The ambidextrous freelancer : understanding the effects of ambidexterity on performance for freelancers working in the Dutch creative industries

(1)

Master thesis

The ambidextrous freelancer

Understanding the effects of ambidexterity on performance for freelancers working in the Dutch creative industries

Track: Entrepreneurship and Management in the Creative Industries Name: Jorik Feskens

Student number: 10626980 Supervisor: Monika Kackovic Date: June 30th 2014

(2)

Abstract

This thesis is focused on the relation between ambidexterity and performance on the level of freelancers in the creative industries. Ambidexterity refers to the concept of organizational ambidexterity which means that an organization should find a balance between exploitational activities and explorational activities in order to survive in the long run. A lot of studies have been done on the relation between ambidexterity and performance on the organizational or business-unit level, and the few studies at the individual level focused on managers within organizations. A research gap can therefore be found in the effects of ambidexterity on performance at the level of freelancers, which is the focus of this thesis. Because earlier research showed that ambidexterity was of especial importance in dynamic environments the empirical setting of this thesis is the Dutch creative industry. By means of a quantitative study using online surveys the ambidexterity-performance relation is measured after which multiple regression analysis (ordinary least squares) is used to analyze the data. The findings of this analysis show that ambidexterity is significantly, but negatively related to performance. This relation can partly be explained by those freelancers that score high on both exploitation and exploration. Furthermore, the relation between ambidexterity and performance is stronger in the Randstad. Some of these findings contradict theory while there may be explanations for this and other findings support theory while having possible biases. Further research is necessary to show the generalizability of these findings.

(3)

1. Introduction

Organizational ambidexterity has been a widely researched topic in the past 15 years (Tushman & O’Reilly, 2013). It is defined as “an organization’s ability to be aligned and efficient in its management of today’s business demands while simultaneously being adaptive to changes in the environment” (Raisch & Birkinshaw, 2008; p.375). These two sides of organizational ambidexterity refer to exploitation on the one hand, and exploration on the other, which should be balanced in order to survive in the long run (Tushman & O’Reilly, 2013). Exploitation is about a company’s current assets, processes and competences and using these as efficient as possible, and exploration is about experimenting with new alternatives (March, 1991). Too much focus on exploitation at the expense of exploration will lead to organizational inertia that prevents the organization from properly adapting to changing environments, which will cause poor performance outcomes in the long run (Levinthal & March, 1993). While, in contrast, too much focus on exploration leads to underdeveloped new ideas when innovations are replaced by new ideas before they have had the opportunity to contribute to the firm's revenue stream (Levinthal & March, 1993). The extensive research on the relation between organizational ambidexterity and performance shows a positive effect (Junni et al., 2013). In other words, organizations that are successful in attaining and retaining ambidexterity are more successful than organizations that focus more on either the

exploitation or exploration part.

Research has shown that this state of ambidexterity can be reached in a couple of ways. The first is sequential ambidexterity, where organizations go back and forth between periods of exploitation and exploration (Brown and Eisenhardt, 1997). The second is simultaneous ambidexterity, which proposes that exploration and exploitation should be strived for simultaneously (Tushman and O'Reilly, 1996). And finally there is contextual ambidexterity, which was proposed by Gibson and Birkinshaw (2004) and will be elaborated

(5)

on later in this thesis. All three have been found to be potentially viable (Tushman & O'Reilly, 2013). Furthermore, exploration and exploitation have even been found to be mutually

exclusive in some cases (Mom et al., 2007), which means that scoring high on exploration does not mean a high score is not possible on exploitation as well, and vice versa.

An examination of the literature indicates that there are topics, research environments and levels of analysis within the construct of organizational ambidexterity that remain fairly unexplored. As Junni et al. (2013) state, “it is important to examine whether the effects of organizational ambidexterity are constant across different levels of analysis or whether they tend to accumulate at specific levels.” But, there are only a few studies done at the level of the individual (p. 301). Moreover, the few studies that were done on the individual level focused on managers or employees working in organizations. A research gap thus lies in the effect of ambidexterity on performance on the level of freelancers, which will be the focus of this thesis. Although managers and freelancers are quite different they do have a common characteristic that makes it possible to study them the same way, namely decision-making authority. Decision-making authority refers to the authority to determine how to divide their time, effort and resources between exploration and exploitation (Mom et al., 2009). As Mom et al. (2009) show that decision-making authority is related to ambidexterity, ambidexterity should also be found in freelancers.

At the same time there seems to be a research gap regarding studying organizational ambidexterity within the creative industry. Only a few studies have focused on this industry, namely Hotho and Champion (2010), who did an empirical study on a computer game firm, and Andriopoulos & Lewis (2010), who focused on product design companies. For these reasons I would like to try to take the research of ambidexterity in the creative industries a little further by choosing to do my master thesis in the creative industry. More specifically, I will focus my thesis on the second and third domain of the creative industry proposed by

(6)

Stam et al. (2008), namely the media and publishing domain and the creative business services domain. I will elaborate on this in the "ambidexterity and the creative industry" section of my literary review.

Studying freelancers in the context of the creative industry is an interesting combination as the creative industry is the fastest growing sector in the Netherlands (6% growth), and 66% of the people working in the creative industry consists of freelancers (creativecounsil.nl). In a rapport made by the “Advies Topteam Creatieve Industrie” (Chijs et al., 2011) on the state of the creative industry in the Netherlands they write that creatives operate in high-risk and fast changing environments, which demand a great deal of adaptability (p. 5). It thus seems that freelancers within the creative industry act in a very competitive environment that could benefit greatly from ambidexterity. This competitiveness is even more enhanced by the possible dispense of the “zelfstandigenaftrek” (tax deduction for self-employed) which would mostly affect creative freelancers (Profnews, 2013). It should therefore be interesting to see if these freelancers in the creative industry benefit from being ambidextrous like managers and organizations do. This thesis’ research question is therefore:

To what extent does ambidexterity affect performance for freelancers in the creative industries?

In order to answer this research question an internet survey was created, that included the exploitation and exploration scales from research done by Mom et al. (2007; 2009). This survey was then spread among creative freelancers through LinkedIn groups and through the help of Freelancers United, an organization that mediates between freelancers and possible employers. The findings were then analyzed with IBM's SPSS by using a multiple regression analysis.

The rest of this thesis is structured as follows. First, in the literary review, the underlying theories are described that research on organizational ambidexterity has brought

(7)

forward through the years. I will then discuss the methods that were used to gather and

analyze my data in the "Research design and methodology" section, followed by a description of the findings in the results section. Then, the findings and implications for future research will be discussed in the discussion and limitations section and finally a summary of this thesis will be given in the conclusion.

2. Literary review

I will start this literary review by elaborating on organizational ambidexterity on the

organizational level because that is where most of the research has been done. After that I will describe the literature on the individual level, and combine this with the literature on the organizational level and the creative industry to formulate the hypotheses aimed at testing the research question.

2.1 Organizational ambidexterity

The first author that used the term organizational ambidexterity was Duncan (1976), but it was James March (1991) that incited the interest of the scientific world. In his view

organizations can only survive in the long term when they find a balance between exploration and exploitation. As described in the introduction exploitation is about a company’s current assets, processes and competences and using these as efficient as possible. March (1991) writes that exploitation can be captured by terms such as “refinement, choice, production, efficiency, selection, implementation and execution”. On the other hand, exploration is about experimenting with new alternatives and can be captured by terms such as search, variation, risk taking, experimentation, flexibility, discovery and innovation (March, 1991). He states that "the basic problem confronting an organization is to engage in sufficient exploitation to ensure its current viability and, at the same time, devote enough energy to exploration to ensure its future viability" (1991, p. 105). Its seems logical for a company to engage in both,

(8)

but the problem arises because there are limited resources (think of people and money) and a company needs to decide how to divide these resources between exploitation and exploration processes. Finding the optimal balance involves possible trade-offs (Simsek et al., 2009), and because, as March (1991) writes, returns coming from exploitation processes are “positive, proximate, and predictable” and the returns coming from exploration processes are “uncertain, distant, and often negative” there is a bias in favor of exploitation with its greater certainty of short-term success. “Yet, without some effort toward exploration, firms, in the face of change, are likely to fail” (Tushman & O’Reilly, 2013). March (1991) even calls refining exploitation more rapidly than exploration self-destructive in the long run. This bias towards exploitation is reinforced by other researchers. Uotila et al. for instance, estimate that 80% of the firms in their sample focus more on exploitation while under-emphasizing exploration (Uotila et al., 2008). Thus, an organization needs to find an optimal balance between exploration and

exploitation in order to survive in the long run. When this balance is found, an organization be described as ambidextrous (Tushman & O’Reilly, 1996). Of course, creating a balance of both high levels of exploration and exploitation is desired. This implies that efficiency is high in current operations while, simultaneously, new opportunities are identified and captured at a high level to prevent organizational inertia and the negative effects of path dependence (Simsek et al., 2009). However, this can be very costly and difficult to achieve (Junni et al., 2013).

The effect of ambidexterity on performance has been researched extensively. Even though the data was dependant on empirical evidence and anecdotes in the beginning (Tushman & O’Reilly, 2013), since then there have been a lot of articles written about this relation. One of those articles is written by Geerts, Blindenhach-Driessen, and Gemmel (2010), who looked at more than 500 firms over a four-year period and found that

ambidexterity had a positive effect on firm growth. It is articles like this, and many more, that

(9)

made Tushman and O’Reilly (2013) come to the conclusion that ambidexterity has been shown to be positively associated with performance.

Since March (1991), different literature streams, including organizational learning, technological innovation, organizational adaptation, strategic management, and organizational design, have contributed to the research on organizational ambidexterity. (Raisch &

Birkinshaw, 2008). Organizational ambidexterity can thus be seen as a touching point between all these literature streams, which makes it a very multidimensional concept and, in turn, a very complex one.

In order to make sense of this complex concept researchers have tried to identify the optimal way to reach this state of ambidexterity. Three of these ways have been described in the literature which are all potentially viable (Tushman and O’Reilly, 2013). The first is sequential, or structural, ambidexterity, which describes how an organization can adapt their structures and processes to reflect changed environmental conditions (Tushman & O’Reilly, 2013). By doing this, organizations “oscillate back and forth between periods of exploitation and exploration” (Brown and Eisenhardt, 1997; Tushman & O’Reilly, 2013).

Tushman and O’Reilly (1996) on the other hand, proposed that organizations should not go from exploitation to exploration sequentially, but that they should do it simultaneously and that the managers are of great importance in this process. Later, they proposed that this could be accomplished by establishing separate aligned organizational architectures (O’Reilly & Tushman, 2008; Tushman & O’Reilly, 2013). With alignment they mean that even though there are different people, structures and processes for explore and exploit subunits, there is also a clear, overarching, consensus of a common vision and strategy to legitimate the need for exploration and exploitation, and with leadership that is capable of managing the tensions that ensue.

Finally, Gibson and Birkinshaw (2004) proposed that the tension between exploration

(10)

and exploitation in an organization could be resolved at the individual level, and called this contextual ambidexterity. They defined contextual ambidexterity as "the behavioral capacity to simultaneously demonstrate alignment and adaptability across an entire business unit" (p.209). In this definition alignment refers to whether all the parts in a business unit are working together toward the same goals, and adaptability refers to the capacity to change activities in the business unit quickly to meet changing demands in the task environment (Gibson and Birkinshaw, 2004). To reach contextual ambidexterity the management system and culture supports workers to pursue exploration and exploitation themselves by "building a set of processes or systems that enable and encourage individuals to make their own

judgments about how to divide their time between conflicting demands for alignment and adaptability" (p. 201) or, in other words, between exploratory and exploitative activities. The result of the research that Gibson and Birkinshaw (2004) conducted within 41 business units was that successful business units scored higher on both alignment and adaptability than less successful units.

2.2 Individual ambidexterity

Until now I have only mentioned ambidexterity at the level of analysis of a whole company or a business unit. The reason for this is that the vast majority of research on ambidexterity is done on these levels of analysis (Tushman & O’Reilly, 2013; Junni et al, 2013). There has, however, been research done on the topic of individual ambidexterity. Gibson and

Birkinshaw’s (2004) contextual ambidexterity, as I described earlier, pertains to the

individuals within organizations. However, their paper cannot so easily be translated to the sample of this thesis, namely self-employed people. This is because their research has been done on individuals within business units, and they only focus on the characteristics of alignment and adaptability. Research on ambidexterity on the level of the manager could be more comparable for studying self-employed. The key characteristic that makes managers and

(11)

self-employed comparably in terms of doing ambidexterity research is decision-making authority.

Decision-making authority refers to the extent to which a manager can decide which tasks the manager performs, how he does this, and his or her ability to solve problems and to set goals (Atuahene-Gima 2003; Dewar et al., 1980; in Mom et al., 2009). Mom et al. (2009) hypothesize that a manager’s decision-making authority is positively related to this manager’s ambidexterity, and show that there is a positive relation between the two after their research on 716 business unit level and operational level managers. Decision-making authority can of course also be found in self-employed as they are basically their own CEO’s and have to determine how to divide their time, effort and resources between exploitation and exploration. Decision-making authority also increases the motivation a manager, or a freelancers, has to perform. It stimulates their willingness to recognize a larger diversity of organizational, market, and technological opportunities while they also become more sensitive to

understanding how to act upon these different opportunities and needs (Miller, 1987; Pierce and Delbecq, 1977; Tushman and O’Reilly 1996; in Mom et al., 2009).

Research on the relation between individual ambidexterity and individual performance is rare. Jasmand et al. (2012) did find a significant, positive effect of ambidextrous behavior on sales performance, but the study was done on customer service representatives and not managers. Most of the studies focusing on ambidexterity and managers relate it to

organizational performance (Junni et al., 2013), like the study on contextual performance by Gibson and Birkinshaw (2004). Because of this, I have to base my argument on the relation between individual ambidexterity and performance on research done on other levels of analysis. As studies on the relation between ambidexterity and performance in organizations, business units, and teams mostly show a positive relation I expect to find the same in this thesis.

(12)

H1: The relation between ambidexterity and performance is positive for freelancers in the creative industry

As Raisch et al. (2009) state; “The possibility that individuals can take on both exploitative and explorative tasks creates a number of challenges that need to be addressed”(p.687). They mention that managing contradictions and conflicting goals, engaging in paradoxical thinking, and fulfilling multiple roles are some of these challenges (Smith & Tushman, 2005; Gibson & Birkinshaw, 2004; Floyd & Lane, 2000; in Raisch et al., 2009). It is challenging for a manager to address these issues, or in other words; to excel at both exploitation and exploration (Gupta et al., 2006). The upside is that Mom et al., (2009) demonstrate that these difficulties are not insurmountable. More so because exploration and exploitation are not mutually exclusive, as I described earlier, which also counts at the manager level (Mom, van den Bosch & Volberda, 2007). “Whereas some managers engage more in exploration activities as compared to exploitation activities, or the other way around, other managers have high levels of both exploration and exploitation” (p.925). Junni et al. (2013) even state that combined measures of ambidexterity capture the performance effects better than balanced measures. With this they mean that the ambidexterity-performance relation is mostly explained by high levels of both exploitation and exploration instead of a proper balance between the two. This leads to the following hypothesis in order to test if this is true on the individual level:

H2: The relation between ambidexterity and performance can be explained by those ambidextrous freelancers that have high levels on both exploration and exploitation.

When researching ambidexterity in freelancers it is necessary to determine what kind of behavioral characteristics to look for. Mom et al. (2007) give an overview what kind of exploration and exploitation behavior constitutes ambidexterity in managers. They begin with explaining that exploration activities of managers include “searching for new organizational

(13)

norms, routines, structures, and systems (Crossan et al., 1999; Nooteboom, 2000; Zollo and Winter, 2002; in Mom et al., 2009), experimenting with new approaches towards

technologies, business processes, or markets (McGrath, 2001; in Mom et al., 2009), innovating and adopting a long-term orientation (Duncan, 1976; Tushman and O’Reilly, 1996; in Mom et al., 2009), and reconsidering existing beliefs and decisions (Floyd and Lane, 2000; Ghemawat and Ricart I Costa, 1993; Rivkin and Siggelkow, 2003; in Mom et al., 2009). On the other hand, managers’ exploitation activities include “using and refining their existing knowledge (Levinthal and March, 1993; in Mom et al., 2009), applying, improving, and extending existing competences, technologies, processes and products (March, 1991; in Mom et al., 2009), focusing on production and adopting a rather short-term orientation (Duncan, 1976; Tushman and O’Reilly, 1996; in Mom et al., 2009), and elaborating on existing beliefs and decisions (Floyd and Lane, 2000; Ghemawat and Ricart I Costa, 1993; Rivkin and Siggelkow, 2003; in Mom et al., 2009). Because Mom et al. (2007;2009) focus their ambidexterity research on managers I found their methods of testing ambidextrous behavior also useful for research on freelancers. I will elaborate on this in the methods section.

2.3 Ambidexterity and the creative industry

As Junni et al (2013) state in their meta-analysis of the effect of ambidexterity on

performance; “several studies have pointed out the organization's environment as a possible moderator of organizational ambidexterity" (e.g., Raisch & Birkinshaw, 2008; Simsek et al., 2009). They go on by writing that, in line with this perspective, prior research has proposed that the effects of ambidexterity could be industry-specific which would mean it has more positive effects in dynamic environments (Simsek et al., 2009). The reason that it has a more positive effect in dynamic environments is that the duration of an existing competitive advantage is very uncertain, and therefore firms need to continuously be on the lookout for

(14)

new opportunities while at the same time exploiting existing resources (Junni et al., 2013; Bingham & Eisenhardt, 2008; Brown & Eisenhardt, 1997). It is thus interesting to look at ambidexterity and the role it plays in different industries. Besides that, it is important to examine the boundary conditions of studies on ambidexterity in terms of the robustness of the ambidexterity – performance relationship across different industry contexts (Junni et al., 2013).

The creative industry is an industry that seems to be forgotten by the literature on ambidexterity. Besides research at product design companies (Andriopoulos & Lewis, 2010) and a computer games company (Hotho & Champion, 2010), there are no articles to be found that combine the literature on organizational ambidexterity with the creative industry, let alone articles on individual ambidexterity. This alone makes it interesting to take a further look at the relation between these two, but moreover the creative industries is characterized by a fast changing and high-risk environment (Chijs et al., 2011) which makes it extremely dynamic. Therefore the creative industry could be an environment where ambidexterity is especially important for organizations and freelancers to survive.

When defining the creative industry it is important to keep in mind that there is not one completely accepted definition to be found. Different researchers have used a wide range of definitions that all have a different view on which industries, firms, and occupations are included in the creative industries (Carey and Naudin, 2006; Markusen et al., 2008; Stam, de Jong & Marlet, 2008). To give an illustration; Braaksma, de Jong and Stam (2005) found that creative industries could comprise between 1.7 per cent (when including only the arts) and 19 percent of the Dutch business population (when knowledge-intensive business services such as consultants and software developers are included). For this thesis, I will use the

classification of the creative industries that is proposed by Stam et al. (2008) who, in turn, based their classification on the works of Florida (2002) and Rutten, Manshanden, Muskens

(15)

and Koops (2004). They argue that the arts and media industries should, “without any doubt, belong to the group, as the nature of the work in these industries clearly reflect Florida’s core of super-creative professions”. In addition, relating to Rutten et al. (2004), the definition should also include creative business services like designers, architects, and software

developers. These creative business services include many creative professionals with jobs of a clearly creative nature (e.g. designing a house, creating an advertising campaign). Stam et al. (2008) therefore propose that the creative industry consists of three domains, namely: 1 arts (e.g. performing artists, visual artists, theatre companies).;

2 media and publishing (e.g. photographers, broadcasters, journalists) 3 creative business services (e.g. technical designers and advertising ﬁrms).

For this study I will focus on the second and the third domain for a couple of reasons. The first relates to the time constraint for writing this thesis; when including the arts the research population would be much larger which means that the necessary research sample will also have to be much larger in order to make reliable statements about the findings. Also, the arts domain is very different from the other two domains. Because artistic performance is valued differently than commercial performance in the arts domain (Stam et al., 2008), including it would require a different way of testing performance. Net income would not be a good measure for artistic performance, so other measures would have to be developed. This, in turn, would make it hard to compare the arts dimension to the other two dimensions. In the creative industries the project based organization is an attractive organizational form, because it ensures bringing about variety and flexibility (Faulkner & Anderson, 1987; Jones, 1996; in Ebbers & Wijnberg, 2009). These project based organizations consist of freelancers or employees from different firms that have the necessary skills for the project to be successful. Most film companies, for example, outsource creative tasks to a wide network of freelance creatives (e.g. directors, actors or cinematographers) and small specialized firms

(16)

(e.g. lighting-, camera- or post-production firms) (Cooke and Lazzeretti, 2008). Other studies also show the heavy reliance of the creative industry on freelancers and related flexible working patterns compared to other sectors of economic activity (Carey and Naudin, 2006). “Employing shifting freelancers and specialized suppliers allows for shifting resources and skills as needed, and hence for flexibility and openness of projects” (Cooke and Lazzeretti, 2008; p.168). The dynamic environment and the importance of freelancers makes the creative industry a very interesting research setting for studying ambidexterity.

Stam et al. (2008) studied the creative industries in the Netherlands and came to the conclusion that geographical differences in the Netherlands have an effect on the

innovativeness of firms in the creative industries. Firms located in urban areas were found to be more innovative than their rural counterparts. This leads to my last hypotheses where I will test if the relation between ambidexterity and performance differs per working area of the creative freelancer. Stam et al. (2008) found differences in the spatial distribution of

employment in the creative industries between the media and publishing dimension and the creative business services dimension. The depiction of this spatial distribution can be found in appendix 3. Assuming that a higher concentration of firms in the same sector results in a more competitive environment that demands a better balance between exploration and exploitation or a higher need for innovation, this leads to the following hypotheses:

H3a: The relation between ambidexterity and performance is stronger for freelancers working in the Randstad than for freelancers working outside the Randstad.

H3b: Exploration is more positively related to performance for freelancers working in the Randstad than for freelancers that do not work in the Randstad.

(17)

3. Research design & Methodology

In this section the research design and research methodologyare described. First, the

empirical setting and the methods of data collection are described. After that, the independent, dependant and control variables that were measured are operationalized. Finally, the methods of analysis and the process of data cleaning are described.

3.1 Emperical setting

The empirical setting for this study is the Dutch creative industry. The Netherlands is a country with one of the highest concentrations of the creative class and creative industries (Florida, 2005). As stated in the literature review I will focus on freelancers in the media and publishing domain and the creative business services domain based on the article by Stam et al. (2008). The concentration of creative freelancers working in the media and publishing domain is highest in the Randstad, which is the main agglomeration in the Netherlands and contains the cities of Amsterdam and Utrecht (see appendix 3). The concentration in the relatively rural area in between these two big cities can be explained by the Dutch

broadcasting industry which is based there. The creative business services are quite evenly spread over the country (Stam et al., 2008).

3.2 Data source and data collection

The Data was collected by means of an internet survey which was initially spread on LinkedIn groups where a lot of creative freelancers are active. These were: Association of Dutch

Desigers (BNO), Beroepsvereniging Nederlandse Interieurarchitecten (BNI), CLICKNL, Dutch Games Association, Freelance grafisch ontwerpers, Freelance ICT professionals, Frennel, Landschapsarchitecten, VEA, de Architect, Educatieve Ontwerpers, and De Invasie. These LinkedIn groups were selected because they fit the second and third dimension of the

(18)

creative industry from Stam et al. (2008), and because they all had more than 20 Dutch Members. The message I used to attract respondents can be found in appendix 4. Afterwards one of the respondents send a message with the names of three organizations were a lot of creative freelancers worked. After emailing those organizations with the same message and the link to the survey, one of them (stichting de fabriek) was willing to spread my survey among the freelancers working there. I also emailed several employment agencies that mediate between creative freelancers and possible employers. One of them, Freelancers United, was willing to work together. Freelancers United is an agency mostly consisting of freelancers in marketing and communication. In exchange for giving them the results of this study they were willing to send the survey to the 500 freelancers working for them, in their name. The message they send to their freelancers can be found in appendix 5. In total, this resulted in a response of 166 surveys. Unfortunately, 35 of these were uncompleted surveys that weren’t useful as they lacked answers on the questions relating to both the independent variable (ambidexterity) and the dependent variable (performance). These responses were therefore excluded, resulting in a total of 131 surveys.

3.3 Measurement of variables and reliability

3.3.1 The independent variable: measuring ambidexterity

Organizational ambidexterity has been researched in a large variety of ways. On one end of the spectrum are articles like the one by Gibson and Birkinshaw (2004) that study

ambidexterity by looking at a few concepts that contribute to the balance between exploration and exploitation like discipline, trust and support. On the other side of the spectrum are articles by Cao, Gedajlovic and Zhang (2009) and He and Wong (2004) that have a more straight forward approach to measuring ambidexterity by looking at the degree of exploration and exploitation that takes place. For this master thesis the latter approach is used to measure ambidexterity because of the time constraint that is present, and the fact that the first option

(19)

would also require some qualitative research.

When choosing to take this approach another problem arises though, namely how exactly you decide to operationalize exploration and exploitation. Most studies on

ambidexterity see exploration and exploitation as two concepts pertaining to organizational learning, namely introduction of new products and markets (exploration) and the

improvement of existing products and markets (He & Wong, 2004; Lubatkin et al., 2006; Smith & Tushman, 2005; Tushman & O’Reilly, 1996; in Cao, Gedajlovic and Zhang, 2009). Thus, “the differences between the two concepts pertain to whether the new learning occurs along the same trajectory as the old one or along an entirely different trajectory (Gupta & Smith, 2006). Some studies though, treat all activities associated with learning and innovation as forms of exploration and reserve the term "exploitation" for activities where the central goal is ongoing use of a firm's knowledge base rather than moving down any kind of a learning trajectory (Rosenkopf & Nerkar, 2001; Vassolo, Anand, & Folta, 2004; Vermeulen & Barkema, 2001, in Gupta & Smith, 2006). Two of those studies are especially useful for this thesis, namely Mom et al. (2007; 2009), which I mentioned in the literature review. Mom et al. (2007) study the effect of knowledge inflows on manager's exploration and exploitation activities. After some minor adjustments the methods they use to measure

ambidexterity in managers can also be used to measure ambidexterity in freelancers. In their later research (2009) they make some minor adjustments in their survey when they focus on investigating manager's ambidexterity and develop a model and hypotheses on the effects of formal structures and personal coordination mechanisms on managers' ambidexterity. This is of course different than what I am researching, because the focus of this thesis lies on the ambidexterity-performance relation which Mom et al. (2007;2009) do not adress. I can, however, use their measures for determining my dependant variable, namely ambidexterity on the individual level.

(20)

The method they used in their 2009 study was asking respondents to rate, on a 1-7 scale, the extent to which fourteen different statements were true regarding exploration and exploitation activities within the past year. Seven of these statements pertained to the firm’s exploration (i.e. searching for new possibilities and activities requiring them to learn new skills or knowledge)and seven pertained to exploitation (i.e. routine activities or activities focused on achieving short-term goals). On this 1-7 point scale a one means not having done this activity at all and a seven means having done it a lot. Only these two numbers have labels, the numbers 2-6 do not. The argumentation for this is that for a scale like this the 'distance' between the points should be equal along the whole scale in order to treat the results as an interval-variable. When you put a label on every number you run into the problem whether, for instance, the interval between "helemaal niet" (not at all) and "zelden" (rarely) and between "regelmatig" (regularly) and "redelijk vaak (quite a lot) is equal. On the other hand, when respondents are able to divide the points for themselves between the two extremes then they will do so in more or less equal parts. A second argument for using the same

strategy Mom et al. (2009) used is that emperical results based on different scale formats may not be comparable (Weijters, Cabooter & Schillewaert, 2010).

The exploitation and exploration scales used by Mom et al. (2009) can be found in the appendix 1. In order to use these scales on freelancers I only had to adjust one statement in the exploration activities, and one statement in the exploitation activities. These statements

pertained to activities that were "existing company policy". They were changed into activities that were respondents' usual/common activities (gangbare werkzaamheden). Other items could be directly translated to Dutch. For the final version of the (Dutch) scales see appendix 2. Even though the scales I used were already tested on reliability by Mom et al. (2009) I did another reliability test on the scales because I had to translate them to Dutch. The exploration and the exploitation scale was tested separately with the "reliability analysis" option in SPSS

(21)

(Pallant, 2001). These tests showed a Cronbach´s Alpha of ,791 on the exploration scales and a Cronbach´s Alpha of ,878 on the exploitation scale. This means they are both above the recommended criteria of 0,70 (Kline, 2000). The Corrected Item-Total Correlation measure also showed good reliability as the lowest measure was ,393 with the rest between ,472 and ,737 (Pallant, 2001).

3.3.2 The dependant variable: measuring performance

In their meta-analysis of the ambidexterity-performance relationship, Junni et al. (2013) give an overview of the ways that performance has been measured in ambidexterity studies. They describe two different kinds of measures that are used, namely objective measures and perceptual measures.

The most common objective performance measures are growth and profitability (Lin et al., 2007; Mudambi & Swift, 2011; in Junni et al., 2013) while perceptual performance measures can for instance be asking how the respondents felt about their performance or how their performance was compared to that of competitors. The successes of the perceptual performance measures all came from a company, business unit or manager level of analysis (Junni et al., 2013) where there are top-down knowledge flows coming from the CEO or top management teams about, for instance, the relative performance of the company (Mom et al., 2007). This is not the case with self-employed, which might result in some of the respondents saying they are doing really good while they actually have no knowledge of their relative standing compared to other self-employed. Therefore performance will be measured with an objective performance measure in this thesis.

When choosing to use objective performance measures a choice has to be made about which measures to use. Like I stated above, the most common measures are growth and profitability. I measured performance through profitability because it is the most used method of measuring performance (Junni et al., 2013). This was done by asking for the respondents’

(22)

after tax income as a freelancer. Because I assumed that a freelancer is unlikely to earn more than 150.000 euro in a year this question came in the form of a drop-down menu that contains choices per 1000 euro going from 0 to 150.000 with an added choice of 150.000+. The

question could also be answered by the option “I would rather not answer this question” (translated in Dutch) in case a respondent did not want to give this information.

3.3.3 Control variables

In order to make reliable statements about the ambidexterity-performance relation I have included nine questions in the questionnaire that can be used as control variables. The first of these questions are pretty straightforward as they concern the gender and the age of the freelancer. The other control variables will be elaborated on below.

Level of education

“Increasing levels of education are associated with increasing cognitive abilities to process information and learning” (Papadakis et al., 1998; in Mom et al., 2009). These higher cognitive abilities may influence the ambidexterity-performance relation in a way that respondents with a higher level of education earn more or pursue a different balance of exploitative and explorative activities than lower-educated respondents. In order to be able to pay attention to this influence I have included a question asking what the level of the

respondents highest completed education is, which can be answered by: no education, LBO/VBO, MAVO/VMBO/MBO, HAVO/VWO/WO- propaedeutic, HBO/WO bachelor or WO-doctoral or master. In international standards, the second answer LBO/VBO means only having finished primary school. The third; MAVO/ VMBO, is the lowest level on Dutch high schools which is often followed by an MBO (intermediate vocational education). The fourth answer consists of the middle and high level on high schools HAVO/VWO or only a

propaedeutic diploma on a university (WO). The fifth answer then consists of a finished

(23)

degree on the level of a higher vocational education (HBO) or a bachelor on the university level. Finally, the last answers pertains to those respondents with a finished master degree or doctorate of a university.

Marital status

Marital status is an important control variable because the balance in exploration/exploitation and the effort put into a freelance career may be affected by how (in)dependant someone is (Blau, 1985). The reason for this is that if you are very dependant of the income you generate with your freelancer activities you might be a lot more competitive than a freelancer that can always rely on the income of a partner. Because of this the survey contained one question related to the marital status of the respondent and one question related to the percentage the freelancer contributed to his or her household (if the respondent was married or living together).

Working sector

In order to test whether the ambidexterity-performance relation is influenced by which sector the creative freelancers is working in I included a question in the survey pertaining to the working sector. This question is a multiple choice question that can be answered by: “IT”, “Fashion design”, “Industrial design”, “Graphic design”, “Architecture”, “Interior design”, “Photography”, “Communication/ commercials”, “Marketing”, “Different form of design” or “Different”. These answer possibilities were based on the LinkedIn groups where the survey was spread. In turn, those LinkedIn groups were selected because they fit the second and third dimension of the creative industry from Stam et al. (2008), and because they all had more than 20 Dutch Members. The percentages of how much respondents from each sector were eventually recorded can be found in the descriptive statistics section of the results chapter.

(24)

Working area

As mentioned earlier Stam et al. (2008) stated that besides there being a larger concentration of creative firms in the urban areas, the companies in those urban areas are also more

innovative than their rural counterparts, which resulted in hypotheses 3a and 3b. In order to test this freelancers were asked to state the geographical area that they worked in. This was measured in a multiple choice question where multiple answers were possible. It contained thirteen possible answers pertaining to each province in the Netherlands and one extra answer possibility for those who also worked outside the Netherlands.

Experience

Because "increased levels of experience are associated with an increased ability to interpret and deal with a larger diversity of ambiguous cues" (Daft and Lengel 1986, p. 555) it may affect the ambidexterity-performance relation. In order to control for experience I included two measures of this in my survey. The first question is about the years of experience in the sector that the freelancer is working in and the second question relates to the years of experience the respondent has as a freelancer. Because the correlation table of my variables (which can be found in the results chapter) showed a large correlation between these two experience variables one had to be chosen for my regression analysis in order to avoid multicollinearity. I chose to use the years of experience variable because the years as a freelancer variable was negatively related to income.

3.4 Methods of analysis

I used multiple regression analysis to answer my hypotheses. This allowed me test how ambidexterity relates to performance for freelancers, and to control for the effects of covariates. The assumptions of normally distributed variables, a linear relation between the independent and dependant variable, reliable measures and the assumption of

(25)

homoscedasticity were all tested in order to avoid type I and type II errors. The results of these tests can be found in the results chapter.

3.5 Data cleaning

In order to answer my hypotheses the raw data had to be cleaned in SPSS. After renaming my variables the variable view contained the control variables “Gender”, “Age”, “Education”, “Marital status”, “Worksector”, “Workexperience”, “ZZPexperience” as well as thirteen variables about the geographical area the respondents works in. Besides that there are “Explor1” until “Explor7” which are the interval variables related to the seven statements regarding explorational activities and “Exploit1” until “Exploit7” related to the exploitational activities. Finally, there is the “Income” variable.

I had to change some data points regarding the working sector of respondents. There were some people that answered this question by “a different form of design, namely:” or by “different, namely:” that answered “namely” by giving a sector or profession that, according to my definition, could be categorized under an existing category. I therefore changed some answers of respondents to the number of the category they belonged to.

In order to answer my hypotheses I also had to compute some variables. First I computed the seven exploration variables into one variable that contained one score for exploration by computing the mean, and after that I did the same with the exploitation variables. Following Mom et al. (2007; 2009), I then centered these variables and computed ambidexterity by computing the interaction between them.

For my second hypothesis the new variable “highlevels” was computed where freelancers with a mean of both exploration and exploitation that was higher than a 5 (on a seven point scale) were a 1, and every respondent below that a 0. The SPSS file was then split on this variable after which another regression analysis was done.

To answer hypothesis 3a and 3b I computed my geographic working area variables.

(26)

Because my internet survey contained a multiple choice question regarding the working area of the freelancer where multiple answers were possible, I had 13 different variables (12 for each province and 1 for “outside the Netherlands”) were a 1 meant a yes and a 0 meant a no. I computed the “Noord-Holland” variable and “Utrecht variable” to get a new variable

(Working Area) were a 1 meant that the respondent worked in the Randstad and a 0 meant that that respondent didn’t. I also computed the log of income in order to make the variable more normally distributed.

3.6 Dealing with missing values

29 of the 131 respondents answered the income question with “I would rather not answer this question”, which accounted for 22,1 percent of the respondents. In order to deal with these missing values I first tested the hypotheses that this data was missing randomly by using Little’s MCAR test (Allison, 2002). To do this the Missing Value Analysis option was used in SPSS were I imputed the quantitative variables age, years of experience, years as a freelancer, exploration and exploitation, as well as the categorical variables gender, marital status,

education, working area and working sector. The result can be found in table 1 below. With a significance level of ,536 the data was indeed missing at random (Allison, 2002). Because of this I was able to use the Expectation Maximization method to fill in the missing data points, which I also did through the missing value analysis option in SPSS (Allison, 2002)..

Table 1: Expectation maximization estimated statistics. Expectation Maximization Meansa

Age Workexperience ZZPexperience Exploration Exploitation Income

4,56 18,4046 8,7786 30,8176 34,9924 37,47

a. Little's MCAR test: Chi-Square = 4,092, DF = 5, Sig. = ,536

(27)

4. Results

In this chapter, the results of the analyses are discussed. First the process of testing for the assumptions of the hierarchical regression model is described. Then a general image of the collected data is given by describing the descriptive statistics and the correlations that were found. And finally the findings pertaining to the hypotheses are discussed.

4.1 Testing the assumptions for the method of analysis

As described in the methods section there are four assumptions that should be met in order to derive results correctly and avoid type I and type II errors (Green and Salkind, 2010). I will discuss the tests of these assumption while referring to the graphs in appendix 6. The first assumption is the assumption of normally distributed variables that states that the error terms are normally distributed at every level of the model (Berry, 1993). Graph 1 until Graph 5 are the results for testing normality of the independent variable ambidexterity, which was done by plotting a histogram of the variable (Green and Salkind, 2010). All the graphs show the normality assumption is met, including those resulting from splitting the file on the "high levels" or "work area" variable. This is important because a regression analysis was also used for testing hypotheses 2, 3a and 3b. The same test was performed in order to test normality of the dependant variable "Logincome", which can be found in graphs 6 until 10.

The second assumption for using hierarchical linear regression analysis states that there should be a linear relation between the independent and dependent variable (Berry, 1993). This was tested by plotting ambidexterity and income against each other in a scatter plot (Green and Salkind, 2010). The result of that test can be found in graph 11 and shows this assumption is met.

The third assumption of hierarchical linear regression is the assumption of

homoscedasticity, which assumes equality of population variances (Berry, 1993). For this, a

(28)

scatterplot was created for ambidexterity and performance that shows the relation between the residual and the predicted value. The result of this are Graph 12 and 13 and show

homoscedasticity in both the independent and the dependent variable.

The fourth and last assumption is that the variables are measured reliably. This test has been done in chapter 3.1.1 of the method section which showed that the exploration and exploitation measures are reliable.

4.2 Descriptive statistics and correlations

In my total dataset of 131 respondents 49.6% are male and 50.4 percent are women. The ages of the respondents are normally distributed, with most of them being in the 35-44 years category (33.6%) and the 45-64 years category (41.2%). Interestingly, 63.4% of the

freelancers are HBO educated and 28.2% even have a Master or doctoral degree which shows that my sample consisted of mostly higher educated people. Almost half of the respondents are married (49.6%), 28.2% is cohabiting and 21.4% is single.

As could be expected because of Freelancers United helping me spread my survey, by far most of the respondents work in communication/advertising. This amounted to 39.7% of the freelancers. The second most common working sector is graphic design which amounted to 20.6% of freelancers and besides that the other working sectors are quite evenly spread with the highest being marketing (7.6%) and the lowest being photography (1.5%). Also expected was that the amount of freelancers working in the Randstad is a lot higher than the amount of freelancers not working there. 74% of my respondents work in the Randstad. The mean of the respondents’ years of experience is 18,4 years, but the frequencies clearly show that freelancers rounded off their years of experience when they had more than 10 years of experience as there are strong peaks at 15 years (15.4%), 20 years (12.2%) and 25 years (11.5%). This was not the case for the years as a freelancer variable where the mean is 8,8 years. Finally, the average income was between 34.000 and 35.000 euro´s of after tax

(29)

income. In order to see how these variables relate to each other a correlation table was created, which can be found in table 2 below.

Table 2: Inter-correlations and descriptive statistics

1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. Mean SD 1. Ambidexterity 1,00 -,10 1,45 2. Exploration -,02 1,00 ,00 1,08 3. Exploitation ,18*_{-,08 1,00} _,00 _1,13 4. Gender -,04 ,05 -,03 1,00 1,50 ,50 5. Age ,02 -,07 ,01 -,05 1,00 4,56 ,95 6. Education -,04 ,15 -,15 ,04 -,07 1,00 5,15 ,70 7. Marital status ,12 ,01 ,10 ,25** -,12 ,02 1,00 1,80 ,88 8. Working sector ,00 ,01 ,08 ,12 -,02 ,21* ,03 1,00 6,73 2,75 9. Years experience ,07 -,15 ,08 -,07 ,68** -,11 -,03 -,06 1,00 18,40 8,65 10. Years Freelancer ,12 ,03 ,07 -,13 ,54** -,17 -,02 -,20* ,62** 1,00 8,78 7,72 11. Workarea ,05 ,07 ,08 ,07 ,02 ,23** ,12 ,29** ,08 -,03 1,00 ,74 ,44 12. Income -,20* -,04 ,01 -,05 ,04 ,15 -,15 ,13 ,14 -,15 ,11 1,00 3,25 ,98 **. Correlation is significant at the 0.01 level (2-tailed).

*. Correlation is significant at the 0.05 level (2-tailed).

There are some logical things that can be found in this matrix, like the strong correlation between age and years of experience and years as a freelancer. This strong correlation has its consequences for the regression analysis. Because there is a strong correlation between those variables they could be measuring the same thing, which could in turn lead to

multicollinearity. I therefore chose to include only one of these three variables in my regression analysis, namely years of experience.

Also interesting is the correlation between the respondents’ level of education and working sector, and the respondents’ working area. This could hint that educated freelancers work in the Randstad and that they tend to work in the same sector, but when making

statements about this correlation one has to keep the frequencies in mind. The fact that by far most respondents in this sample work in the Randstad and most of them are highly educated could give a biased image. Also interesting is that exploitation is significantly related to ambidexterity while exploration is not, for which I have no explanation. The most important

(30)

correlation in this matrix is the correlation between ambidexterity and income. This

correlation means that there is a significant effect between the independent and the dependant variable in my thesis.

4.3 Hypothesis testing

As mentioned in the methods chapter I used multiple regression analysis to answer my hypotheses. In order to answer my first hypothesis “ambidextrous freelancers perform better than non-ambidextrous freelancers” I created six models to test if ambidexterity was actually a significant predictor of performance. In model 1 of this hierarchical regression analysis all the control variables were put in and in model 2 the centered exploration and exploitation variables were added. After that I chose to analyse the interaction terms in separate models because when including them all in one model every variable that was significant before became insignificant. This could be because of multicollinearity, or because the three last interaction terms are variables that are combined with ambidexterity, which is already an interaction term. When looking at the interaction terms separately, it is possible to see which interaction term has the most explanatory power. The results of this analysis can be found in table 3 below.

As can be seen at the bottom of the table the predictive power of the models increased when the independent variables and interaction terms were added. The variability in the dependent variable that can be accounted for (R-square) first went up from ,095 to ,099 when the exploration and exploitation variables were added. When ambidexterity was added the R-square further increased to ,135. The insignificance of the ambidexterity*experience

interaction term shows that while years of experience is a significant and positively related predictor of income in every other model, the combination of ambidexterity and experience does not increase the explanatory power. The last two models further enhanced the R-square to ,139 and ,148 respectively.

(31)

Table 3:Results of Multiple Regression Analyses (Ordinary Least Squares)

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6

B Sig. B Sig. B Sig. B Sig. B Sig. B Sig.

Control variables Gender -,032 ,855 -,024 ,894 -,048 ,782 -,017 ,923 -,018 ,919 -,015 ,931 Education ,201 ,117 ,211 ,110 ,204 ,116 ,213 ,108 ,224 ,084 ,191 ,138 Marital status -,189 ,064 -,187 ,070* -,137 ,174 -,188 ,068 -,164 ,102 -,131 ,190 Working sector ,034 ,301 ,032 ,331 ,029 ,370 ,032 ,330 ,018 ,579 ,027 ,401 Years of experience ,028 ,043** ,027 ,055* ,017 ,082* ,016 ,120 ,020 ,045** ,020 ,049** Working area ,120 ,568 ,122 ,562 ,136 ,509 ,100 ,635 ,146 ,479 ,101 ,623 High levels ,023 ,904 ,127 ,651 ,288 ,311 ,029 ,919 -,003 ,991 ,278 ,320 Independent variables Exploration -,073 ,468 -,109 ,273 -,056 ,577 ,028 ,787 -,094 ,340 Exploitation ,016 ,874 -,015 ,877 ,015 ,879 ,107 ,316 -,014 ,883 Interaction effects Exploration*exploitation (Ambidexterity) -,153 ,014** Ambidexterity*years of experience ,008 ,208 Ambidexterity*highlevels -,283 ,010** Ambidexterity*workarea -,187 ,005** R-square ,095 ,099 ,135 ,102 ,139 ,148 Adjusted R-square ,036 ,024 ,063 ,027 ,067 ,077

Notes: Dependent variable: LogIncome; N=131; *p<0,10; **p < 0,05

(32)

When ambidexterity was included in model 3 the regression analysis showed

ambidexterity to be a significant predictor of performance (P=0,014). However surprisingly, the Beta that pertained to this relation was negative (-,115) which means that those freelancers that scored high on ambidexterity actually scored lower on income in general. This means that H1: “The relation between ambidexterity and performance is positive” is rejected.

As expected gender did not make a difference. Less expected was the insignificance of education (p=0,116). Working sector was also not a significant predictors of income (P=,370) as well as working area (P=,509). Years of experience did have predictive power that was significant on the 0,10 level (P=,082) and was shown to be positively related to income, which could be expected.

After this, the second hypothesis: “The relation between ambidexterity and

performance can be explained by those ambidextrous freelancers that have high levels on both exploration and exploitation.” was tested. To do this, the new variable “highlevels” was computed that gave freelancers with a mean of both exploration and exploitation that was higher than a 5 (on a seven point scale) a 1, and every respondent below that a 0. This

highlevels variable was then multiplied by the ambidexterity variable to create the interaction term that can be found in model 5 of table 3. In this interaction term only the ambidextrous respondents that scored high on both exploration and exploitation received their ambidexterity score and I could therefore test whether the “highlevels” respondents were responsible for the (negative) effect of ambidexterity on performance. With a P-value of 0,01 this interaction term has a higher predictive power for income than ambidexterity. The results also show that the Beta decreased even further; from -,153 to -,283. This, together with the P-value, means that the negative relation between ambidexterity and performance can indeed be partly explained by those respondents that score high on both exploration and exploitation. To double check this the SPSS file was split on the high levels variable, and another

(33)

regression analysis was done like the one stated in model 3 of table 3. The results of this regression analysis can be found in table 4 below. The table shows that ambidexterity as a predictor of performance was not significant for freelancers with low or average levels of both exploration and exploitation (P=0,348) while this relation was very significant for freelancers with high levels of both exploration and exploitation (P=0,01). Like before, the (negative) effect of ambidexterity on income was stronger in the high levels group (B=-0,245) in

contrast to the low or average group (P=-,077), and finally the R-square strongly increased in the high levels model. These findings are in line with the former analysis and I can therefore conclude that the second hypothesis is confirmed with the side note that it pertains to a negative relation instead of a positive relation.

Table 4: regression analysis split by highlevels variable.

High levels No Yes

B Sig. B Sig. Gender -,129 ,551 ,369 ,245 Education ,212 ,168 ,074 ,758 Marital status -,208 ,083 ,083 ,639 Working sector ,033 ,395 ,070 ,348 Years of experience ,022 ,071* ,019 ,270 Working area ,209 ,391 -,702 ,198 Ambidexterity -,077 ,348 -,245 ,010** R-square ,134 ,321 Adjusted R-square ,065 ,151

Finally, hypotheses 3a and 3b were tested. To get a quick impression of the difference in ambidexterity and exploration between living in the Randstad, or outside the Randstad I splitted the SPSS file on the work area variable and computed the means of ambidexterity and Exploration. This showed that freelancers working in the Randstad are more ambidextrous, and pursue explorational activities to a higher extent than freelancers that are not working in

(34)

the Randstad. This can be seen in table 5. However, this does not say anything about the reliability of these numbers or their effect on performance. Therefore an interaction term consisting of the ambidexterity variable and the workarea variable was created after which it was included in the regression analysis which can be found in model 6 of table 3.

Table 5: Exploration and ambidexterity compared by working area.

Work area Exploration Ambidexterity

Not working in Randstad (Mean) 4,2702 -,2112

Working in Randstad (Mean) 4,4489 -,0614

Model 6 of table 3 shows that the predictive power seemed to have increased in the

ambidexterity*workarea interaction with a P-value that went from ,014 to ,005. This shows that the relation between ambidexterity and performance is in fact somewhat moderated by whether or not the freelancer works in the Randstad. Also, the Beta shows that the strength of this relation is a bit stronger as it went from a Beta of -,153 to a Beta of -,187. This means that ambidextrous freelancers working in the Randstad actually performed somewhat worse than ambidextrous freelancers that do not work in the Randstad. Therefore H3a: “The relation between ambidexterity and performance is stronger in the Randstad than outside the Randstad” is confirmed, again with the sidenote that it pertains to a negative relation.

Hypothesis 3b: “Exploration is more positively related to performance for freelancers working in the Randstad than for freelancers that do not work in the Randstad ” was tested in two ways. First I tried to test the effect of exploration itself on performance by splitting the SPSS file on the workarea variable and doing a regression analysis like the one in model 3. This resulted in the coefficients that can be found in table 6. Unfortunately, exploration was found to be insignificant at a P-value of ,199 outside the Randstad and ,499 in the Randstad, so no statements can be made whether scoring higher on exploration in the Randstad has a different effect on performance than for freelancers that are not working in the Randstad. Even though this answered hypothesis 3b I wanted to exclude the possibility that this

(35)

hypothesis was still true for those extreme cases scoring higher on exploration than on exploitation. For that I created a dummy variable that gave a 1 to those respondents that scored 1,5 times higher on exploration than on exploitation to see whether more exploration focused freelancers scored better than ambidextrous freelancers. Unfortunately, this also didn’t come up with an answer to hypothesis 3b as the P-value of this dummy variable is ,975 as can be seen in table 7. Therefore this hypothesis can neither be confirmed nor rejected.

Table 6: Regression analysis - split by working area

Working area Not Randstad Randstad

B Sig. B Sig. Gender ,158 ,674 -,060 ,774 Education -,065 ,734 ,303 ,076 Marital status -,286 ,194 -,131 ,272 Working sector ,019 ,804 ,029 ,444 Years of experience ,049 ,003** ,008 ,581 Exploitation ,005 ,978 ,048 ,591 Ambidexterity ,255 ,124 -,159 ,021** Exploration ,249 ,199 -,062 ,499 R-square ,343 ,136 Adjusted R-square ,133 ,057

Table 7: Regression analysis including dummy variable - split by working area

Working area Not Randstad Randstad

B Sig. B Sig. Gender -,053 ,884 -,065 ,754 Education -,100 ,614 ,274 ,101 Marital status -,096 ,629 -,121 ,312 Working sector ,001 ,988 ,031 ,402 Years of experience ,044 ,003** ,008 ,549 Ambidexterity ,335 ,099* -,158 ,030** Explor1.5*Exploit ,941 ,230 -,022 ,952 R-square ,334 ,128 Adjusted R-square ,154 ,059 34

(36)

5. Discussion and limitations

By testing the four hypotheses I tried to get an understanding of how ambidexterity affects performance on the level of analysis of the individual, and more specifically freelancers. Some findings contradict theory while there may be explanations for this and other findings support theory while having possible biases that should be discussed. I will therefore elaborate on these findings in this section.

With regards to the the first hypothesis the findings showed that ambidexterity is a significant negative predictor of performance. This was not expected as the meta-analysis of the ambidexterity-performance relation by Junni et al. (2013) showed that combined

organizational ambidexterity was positively and significantly associated with performance. Combined means that ambidexterity was measured by a multiplication of separate exploration and exploitation scales, which is what I did. My findings therefore contradict these theories, but one has to realize that most of these studies were done on higher levels of analysis such as the organizational, business unit or team level. Very few studies were done on the individual level (Junni et al., 2013). Cao et al. (2009) do state that managers in resource constrained contexts may benefit from managing tradeoffs between exploration and exploitation, while managers in firms with sufficient resources should simultaneously pursue exploration and exploitation. However, the last part of that statement is not applicable to freelancers as they don’t have multiple employees that can work on exploration instead of them let alone sufficient resources to do it themselves (think about time and income). This automatically means that the first part of their statement is true for freelancers; they have to manage the tradeoffs between exploration and exploitation. But this says nothing about balance that should be aimed for when managing this tradeoff. Concrete evidence that ambidextrous managers perform better or worse than non-ambidextrous managers is thus lacking, and studies of this effect on freelancers are non-existent. This is why more research should be

(37)

done on the individual level on analysis in order to find support for my findings and to know whether ambidexterity is beneficial or harmful for managers or freelancers.

Besides ambidexterity the variable pertaining to a freelancer’s years of experience was also significantly related to performance. In order to get a better image of how these two significant predictors relate to each other I plotted them against each other. The result is graph 13 below, which shows that ambidexterity has a negative effect whether someone is

experienced or not. The experience does however take the whole ambidexterity-performance relation to a higher level of income when it increases. When looking at this graph one has to keep in mind though that the interaction term of these variables (model 4, table 3) has an insignificant P-value, which may create a wrong image.

Graph 13: Ambidexterity and Experience plotted against Income.

The analysis of the second hypothesis showed strong evidence that the ambidexterity-performance relation could indeed be explained by those ambidextrous freelancers that have high levels on both exploration and exploitation. However, one has to keep in mind that the analysis that was used to answer this hypothesis could be biased because the high levels group

The ambidextrous freelancer : understanding the effects of ambidexterity on performance for freelancers working in the Dutch creative industries