference between how both sides experience the relationship (Brattström and Richtnér 2014; Oosterhuis et al, 2013). By using dyadic data on the relationship it is possible to test for these differences and learn where the differences in perceptions are. From that, possible implications of these differences can be determined which was done in the next two chapters of this paper. The dataset that was used in this paper came from work done by Berger (2015) on the subject of interfirm absorptive capacity. In their paper the main focus was on the dimension and drivers of interfirm absorptive capacity and the influence it had on ambidexterity, but constructs measured in that study are also applicable in this study.

Data Collection

For his paper, Berger (2015) collected data from both sides of buyer-supplier relationships, thereby preventing the presence of high levels of measurement errors. These high levels of measurement errors can be caused because a relationship is not directly observable (Berger, 2015). The first step they took to collect the data was inviting heads of purchasing and high-ranking technology managers from purchasing organizations and ask them to supply contact data of four persons in their

companies that were crucial to their customer relationships. As their main focus in the paper was on ACAP, the researchers only included manufacturing firms while they have a higher probability of systematically acquiring the external knowledge that is needed in ACAP (Arbussà, & Coenders, 2007). To create more variance in their data, businesses from the automotive, chemicals, machinery, pharmaceuticals, semiconductors, and electronics industry were used. The choice for this was made because it was described in existing literature that there were different knowledge strategies used in those industries (Lichtenhaler, & Ernst, 2007).

(28)

28 contacts supplied by the heads of purchasing and high ranking technical managers), who were all located in the Netherlands, were then interviewed according to a standardized questionnaire. Informants on the supplier side of the relationship (the contacts supplied by the buyer informants), who came from all over the world, received and returned the same standardized questionnaire through email. It was promised that all the collected data would be treated with complete

confidentiality, therefore only company codes were incorporated in the database to still be able to match both sides of the relationship. To monitor how involved and knowledgeable the respondents were, their currents working positions, the number of years they were engaged in the relationship and the time they spent on the relationship in percentages of their total working time. Using this procedure in the period between June 2011 and April 2013, data was collected from 166 matched-pair buyer-supplier relationships.

Measures

The questionnaire that was used to gather this data used scales that were already existing from previous research. Some of the scales were previously used in literature on buyer-supplier

relationships, while others were created in light of subjects like firm learning, intra-firm learning, and interfirm learning but not used in relation to a buyer-supplier relationship before. The use of these measures was extended to buyer-supplier relationships in the questionnaire of Berger (2015). Furthermore the study made use of first-order and second-order constructs, with first-order constructs being latent and having observable variables as indicators, and second-order constructs have first-order constructs as their indicators. An example is are the first-order constructs flexibility, information exchange, and solidarity that are measured by questions in the questionnaire, and together are the indicators for the second-order construct relational norms. Apart from the distinction between first and second-order constructs, the researchers also make a distinction between reflective and formative scales used in their study based on the causal link between the indicators and the latent construct. ‘Reflective indicators ‘reflect’ the latent construct; changes in the value of the latent construct determine changes in the formative indicator values of the indicator’ (Berger 2015 p. 91). This means that high levels of correlation can be expected amid reflective indicators. Formative indicators on the other hand have a causal link to the latent construct, so change in the level of the indicator causes change in the latent construct. Different formative indicators should measure different aspects of the latent construct and therefore are not related (Edwards, 2010).

(29)

29 the questionnaire was offered in both Dutch and English depending on the preference of the

respondent. Both versions were based on the English original, then translated in Dutch, and then back-translated to English again. Both version had been checked by people who were fluent in both languages and adjusted the Dutch questions where necessary to match the English version.

The scales used for all the questions were 7-point-likert-scales labelled from 1= strongly disagree to 7= strongly agree. Some scales however were labelled differently, like the questions on market turbulence related to differences in customers for example, which are labelled from 1= very similar to 7= very different. Also the scale for explorative learning was labelled different with 1= to no extent 7= to a great extent. The English version of the questionnaire is presented in Appendix 7.2.

Several different concepts and measures were taken from the paper of Berger (2015). Appendix 7.1 shows which constructs and measures this are, and also what source they came from.

The variable ‘control’ is measured by the concept of contracting and its related measures that were used in the paper of Berger (2015), who see it as ‘the focus and the degree of formal contractual control in the buyer-supplier relationship (p. 92). As indicated before, contracting is one of the most important forms of control according to previous research (Poppo and Zenger, 2002). Also in past research contracting is often used intertwined with control, using control and contracting to describe the same concept (Cao, & Lumineau, 2015; Dekker, 2004; Woolthuis et al, 2005). The measures for contracting were obtained by Berger (2015) from Desphande and Zaltman (1982), Cannon and Perreault (1999), Buvik, & Reve (2002), and Jansen, Van de Bosch, & Volberda (2006).

(30)

30 This is in accordance with the definition that is used in this paper were PACAP comprises of

acquisition and assimilation and RACAP of transformation and exploitation (Zahra & George, 2002). For acquisition the measures came from studies done by Szulanski (1996) and Camisón and Forés (2010) and for assimilation the measures came from work done by Jansen et al. (2005) and Camisón and Forés (2010), together forming PACAP. For transformation on the other side the measure came from work done by Bontis, Crossan, and Hulland (2002), and Cadiz, Sawyeer and Griffith (2009) and for exploitation the measure came from studies by Jansen et al. (2005), Cadiz et al. (2009), and Camisón and Forés (2010), together forming RACAP.

Market turbulence is also directly measured in the work of Berger (2015), and therefore his concept and measures are the same as the used in this paper. Market turbulence is a second order construct that is measured by the construct environmental dynamism and market diversity from the studies of Selnes and Sallis (2003), and Achrol and Stern (1988).

Explorative and exploitative learning are as in Berger (2015) his paper measured by a measure developed by Lane, Salk, and Lyles (2001) for explorative learning performance and a measure of Selnes and Sallis (2003) for exploitative learning performance.

As a control variable industry will be used as it is known from previous literature that there are differences in knowledge strategies between industries (Squire, Cousins, & Brown, 2009). High technological industries like the pharmaceutical industry for instance is known for its focus on knowledge transfer (Santos, 2003). The different industries were measured on the questionnaire in three different categories that were selected on prior literature highlighting the differences in knowledge strategies in these industries (Lichtenhaler & Ernst, 2007).

Statistical Procedures

Looking at the conceptual model in figure 1.3 an important thing to note is the moderation effect of market turbulence on four other relationships. Furthermore, there is also the interaction effect between control and trust on both PACAP and RACAP. In a moderating there is an independent variable (IV), dependent variable (DV) and a moderating (M) variable. To test for moderation, the effect of the IV on the DV, M on the DV, as well as the interaction effect of the IV and M on the DV need to be determined. In case there is a significant interaction effect between IV and M on the DV there is an interaction effect, meaning that the effect of IV and M on DV depends on the level of M (Baron & Kenny, 1986). In an equation this looks like: DV= b0 + b1*IV + b2*M + b3*(IV*M) + err, and to test this moderation a regression analysis could be used (Little, Card, Bovaird, Preacher, &

(31)

31 (Baron & Kenny, 1986). So if this assumption is violated this means that the results of the analysis will be biased. This measurement error related issue is problematic in all regression analyses, though it is especially precarious in the case of an interactive term like is present in this paper. A solution to this is using structural equitation modelling (SEM) which has an advantage in that case because it does take measurement errors into account (Little et al, 2007). Furthermore regression does not allow for interaction between variables as it sees all variables independent. SEM on the other hand is more realistic as it allows for correlation between all variables, making it more reliable (Shu-Cheng, Hueimei, & Chang-Yung, 2010). Because of those advantages over regression it will result in more reliable results by reducing the overall error (Hair, Hult, Ringle & Sarstedt, 2014), which is why it will be used method used in this paper.

The next choice that needed to be made was which form of SEM would be used as multiple forms are available, of which the most documented and well-known are covariance-based SEM (CB-SEM) and variance-based SEM (VB-SEM) (Chin, 1998; Dijkstra, & Henseler, 2015; Wong, 2013). PLS-SEM is the most developed and sophisticated of the CB-SEM (McDonald, 1996) and has the advantage over the CB-SEM that it allows for models that are more complex, plus it can handle more diverse data like reflective and formative scales (Hair, Ringle, & Sarstedt, 2011). As the conceptual model in this paper is quite complex and both formative and reflective scales are used, PLS-SEM will be the method used in this paper.

More specifically the software package SmartPLS (Ringle, Wende, and Will 2005) was used, that just like most other PLS-SEM software provides a graphical way of modelling and implementing the basic PLS algorithm (Hair et al, 2011).

After selecting an algorithm, it also had to be determined which sampling method was going to be used. One of standard sampling methods in SmartPLS is bootstrapping (Hair et al, 2011), which repeatedly draws random samples with replacement from a sample to estimate adequate p-values (Efron, 1979). As this form of sampling results in the most stable resampling coefficients for samples of over a 100 cases (Berger, 2015), it is the sampling method that will be applied in this paper as the smallest sample size is (n=163). The number of subsamples that was used for bootstrapping in this paper was 5000 as it that is generally accepted as the minimum amount that is necessary in PLS-SEM (Hair et al, 2011).

(32)

32 Then all that had to be assessed was the appropriateness of the sample size of 163 (due to deletion of some cases which is further elaborated on in the next paragraph) used in this research. As PLS-SEM is a relatively young data analysis method there are no strict guidelines yet concerning minimum sample size (Wong, 2013). Literature however states that to begin with, a sample size of between 100 and 200 would lead to good results (Hoyle, 1995). Though more specific requirements are also present: 1) the largest amount of indicators to one formal construct, times ten (in this paper eight) 2) the greatest number of structural paths to a dependent variable, times ten (in this paper five) (Hair et al, 2011). Hence, it can be concluded that the sample size that is used in this paper is indeed

(33)

33

4 Data analysis and results

In general doing a PLS-SEM analysis consists of two steps. First the outer model is evaluated and after it is proven to be strong enough the second step is to assess the inner or structural model (Hair et al, 2011; Wong, 2013).

The outer model corresponds to the measurement model, which makes an estimation of the relation between latent variable and its indicators. Prior to the assessment of the whole structural model it is crucial to first test the latent constructs reliability and validity (first step). After the outer model has been confirmed to be adequate enough the next step is to assess the inner model which estimates the relations or paths between the independent and dependent latent variables (second step) (Henseler, Ringle, & Sinkovics, 2009; Wong, 2013).

However, before both models can be evaluated they first need to be build. It is important to note that constructing a model which includes second-order constructs or hierarchical component models, like this paper does, needs extra attention. A failure of the majority of conducted research is that it does not explain which second-order construct approach was adopted, and why it was used (Ringle, Sarstedt, Straub, 2011). Of the research that does indicate which approach was adopted, about 70% used the ‘indicator reuse technique’, even though under some conditions this is not the right approach (Ringle, Sarstedt, Straub, 2011; Wold, 1986). As all latent variables need indicators in the model, the reuse approach uses the same indicators for the first and second order construct. One of the complications making this approach inapplicable is that in case the second-order construct has a predecessor, the path coefficient between the two will always be approximately zero as all the variance is already explained by its first order-constructs (Ringle, Sarstedt, Straub, 2011; Wold, 1986). A solution according to literature is that in models that include a second order constructs with a predecessor, a mixture of the indicator reuse technique and a latent variable two stage approach should be apployed (Ringle, Sarstedt, Straub, 2011). In this approach the first step is the reuse technique, which provides latent variable scores that will then be saved. Those saved latent variable scores will then be employed in the second step as indicators of the second order-variable, that now forms a first-order construct, in a newly formed model (Becker, Klein, & Wetzels, 2012; Ringle, Sarstedt, Straub, 2011). On the ground that in this paper two second-order constructs with

(34)

34

Data Preparation

To prepare the data it first needs to be checked for missing values, and after that for unusual responses or outliers (Hair et al, 2013).

Missing data can form a problem, so it is important that it is dealt with sufficiently. If the percentage of missing data for an observation (correspondent) exceeds fifteen percent, it is generally advisable to delete this observation from the data (Hair et al, 2013). To analyse this, IBM SPSS was used to generate a new variable that counted the missing answers per respondent. As in this paper initially 62 indicators were used, this meant that a correspondent with more than nine missing values had to be deleted. This was the case for three respondents, and they were therefore deleted from the dataset. Important to mention is that all of these three respondents were suppliers, meaning that the separate dyads of the buyers and suppliers will have an unequal number of observations (suppliers=163, buyers=166).

The remaining values that were missing also had to be taken into consideration. Because literature indicates that expectation maximization is the most preferred method in case of SEM (Gallagher, Ting, & Palmer, 2008) this was first applied. However, as the Little’s test was significant, that meant that is was not a suitable method in this case. Moreover, it is stated by Hair et al (2013) that there is no knowledge on the suitability of replacement methods in PLS-SEM other than the default methods available in SmartPLS (mean replacement, case wise deletion, and pairwise deletion). And even though mean replacement can decrease variability, in case of low levels (rule of thumb <5%) of missing data mean replacement is the recommended method for SmartPLS (Hair et al, 2013). As missing data levels are extremely low (57 indicators <1%, 4 indicators 1.5-2.1%, and only 1 indicator 5.1%) mean replacement will be used.

Subsequently, unusual responses or outliers were checked. There were no observations found that had an anomaly index greater than 2, indicating that there were no problems with unusual

responses.

DOES SMART ALLIANCE GOVERNANCE CREATE SMART FIRMS?