Determinants of Decentralization within Software Development

Determinants of Decentralization

within

Software Development

Rens van Eijk

Amsterdam School of Economics (ASE) University of Amsterdam In collaboration with No Nonsancy

Author Notes

This master thesis was produced for the master managerial economics and strategy within the master business economics of the University of Amsterdam. Teachers: J. van de V and S. Dominguez-Martinez. Student: R. van Eijk, student number: 11371919. For correspondence concerning this thesis please contact R. van Eijk at rensveijk@gmail.com.

This document is written by Rens van Eijk who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

Index

Abstract

This thesis is written to acknowledge an empirically construct the determinants for the degree of decentralization. A survey under ICT entrepreneurs in the Netherlands is conducted to determine if there is a relationship between the degree of decentralization and 1. information asymmetry, 2. conflict of interest and 3. skills. The results from this thesis suggest that the information gap positively correlates with more delegation. The other two determinants do not find significant results. Furthermore, delegation is found to be decision specific. Meaning that under strategic decisions the firm might be very centralized, while at the same time decentralized under product decisions. [101]

Keywords: Software development, decentralization, skills, conflict of interest,

Introduction

We are living in a fast-moving world with technology improvement on a daily basis. To illustrate this Gordon Moore, the cofounder of Intel introduced Moore's law. Which indicates that the number of transistors (i.e. semiconductor devices used to amplify or switch electronic signals and electoral power) in a circuit doubles every two years. Without making this thesis to technical this means our digital technologies improve exponential. The development of organizations on the other hand takes a lot of time. One reason for this slow development is due to strong cultures within firms. Organizational cultures can be seen as an "imprint" of historic behaviour from people within the firm that stipulate its values, beliefs and principles. When persisting in the same form can result in remarkable levels of inertia (Marquis and Tilcsik, 2013). Another explanation comes from Fred Brook, who wrote the book 'The Mythical Man-Month', a software engineering book in which he explains that adding more people to a project not necessarily results in faster or more efficient progress but instead slows it down. This is known as Brook's law, saying; 'adding more people to a late project makes it later' (Brook, 1974). New people first have to get familiar with the project before they can add value and in the short run explaining does not help increase the productivity, but in the contrary it decreases. Comparing the exponential growth of technical development and the slow change within organisations lead to a gap, stipulating that its hard for firms to cope with this fast developing and changing world. This asks for a change of strategy where firms can deal with this change and adapt. Software development was usually done on project basis by making use of the waterfall method. This basically comes down to a process that flows from top to bottom, while going through multiple stages like analysing, design, construction and implementation which dates back to the 80's. So, is this still the best way to work within organisations? More recent used ways of software development are called 'Agile working', where emphasis is laid on the end product with continues feedback loops to the customer and other stakeholders (Gerrits, et al. 2013). This ensures that fast changing technologies can easily be adapted within the projects process and might be the solution to overcome the increasing gap between technology and organisations.

The Agile method can be seen as an iterative and incremental approach of software development and is used in more organizations everyday. Big enterprises like Spotify, Google, Netflix and ING are using this type of project development and most commonly used frameworks are; Scrum, Kanban, Lean and Prince2 (ING, 2015). These frameworks are using

Agile working typically focuses on the inner structure of organisations and is thus closely related to topics within the organizational architecture theory. Questions about who has the right to decide within a company and who has the effective control over decisions are asked in both. For example in organizational architecture theory Aghion and Tirole (1997) define formal authority and real authority within a firm. Formal authority can be seen as the ultimate decision right a principal (e.g. the CEO) of a firm has. One can imagine that the tasks of a CEO are piling up his desk every day and this might lead to rubberstamping of certain projects or (important) decisions, i.e. giving the decision rights to some one else within the company. In this case the employee in question has the real decision rights within the company. This real authority can be seen as a way of decentralization of the firm. Decision rights are delegated to the agent (manager), who potentially possesses more information. Whereas in more centralized firms the decision rights are always in the hands of CEO. The research question to investigate the determinants for the degree of decentralization in this thesis is:

Research question

Does a relationship exist between specific firm characteristics and the allocation of decision rights in small/medium sized firms?

The specific firm characteristics might be seen as determinants for the degree of decentralization. I will mainly focus on the information gap (i.e. who has the information within the firm? Is this the principal or the agent), the conflict of interest between the principal and the agent and the skills of the employees. The rest of the thesis is structured as follows: After this introduction the hypotheses will be discussed by mainly making use of the organizational architecture theory (Theoretical framework). In the Methodology is explained why certain choices were made considering the research methodology. Where after I will present a brief overview of the Data followed by an interpretation and in the

Conclusion section I conclude the thesis with general comments, a discussion and

Theoretical Framework

Decentralization can be defined as 'the delegation of power from a central authority to regional and local authorities' (Merriam-Webster, 2017). In this thesis I empirically test determinants for this delegation of power. Where a distinction is made between three determinants for decentralization. 1. The information asymmetry between the employer and employee, 2. The conflict of interests between the employer and the employee and 3. The skill level of the employee.

Information asymmetry

At the basis of my thesis lays the principal-agent theory. When I speak of the principal in this thesis I mean the employer or entrepreneur who started the company or is in charge of the company. The agent in this thesis refers to an employee/developer within the company. When considering the principal-agent theory, the principal and the agent can cooperate to create mutual benefits. As an example we can think of a developer with a great idea but not the funds to realize it and an entrepreneur who has the money but not the idea. By working together these two parties can realize the idea and gain from benefits like a return on the investment or profit for the entrepreneur when a new enterprise is created. To motivate the developer to invest effort he or she needs to be incentivized and this can be done by giving the developer a share of the profit in terms of a salary. Principal-agent theory tries to overcome these conflicting objectives by certain typical elements. The principal is the person that offers the contract to the agent. Where after the agent can either accept or reject the offer and afterwards makes decisions on e.g. the amount of effort dispensed and which project to choose. Within this structure the objectives are not fully aligned and need to be written down in a contract between the principal and the agent.

To define who has authority within this newly created enterprise I look at Aghion and Tirole (1997). They define authority as the right to make decisions concerning the use of the new enterprise. Where a separation is made between formal authority and real authority. Formal authority can be seen as the ultimate decision right over a specific asset. The owner (entrepreneur) of this asset (new enterprise) has these rights and can decide to leave the control over the enterprise to the developer. In this case the entrepreneur gives the real authority over the enterprise to the developer. Please note that the ultimate control over the enterprise resides at the entrepreneur (i.e. the entrepreneur still has the formal authority). In the analysis of Aghion and Tirole (1997) they suggest that delegating authority to the agent

of control over the choices made by the agent. When the agent is more willing to participate in the contract and shows more initiative to perform a task a logical consequence is a gain in knowledge. This might result in an information asymmetry, where the agent is better informed about a task or project than the principal. The agent can use this information and take advantage of the uninformed principal. Aghion and Tirole (1997), describe this information gap and argue that when the knowledge gap between the principal and the agent becomes bigger there tends to be more delegation.

The article of Aghion and Tirole (1997) gives an example of these situations by illustrating three projects. Where project 1 is better for the principal, project 2 is better for the agent and project 3 is very unprofitable for both (Note: The parameters α and β range from [0,1].

Table 1). Since the agent has better knowledge about the situation he gets real authority of the principal and may select in which project the company will invest. In this situation, it might be the case that the agent prefers a different project than the principal. Knowing this fact, it is still better to give the decision rights to the agent. Since when the principal decides which project to invest without gathering information about the projects, the chance of selecting project 3 (the very bad outcome) has a probability of !_! . This could unlatch a conflict of interest where the agent can select the (for him) best outcome of the project in terms of reward in relation to work effort. Thus, the authors argue that when decisions (or activities) are relatively unimportant for the principal, authority is more likely to be delegated to the agent.

Project Profit Principal Profit Agent

1 B !b

2 !B b

3 Very bad Very bad

Note: The parameters α and β range from [0,1].

Table 1 Profit indication for specific project

This work is further developed by Baker, Gibbons and Murphey (1999) in which the authors argue that authority can not be formally delegated form the principal to the agent, as decision rights stay with the principal. However, it is possible to informally delegate the decision rights to the agent through self-enforcing rational contracts. Meaning persons higher in the hierarchy

of a company overrule the decision. In their work Baker, Gibbons and Murphey (1999) examine the feasibility of informal authority and show that the agents and principal alike are tempted to select short term beneficial projects over long term beneficial contracts. They illustrate this by describing two models. A model where the boss is informed about the project and 2. A model where the boss in uninformed about the project. In the former model the boss ratifies all projects the agent proposes, even though it might be the case that this is not in the best interest of the principal. This informal way of delegation is feasible if the principal values her reputation for delegating authority over what she would gain from breaking the promise to ratify all projects. The latter model the principal does not have the necessary information to assess the project, but does observe the results after from the project when the it has been implemented. When the principal is uninformed, and delegates the project choice to the agent, the agent might select a project that is good for him but not beneficial for the principal. The principal will notice this project selection afterwards and from then on have less trust in the agent. This distrust results in a shift in authority decisions of the principal, who from now on will choose centralization. When the agent selects a project that is beneficial for the principal. The principal is satisfied and keeps on delegating the authority to the agent.

Both in the article of Aghion and Tirole (1997) and in the article of Baker, Gibbons and Murphey (1999), the size of information asymmetry or so-called information gap indicates whether or not the principal delegates the decision rights to the agent. A bigger information gap, i.e. a better-informed agent, results in more delegation of authority to the agent. This theory is empirically tested by several articles. For example, Colombo and Delmastro (2004) distributed a questionnaire under plant managers of Italian metalworking plants and found that when the plant's organisation is complex, the superior's information regarding the plant's internal activity become limited. This results in a rise in the stimulus to delegate authority to the plant manager. Another study by Abernethy, et al. (2004), conducted a survey under 78 firms listed on the Amsterdam Stock Exchange again found a positive relationship between information asymmetry and decentralization. Within the delegation process it is important to realize that the decisions within the firm come from individuals not simply form the firm as a hole (Graham, et al. 2015). The authors found that directors of firms are less likely to delegate decision rights when they are knowledgeable about the subject. Combining the findings of the studies above I developed my first hypothesis.

There will be a positive relationship between the information asymmetry and the level of decentralization.

This information asymmetry can thus be seen as a determinant of decentralization and based on the above literature I expect a positive correlation. The degree of delegation is not monolithic (Graham, et al. 2015). Meaning that the degree of delegation varies across different decisions. For example, the principal might delegate decisions when more input is needed from their developers and less when their information advantage is bigger.

Conflict of interest

Apart from the information asymmetry Aghion and Tirole (1997) and Baker, Gibbons and Murphey (1999) speak of a conflict of interest. By this they refer to an agent and principal wanting to select different projects based on their beneficial gains from these projects. This conflict of interest can be seen as heterogeneity or a weak alignment between the boss and the agent. Looking at the influences of heterogeneity within an organisation, I found some interesting articles. For instance, when leaders of the teams are either very informed or not informed at all, leaders prefer a heterogeneous team composition. As they benefit from less correlation in the information they receive from the team (Mello and Ruckers, 2006). Although, to work together there must be some sort of homogeneity in order to agree with each other (Marinez-Moyano, 2006). On the other hand, Hamman and Martinez-Carrasco (2015) show that a decentralized organization increases the likely hood of heterogeneity within the teams. They test this in an environment where the manager needs to determine the allocation of decision rights and has to choose her team composition. Dominguez Martinez and Sloof (2017) show that when there is a weak alignment between the principal and the agent this ensures a more restricted form of delegation, i.e. when the heterogeneity between the entrepreneur and the developers is larger this results in more centralization. In an even broader perspective Acemoglu, et al. (2006) finds a positive relationship between a more heterogeneous environment and more decentralized firms. Intuitively they argue that when the environment is more homogeneous, more information about the recommended actions are known making the firm more responsive to public information. While when the environment is more heterogeneous less information can be found publicly and greater reliance on the employees is required. Meaning the emphasis should lay on delegation, making the firm more decentralized. Heterogeneity is a determinant of decentralization and determines the level of alignment or conflicting objectives between two parties.

This conflict of interests can be linked to our situation between the entrepreneur and the developer and the informal authority within organisations. The developer is better informed about the possible projects and can (when authority is delegated) make the choice for a certain project. This however might not be the most efficient outcome for the entrepreneur. The entrepreneur, who seeks for the highest profit from his investment, lets the developer invest as much work effort as possible. The developer on the other hand wants to receive incentives (e.g. salary, rewards or bonuses) for the accomplished tasks with as minimal work effort as possible and to stay motivated. To verify that the developer is investing work effort the entrepreneur can choose to monitor the developer. As a result, the developer puts in less work effort because he senses the feeling of being monitored. This theory is called the crowding-out theory (Falk and Kosfelt, 2006). The authors test this by letting principals set different minimum performance requirements for the agents where after they may choose their effort level and see that many agents exhibit control-averse behaviour. Meaning a higher minimum performance requirement set by the principal results in a lower amount of effort given by the agent. The other way around when monitoring does not influence the motivation of the employee this is called the disciplining effect. The nature of the employment relationship defines whether the crowding-out effect or the disciplining effect applies, i.e. they argue that in a smaller setting the relationship between the principal and agent become more personal. This smaller setting results in a decrease in effort when monitored and thus crowding-out theory applies. Dickinson and Villeval (2008) show this by conducting an experiment where the agent had to perform a motivational task and the principal could monitor the agent. The different treatments where 'distant' and 'interpersonal', in the former principal and agent where matched as strangers and assigned anonymously. In the latter treatment matches where linked with each other for 10 rounds to develop an interpersonal relationship. The crowding-out effect can best be defined as an increase of monitoring, which can be perceived as distrust by the agent and this induces them to reduce work effort (Frey and Reto, 2001). An example is for instance when a principal decides to install cameras on the work floor to monitor the agents, this extra form of control is perceived as distrust by good agents and influences their behaviour by working less hard. The disciplining effect can best be described as an increase in the work effort by monitoring (Frey, 1993).

In a quick recap, there are two contradicting theories; the crowding-out theory and the disciplining theory. When the crowding-out theory applies monitoring decreases the motivation of the employee. While Aghion and Tirole (1997) describe that delegation of authority increases the incentive to acquire information (i.e. increasing the motivation of the employee). I tried to capture the heterogeneity and unmotivated employees in one determinant for decentralization, naming it conflict of interests. When the entrepreneur and the developer become less aligned, they become less reliant on each other making the firm more centralized.

Hypothesis 2

When there is more conflict of interests between the entrepreneur and the developer within the firm, this is positively correlated with less decentralization.

This conflict of interests, might influence the delegation within a firm and by making use of the above literature I expect a negative relationship. Meaning an increase in the amount of delegation within a firm can be negatively correlated with the heterogeneity between the entrepreneur and the developers.

Skills

Within the literature another determinant for decentralization is skills. Overall the global level of education is growing over time. This results in an increase in high skilled workers across the world (OECD, 2017). Combine this with the fact that a Tayloristic approach (i.e. a highly structured way of working with imposed tasks) makes room for more decentralized types and forms of organisation. This might suggest that the skills of the labour force are positively linked to this Organisational Change (OC) (i.e. a change from centralization to a more decentralized organization). Caroli and van Reenen (2010) refer to this as the 'Skilled-Biased Organisational Change' hypothesis. Where an increase in the number of skilled workers in a firm increases the possibility of a OC. The connection mainly stamps from the fact that skilled workers have a higher ability to handle information and perform better in general. This makes them better prepared to take on responsibilities and deal with uncertainty, aspects that typify an OC. Caroli and van Reenen (2001) test this by analysing the influence of OC on workers with different skill levels in a British and a French dataset. The authors define an OC by the delayering of managerial functions and the decentralization of authority and argue that organizational changes reduce the demand for unskilled workers. More support for this link can be found in multiple articles, e.g. Greenan and Guellec (1998) show a positive link between skills of the labour force and the level of autonomy of the workers. By analysing the

combined French datasets of a labour force survey amongst blue collar workers and a business survey conducted by 7089 firms. Further research of Greenan (2003) again finds support for the positive link, this time by conducting a survey in France manufacturing firms during 1993. Not only OC is examined, but often the link between technological change and technical excellence is also emphasized. Though the OC is found to have stronger effects then a TC (Piva, Santarelli and Vivarelli, 2003; Caroli and van Reenen, 2010). Based on the above discussed academic papers I expect a positive relationship between skilled workers and the decentralization of a firm i.e. when the skill level of the employees is higher this correlates with a more decentralized firm.

Hypothesis 3

More skilled workers within a firm can be positively correlated to more decentralized organisation.

Methodology

To form a data sample, I made use of multiple channels. This will first be discussed, where after the structure of the questionnaire is explained. The questionnaire consists out of four parts each using questions from different academic articles. These questions are used to construct the variables to test my hypotheses. The construction of each variable will separately be discussed. Where after the structure of the model is explained.

To answer the research question and test the hypotheses, I developed an online questionnaire where mainly ICT entrepreneurs are asked to answer questions about their firm. This is likewise the target audience of the questionnaire. Specifically, the ICT sector is chosen because I make use of the SAMI. This test emphasizes Agile working, which is mainly seen in software firms. To test for decentralization within the firm I made use of Graham, et al. (2015). Within this empirical article the authors use a dataset about executive decision making. CEOs and other high-level managers were asked to answer who made the decisions upon five different important decision topics. I used the same perspective by focusing on the entrepreneurs of the companies. The target audience is fairly specific, i.e. not everybody is suited to participate in the survey. This distribution of the survey is done via the use of specific channels, being: 1. A LinkedIn campaign, where I could target specifically at SME ICT entrepreneurs. Those receive a so called 'In mail' (message via LinkedIn) with the question if they want to participate in the research. 2. Via an EBook written and distributed by No Nonsancy called 'Waarom sommige ICT-ondernemers succesvol zijn en anderen blijven ploeteren.' which translates to 'Why certain ICT-entrepreneurs are successful and others keep slogging.', specifically written for the target audience. This EBook is made public on the website of No Nonsancy and to download a copy you need to write down your contact information. All this data is collected in a database, which contains 68 contacts. These contacts where reach via mail with information about my research, followed by a call to participate in the survey. This way of gathering participants should be handled with care, because all participants that downloaded the EBook before participating in the survey might potentially be seeking for help. Which gives a biased result. To cope with this problem I also used other channels. 3. By distributing the survey on an ICT conference. The 31st of May an ICT conference was held in Utrecht, which was called the conference for Electronics and Applications. Here I asked exhibitors to fill the survey on the spot. 4. By asking all

ICT-entrepreneurs and CEOs within the community of No Nonsancy to fill the survey and share the link within their ICT-network.

When looking at the industry information of ICT firms within the Netherlands, 95% has less then 10 employees and is according to the law denoted as SME (Rabobank, 2017; European Commission, 2015). For the precise definition of SME please look at Note: The threshold for each category is conform the law of the European Commission (2015). Table 2.

Enterprise Category Number of employees Annual revenue

Medium < 250 < 50 million euro

Small < 50 < 10 million euro

Micro < 10 < 2 million euro

Note: The threshold for each category is conform the law of the European Commission (2015).

Table 2 Enterprise Category

The questionnaire was made with the use of SurveyMonkey, this is an online survey medium that allows participants to fill-in the survey by making use of the following link: https://nl.surveymonkey.com/r/softwareontwikkeling and consists out of four parts: 1. General data of interviewee. Think of questions about firm size, tenure, historic and forecasted firm growth over a time span of three years and education. 2. Questions about the structure of the firm. These questions are used to determine the level of centralization or decentralization when it comes to decision making. 3. Randomized questions measuring the three independent variables to test for the personal relationship between the principal and the agent and the skills of the employees. 4. Questions concerning who is better informed in relation to certain topics to test for information asymmetry. Via the link above, the survey can be viewed online. For an overview of the entire survey, please view Appendix IV.

The survey uses questions taken from previous research to give more validation and to make comparisons with previous studies possible. The questions in the survey are based on three other surveys. Questions in part two about the way the firm is structured, are taken from Graham, et al. (2015) and Gordon and Narayanan (1984). In which the participants need to indicate who takes the decisions on specific topics (e.g. Strategy or investments). As my

For example, in Graham, et al. (2015) one of the decision situations is the allocation of capital across divisions. I decided not to include this question due to the fact that some SMEs do not have multiple divisions. To make the survey more suitable for software developing ICT-entrepreneurs, I incorporated decision situations which are highly common in those firms. For example, 'Who is taking the decisions concerning the usability of the software?' and 'Who is taking the decisions concerning the architecture/design of the software?'.

In 2001 the Manifesto for Agile Software Development is drafted where twelve principles structure and define Agile working (Manifesto, 2001). To test how Agile your organisation or project is or to see if your firm is Agile ready. Sidky (2007) developed a test called the Sidky Agile Measurement Index (SAMI) in which he restructures the twelve principles and converges them to the five most important principals, being: Technical excellence, customer value, human centeredness, frequency and collaboration. Since the questions of the SAMI test measure the characteristics of an organisation and the variables skills and conflict of interest can also be seen as characteristics of an organisation. The questions for these variables are based on the SAMI test developed by Sidky and corresponding to the formulated hypothesis explained in the theoretical framework section of this thesis.

In the final part of the survey, the information gap between the principal and the agent is measured. This is incorporated to test whether the data supports the theory developed by Aghion and Tirole (1997). To measure this the questionnaire of Abernethy, et al. (2004) was useful. The authors used specific questions to see who has the most information within the firm on specific topics. In this Thesis, the same type of questioning is used, although the questions are altered to be suitable for software development ICT-entrepreneurs.

Skills

The measure 'Skill' measures the skill level of the employees in the specific firm. I attempt to capture what Caroli and van Reenen (2001) refer to as 'skill-biased organizational change' by asking ICT entrepreneurs to indicate to what extent certain situations are applicable along a Likert-type scale of 1 to 5 where 1 = not applicable and 5 = applicable. The situations used to capture the skill-biased situations are taken from the SAMI test for Agility in an organisation (Sidkey, 2007). One of the question to measure skills is 'Your employees are professionals in their field.' please view Appendix V for the other questions. The variable 'SKILL' is the

average score of all questions regarding Skills. To see whether the questions for 'SKILL' measure the same a Cronbach alpha test is conducted. A threshold of 0.7 is used conform Gordon and Narayanan (1984), so when the Cronbach is above 0.7 the tested questions measure the same. The value is 0.72 so the questions for 'SKILL' measure the same.

Conflict of interests

The personal relationship measure is used to capture the personal relationship between the principal and the agent and the motivation of the employees (CONF). Which corresponds to one of the 5 principles from the SAMI test measuring collaboration (Sidkey, 2007). The questions related collaboration ask ICT entrepreneurs to indicate to what extent these situations are relevant along a Likert-type scale of 1 to 5 where 1 = not applicable and 5 = applicable. Since I want to measure conflict of interest, the scores on behalf of the questions regarding collaboration are reversed (i.e. when the participant answered 1 this is translated in to a 5). Assuming that there is a conflict of interest when the entrepreneur and the developer do not want to collaborate. When testing for the Cronbach alpha two questions where reversed by the program due to too far apart answers. The two concerned questions are Q27 'The customer must have the authority to directly have a say in what is being developed in any update or release.' and Q36 'The titles and functions of others intimidate employees in the organization.'. Question 36 can be experienced as a negative question and this might be the result of biased data. Question 27 is more about the relationship between the customer and the firm. Because of these reasons I took both questions out. After taking out these two questions the Cronbach alpha is 0.77 indicating that the questions to determine if there is a conflict of interest the between the principal and the agent measure the same. As for 'SKILL' the variable 'CONF' is the average score of all questions regarding conflict of interest.

Decentralization

To measure decentralization, I made use of existing theoretical and empirical research. Gordon and Narayanan (1984), Abernethy, et al. (2004) and Graham, et al. (2015) all measure decentralization. The approach of Gordon and Narayanan (1984) and Abernethy, et al. (2004) is to ask who takes decisions within specific situations and then take the average of all the questions to create one measure of decentralization. In Graham, et al. (2015) several decisions questions are asked but instead of taking the average of all questions they compare the determinants on all decision questions separately. Within this Thesis I use the technique of Abernethy, et al. (2004) and Gordon and Narayanan (1984), where the average of all decision

(2015). In this adapted version, I cluster decision questions that, according to other studies, score the same on the degree of decentralization. strategic and financial investment decisions usually cost a considerable amount of money and are therefore important for the entrepreneur. When decisions are important for the entrepreneur he or she will gather information. According to the theory better informed entrepreneur tends to take decisions in their own hands, i.e. decisions involving a large amount of financial resources are centralized (Colombo and Delmastro, 2004; Graham, et al. 2015). Colombo and Delmasto (2004) found that decisions regarding human recourses are usually taken by the floor manager. In my analysis, there is no middle or floor manager. So, it is hard to determine whether there will be more centralization or decentralization on this topic. Marketing related I found no evidence suggesting these decisions are more centralized or decentralized. This is why I decided to take these two decisions together. Information about the planning, architecture and user experience of software usually asks for a deep understanding of the matter. I expect the developers to have most information regarding those topics. Resulting in more decentralization for these decisions. Based on this explanation I divided the questions into three groups: 1. DECEN1 (strategy and financial investments), DECEN2 (marketing and human recourses) and 3. DECEN3 (planning, architecture and user experience). The measure asks who takes the decisions in certain specific situations occurring mainly in ICT companies (i.e. strategy, software development planning and the architecture of new software). Respondents rated decision choices on a five-point Likert-type scale where 1 = the ICT entrepreneur takes all the decisions and 5 = others take all the decisions. Since some decisions might be centralized and others more decentralized within the same firm it is not useful to do a Cronbach alpha test to see whether the questions measure the same.

Information asymmetry

The variable 'INFO' measures who (i.e. the entrepreneur or the developer) has most information regarding the specific topic. The topics are; strategy, planning, software architecture and user experience. These topics relate to the topics asked to test for decentralization and used the same structure as in Abernethy, et al. (2004). The questions related to the information asymmetry within the firm, ask ICT entrepreneurs to indicate to what extent these situations are relevant along a Likert-type scale of 1 to 5 where 1 = the developer has all the information and 5 = the entrepreneur has all the information and the average score of all topics results in the variable 'INFO'. The scoring of 'INFO' is the reverse of 'DECEN'. So, to compare and relate the two variables I reversed the answers regarding

'INFO' (i.e. when the participant answered 1 this is translated in to a 5). For the same reason as the questions for decentralization I again do not use the Cronbach Alpha to see whether the questions measure the same. Apart from the average score variable 'INFO', I divided the questions in three groups. For this division, I used the same reasoning as for the decentralization groups. Resulting in: 1. INFO1 (strategy), INFO2 (planning, architecture and user experience).

Control variables

The control variables are; tenure, firm size (i.e. number of employees), historical revenue and forecasted revenue. The same control variables are used as in other empirical studies regarding determinants of decentralization (Colombo and Delmastro, 2004; Abernethy, et al. 2004; Graham, et al. 2015). When looking at size in relation to decentralization all three studies named above indicate that bigger organizations tend to be more decentralized. Graham, et al. (2015) also looks at tenure and finds that an increase in tenure results in less decentralization. When looking at the growth variable Abernethy, et al. (2004) finds that an increase in the growth results in more delegation. For the control variable 'GROWTH' I took the historical and forecasted revenue (both on an three year scale) together. In Table 3 a summary of the control variables is given with a sign coefficient prediction based on the literature.

Control Variable Sign coefficient Source

Size + Colombo and Delmastro,

2004; Abernethy, et al. 2004; Graham, et al. 2015

Tenure - Graham, et al. 2015

Growth + Abernethy, et al. 2004

Table 3 Control variables and sign coefficient

Model

The hypotheses are tested using a OLS regression model that describes the determinants of decentralization. The empirical relation looks as follows:

!"#"$_! = !_!+ !_!!"#$$_! + !_!!"#$_! + !_!!"!"_! + !_!!"#$%"_! + !_!!"#$_! + !!!"#$%&! + !!Entre! + !!

I also look at the individual determinants 'SKILL', 'CONF' and 'INFO' as determinant of decentralization, with and without the use of the control variables 'TENURE', 'SIZE' and 'GROWTH'. When I consider the variables 'DECEN' and 'INFO', it is possible to see that specific topics score differently (i.e. the topic strategy might score a 1, where user experience might score a 5). This is why apart from the full model; another model is created where the variables 'DECEN' and 'INFO' are split in multiple groups. The variable 'ENTRE' is added to the model to control for the fact that in the sample both entrepreneurs and developers filled the questionnaire. The variable 'ENTRE' can be seen as a dummy variable, meaning the score of this variable will be '1' if the questionnaire is filled by an entrepreneur and indicates '0' otherwise.

Data

Variable Obs Mean (Std. Dev.) Min Max

Decentralization 42 3.06 (0.63) 1.00 4.00

Skills 42 3.90 (0.43) 3.17 4.75

Conflict of interest 42 2.26 (0.39) 1.57 3.21

Information asymmetry 42 3.30 (0.78) 1.50 5.00 DECEN1 (strategy and

financial investments)

42 2.68 (1.12) 1.00 4.50

DECEN2 (marketing and human recourses)

42 2.99 (0.98) 1.00 5.00

DECEN3 (planning, architecture and user experience)

42 3.37 (0.95) 1.00 5.00

INFO1 (strategy) 42 2.10 (1.12) 1.00 5.00

INFO2 (planning, architecture and user experience) 42 3.70 (0.96) 1.34 5.00 Annual sales 40 3.53 (1.18) 1.00 5.00 Tenure 42 13.17 (16.35) 0.00 100.00 Company size 41 34.34 (32.58) 1.00 100.00 Company growth 41 3.88 (1.07) 2.00 5.00 Male 42 1.00 (0.00) 1.00 1.00 Education 42 7* 4.00 8.00 Profession 42 1* 0.00 3.00 Sector 42 1* 0.00 1.00

Note: The variables indicated with a star (*) are categories. Thus, for these variables the median is given instead of the mean.

Table 4 Descriptive of the dataset

Note: The variables indicated with a star (*) are categories. Thus, for these variables the median is given instead of the mean.

Table 4 presents a summary of the most relevant variables. For a summary of the entire dataset please view Appendix I. The sample consists of 42 participants with an average education of an applied university, indicated by a median of 7. For the full list of possible education options see Appendix IV. The first thing that follows for the data is that on one exception all participants are males. This can be easily explained by the fact that the target audience is ICT entrepreneurs, which is an audience mainly consisting of men.

When moving on to the sector it is possible to see that most participants in the survey are working in ICT (indicated by a 1). When the respondents work at another type of firm this is indicated with a 0. The participants where asked to fill in what sector they worked. When looking at the received data (Appendix III), its possible to see that 23 participant filed a

When looking at the pursuits of hardware and electronics firms this mainly involves project based development of products. Meaning these companies are similar to companies in the ICT sector. In Appendix III all highlighted sectors are counted as ICT. Meaning 8 observations (19% of the sample) are not directly from the ICT sector.

The average number of employees working at the participating firms is around 34 and the average tenure of participants is 13 years. This latter fact is interesting, since the average existence of the participating firms is more then 23 years. The explanation for this difference is two-fold, the developers participating in the survey usually did not start working for the firm from the beginning or the entrepreneurs behind the firm took over the firm. When looking at the annual sales of the participants the average lays between the €1.000.000,- and the €5.000.000,- euro. For the historic growth and the expected future growth of the participating firms the average is between the 5% and the 10% annual growth in a time frame of three years. This indicates that the business sectors perspective is quite good. The average amount of annual sales and average number of employees of my sample relate to the laws that determine SME (European Commission, 2015). Considering the fact that 95% of the ICT industry consists out of SME, I can consider my sample a good reflection of reality.

Results

With a dataset of 42 observations I measure the data on skills, conflict of interest and where the information lies within the organisation and how they determine decentralization. First every variable is individually discussed in the result section. Where after the results of the model will be shown. After the model, the results of the individual analysis are discussed.

Skills

The first variable discussed is 'SKILL', which mainly concerns hypothesis 3, 'More skilled workers within a firm can be positively correlated to more decentralized organisation'. The corresponding questions from the survey for each variable can be reviewed in Appendix III. Overall the questions are answered with an average of 3.90, meaning the questions stated are mainly answered with 'often applicable'.

Conflict of interest

The second discussed is the variable 'CONF', which measures the personal relationship between the entrepreneur and the developers and mainly concerns hypothesis 2, 'When there is more conflict of interests between the entrepreneur and the developer within the firm, this is positively correlated with less decentralization.'. On average the questions measuring the personal relationship are answered with 3.74, meaning the questions where answered with 'often applicable'.

Decentralization

When looking at the summary of the data concerning the variable 'DECEN' what immediately be can seen from the data is that decisions concerning strategy and finance (DECEN1) are more taken by the entrepreneur (average is 2.68), while decisions concerning the marketing and human recourses (DECEN2) are taken together (average is 2.99) and decisions concerning software planning, architecture of the software and the user experience (DECEN3) are more taken by the developers (average is 3.37). For a full distribution please see Figure 1

and Fout! Verwijzingsbron niet gevonden.. To see if the different groups for decentralization match with the results of my sample I made a correlation table (Table 6). Here it is possible to see a high correlation between strategy and financial investments, a high correlation between marketing and human recourses and high correlations between the planning, architecture and user experience of software. From this I can verify that the division

of the Wilcoxon signed-rank test. Via the use of this test I can see that the three groups are significantly different. For the test results please view Appendix VI. The evidence of Figure 1

and Table 5 indicate that the degree of delegation is not uniform across the firm. Entrepreneurs are more likely to take decisions regarding strategy and investments, while tend to delegate decisions with regard to planning,

architecture of the software and the user experience.

Note: In parentheses are the standard deviations.

Mean N DECEN _3.06 (0.10) 42 DECEN1 (strategy and financial investments) 2.68 (0.17) 42 DECEN2 (marketing and human recourses) 2.99 (0.15) 42 DECEN3 (planning, architecture and user experience) 3.37 (0.15) 42 Table 5 Magnitudes of delegation in firms

Figure 1 Delegation within software development

Strategy Financial

investments

Marketing Planning Architecture User

experience Human recourses Strategy ₁ Financial investments 0.8291 1 Marketing _{0.5786 0.6360 1} Planning _{-0.1840 -0.1045 -0.0234 1} Architecture _{-0.2981 -0.3323 -0.2399 0.7537 1} User experience _{-0.2318 -0.1933 -0.0888 0.7719 0.8494 1} Hunan recourses _{0.2440 0.3294 0.5884 0.3270 0.1577 0.3406 1} Table 6 Correlations between decentralization questions Asymmetric information

Hypothesis 1 states 'There will be a positive relationship between the information asymmetry and the level of decentralization' this is mainly concerned with the information asymmetry between the entrepreneur and the developer and is denoted by the variable 'INFO'. It is immediately possible to see that the entrepreneur has more information regarding the strategy

Note: These graphs show the degree to which the respondent describe their involvement in decisions regarding strategy, investment, marketing, human recourses, planning, architecture and user experience. 'DECEN1' indicates delegation decisions regarding strategy and financial investments, 'DECEN2' indicates delegation decisions with regard to human recourses and marketing and 'DECEN3' indicates delegation decisions regarding planning, architecture and user experience of the software. A response of '1' means the respondent makes the decision solely by herself, while a response of '5' means that the decision is delegated to other with no involvement by the respondent.

architecture, user experience of the software and the planning ('INFO2', average of 3.70). To see whether 'INFO1' and 'INFO2' have the same distribution I made use of the Wilcoxon signed-rank test. Via the use of this test I can see that the two groups are significantly different. For the test result please view Appendix VI. To see if the different groups for 'INFO' match with the results of my sample I made a correlation table (Table 7). Here it is possible to see a high correlation between the planning, architecture and user experience of software. From this I can verify that the division of the two groups is right.

Strategy Planning Architecture User experience

Strategy 1 Planning 0.3077 1 Architecture -0.1531 0.5359 1 User experience -0.1277 0.4750 0.7042 1 Table 7 Correlations between information asymmetry questions Model

Table 8 presents the regression results. Model (1) regresses the skill determinant on decentralization, model (2) regresses the conflict of interest determinant on decentralization, model (3) regresses the information asymmetry determinant on decentralization, model (4) regresses all three determinants at once on decentralization, model (5) - (7) the control variables (i.e. growth, tenure and size) are added to the regression and finally in model (8) all variables are included jointly. Recall that the dependent variable (i.e. the degree of decentralization) rates decision choices on a five-point Likert-type scale where 1 = the ICT entrepreneur takes all the decisions and 5 = others take all the decisions. Meaning that the higher the value of the dependent variable the more decentralization there is within a firm.

When looking at the information asymmetry determinant within the full regression model (8) a positive and significant correlation is found with respect to decentralization. Meaning when there is more information asymmetry between the entrepreneur and the developer this can be seen as a determinant for decentralization. This result is significant on a significance level of 5% and under these circumstances with the control variables I can accept hypothesis 1 'The information asymmetry between the entrepreneur and the developer can be positively correlated with a more decentralized firm.'. This can be related to the theory developed by Aghion and Tirole (1997) explaining that an increase in the information gap between the principal and the agent would result in a increase in the amount of delegation and through the years this is emphasised and substantiated by several empirical studies (Gordon

and Narayanan ,1984; Abernethy, et al. 2004; Colombo and Delmastro, 2004; Graham, et al. 2015).

Model (8) finds a correlative relationship for the conflict of interest between the entrepreneur and the developer with respect to decentralization. Which is negative although insignificant on a significance level of 5% with a significant level of 32%. The same results (i.e. a negative correlative relationship between 'CONF' and 'DECEN') are found in the other regressions of Table 8, although not significant on a 5% significance level. This means that the results with regard to a conflict of interest between the principal and the agent are inline with the theory and hypothesis 2 'When there is more conflict of interests between the entrepreneur and the developer within the firm, this is positively correlated with less decentralization.' but I can not accept the hypothesis due to insignificant values. Recalling the theory, more homogeneity would result in trust (i.e. reputation of the developer) and this leads to more delegation (Graham, et al. 2015;)). Meaning that less alignment (i.e. more conflict of interest) between the principal and the agent would lead to less delegation (Dominguez Martinez and Sloof, 2017).

Model 1 2 3 4 5 6 7 8 Skills -0.01 (0.98) (0.93) 0.03 (0.60) 0.12 (0.25) 0.35 Conflict of interest -0.02 (0.94) -0.13 (0.74) 0.04 (0.88) -0.36 (0.32) Information asymmetry 0.16 (0.18) 0.17 (0.17) 0.21 (0.08) 0.26* (0.05) Growth 0.02 (0.77) (0.71) 0.03 (0.82) -0.02 (0.49) -0.06 Tenure -0.01 (0.11) -0.01 (0.13) -0.01* (0.04) -0.01* (0.02) Size 0.01** (0.01) 0.01** (0.01) 0.01** (0.01) 0.01** (0.01) Entrepreneur 0.26* (0.02) 0.26* (0.03) 0.26* (0.02) 0.24 (0.06) 0.25* (0.02) 0.25* (0.03) 0.24* (0.01) 0.20 (0.08) N 42 42 42 42 40 40 40 40 R2 0.14 0.04 0.18 0.19 0.35 0.35 0.41 0.43 Adjusted R2 -0.1 0.02 0.14 0.10 0.26 0.25 0.32 0.31

Note: For all OLS regressions the dependent variable is decentralization. In model (1) only regresses the skill determinant, model (2) only regresses the personal relationship determinant, model (3) only regresses the information asymmetry determinant, model (4) regresses all three determinants at once and in model (5) - (8) the control variables (i.e. GROWTH, TENURE and SIZE) are added to the regression. In parentheses are the p-values. * p<0.05, ** p<0.01, *** p<0.001.

For the skill determinant, the result of model (8) indicate a positive correlative relationship between the skill determinant and the decentralization. Unfortunately the value is insignificant on a significance level of 5%. Although, despite the p-value of 0.25 the result is inline with hypothesis 3 'More skilled workers within a firm can be positively correlated to a decentralized organisation'. The theory predicts the same positive relationship between skills and the decentralization within a firm. For instance, Caroli and van Reenen (2001) indicate that an OC reduces the demand for unskilled workers.

Looking at the control variables it is however possible to see some significant results. Lets first discuss the organisation size. In model (5) until (8) the control variables are added and it is possible to see a slightly positive relationship between the size (i.e. amount of employees) of a firm and the degree of decentralization. This slightly positive relationship is significant on a 1% significant level. This is inline with the literature, which suggests that a positive relationship as well (Colombo and Delmastro, 2004; Abernethy, et al. 2004; Graham, et al. 2015). The tenure of the entrepreneur is according to the data of Table 8 negatively related to the degree of decentralization. Meaning, when an entrepreneur is working more years at the organisation this can be related to more centralization. This is inline with the study of Graham, et al. (2015) and can be reasoned according to the information asymmetry theory as well. When the entrepreneur is working longer at the firm he or she has had more time to gather certain knowledge. Making the entrepreneur better informed resulting in less delegation (Aghion and Tirole, 1997; Bakker, Gibbons and Murphy, 1999). The control variable 'GROWTH' indicating historic and future growth on a three year scale, is not interpretable from Table 8, due to insignificant results.

Individual analysis

Apart from the main analysis with regression Formula 1 I took the dependent variable apart in three sub variables 'DECEN1', 'DECEN2' and 'DECEN3'. This type of analysis is more inline with the work of Graham, et al. (2015) where the decision with regard to specific topics are individually reviewed.

In Table 9 I did an individual analysis with regard to the information asymmetry. When 'DECEN1' is related to 'INFO1' a positive relationship should be seen in the regression. On a significance level of 5% this relationship is significant with a value of 0.49%. Meaning decisions regarding strategy and financial investments can be positively related with the

informed about the strategy this means its more likely he or she will also take the decisions regarding strategy and financial investments).

In this analysis, I can neglect the relationship between 'DECEN2', 'INFO1' and 'INFO2', this mainly due to the fact that these variables measure different topics. 'DECEN2' measures the delegation with respect to human recourses and marketing, while 'INFO1' measures the information gap with respect to strategy and 'INFO2' measures the information asymmetry with regard to the planning, user experience and architecture of the software. I can how ever relate 'DECEN3' and 'INFO2', both measuring the planning, user experience and architecture of the software. When looking at the correlative relationship between 'DECEN3' and 'INFO2' it is possible to find a positive correlation which is not significant on a significance level of 5%. Although, the value is not significant this might suggest that when the developer has more information about the architecture or user experience of the software, this is positively related to the entrepreneurs delegating these decisions. This finding is inline with the theory of Aghion and Tirole (1997), who indicated that when a agent had more specific knowledge there tend to be more delegation.

Model (1) (2) (3) (4) INFO1 _0.57* (0.05) 0.28 (0.39) INFO2 _0.24 (0.44) (0.42) 0.26 Entrepreneur _1.16* (0.02) _(0.81) -0.16 (0.30) 0.58 -0.11 (0.87) Entrepreneur* INFO1 _(0.17) -0.27 _(0.98) 0.01 Entrepreneur* INFO2 0.02 (0.13) -0.01 (0.99) Growth _-0.02 (0.88) (0.57) 0.09 Tenure _-0.02 (0.05) -0.01 (0.41) Size _0.01 (0.26) 0.01* (0.02)

N _{42 42 40 40} R2

0.30 0.08 0.38 0.22

Adjusted R2

0.24 0.01 0.26 0.08

Note: For all OLS regressions the dependent variable is decentralization on specific topics. Model (1) and (3) indicates delegation decisions regarding strategy and financial investments, model (2) and (4) indicates delegation decisions regarding planning, architecture and user experience of the software. 'INFO1' indicates who has most information with regard to strategy, 'INFO2' indicates who has most information regarding the planning, the architecture and user experience of the software. The control variables (i.e. GROWTH, TENURE and SIZE) are added to model (3) and (4). In parentheses are the p-values. * p<0.05, ** p<0.01, *** p<0.001. Table 9 Information asymmetry regressions

In Table 10 I relate the conflict of interest variable on the separate decentralization variables ('DECEN1', 'DECEN2' and 'DECEN3'). Unfortunately, these results are insignificant meaning I can not interpret these data as results. However, when looking at model (5) and (6) what can be seen is that the values become a bit more significant. Suggesting the model does become more informative. Model (5) and (6) might be interpreted as an increase in conflicting interest between the entrepreneur and the developer can be related to a decrease in the delegation.

Model (1) (2) (3) (4) (5) (6) Conflict of interest _(0.47) 0.65 _(0.31) -0.83 _(0.55) -0.53 _(0.44) 0.72 _(0.23) -0.94 _(0.27) -0.95 Growth _-0.04 (0.81) -0.09 (0.49) 0.20 (0.19) Tenure _-0.02 (0.08) -0.01 (0.64) -0.01 (0.56) Size _0.01 (0.43) 0.01 (0.17) 0.01* (0.01) Entrepreneur _1.90 (0.24) (0.42) -1.14 (0.47) -1.10 _(0.28) 1.74 _(0.32) -1.37 _(0.17) -2.10 Entrepreneur* Conflict of interest -0.36 (0.40) 0.44 (0.26) 0.28 (0.50) -0.30 (0.49) 0.49 (0.20) 0.54 (0.19) N _{42 42 42 40 40 40} R2 0.22 0.21 0.02 0.31 0.30 0.21 Adjusted R2 0.16 0.15 -0.06 0.19 0.17 0.06

Note: For all OLS regressions the dependent variable is decentralization on specific topics. Model (1) and (4) indicates delegation decisions regarding strategy and financial investments, model (2) and (5) indicates delegation decisions with regard to human resources and marketing and model (3) and (6) indicates delegation decisions regarding planning, architecture and user experience of the software. In model (4) - (6) the control variables (i.e. GROWTH, TENURE and SIZE) are added to the regression. In parentheses are the p-values. * p<0.05, ** p<0.01, *** p<0.001.

In Table 11 the determinant skill is related to the degree of decentralization on specific decisions regarding a firm. Unfortunately, these results are insignificant meaning I can not interpret these data as results. However, if for a moment the significance is neglected it is possible to see a positive relation between skills and the degree of decentralization for the models (3) until (6). Model (1) (2) (3) (4) (5) (6) Skills -0.12 (0.74) -0.02 (0.95) 0.11 (0.77) 0.07 (0.87) 0.02 (0.96) 0.21 (0.59) Growth -0.03 (0.83) (0.39) -0.12 (0.29) 0.16 Tenure -0.02 (0.08) -0.01 (0.70) -0.01 (0.54) Size 0.01 (0.36) 0.01 (0.24) 0.01* (0.02) Entrepreneur 0.56** (0.01) 0.48** (0.01) -0.06 (0.72) 0.60* (0.03) 0.43* (0.01) -0.10 (0.56) N 42 42 42 40 40 40 R2 0.21 0.18 0.01 0.30 0.26 0.17 Adjusted R2 0.17 0.14 -0.04 -0.20 0.15 0.05

Note: For all OLS regressions the dependent variable is decentralization on specific topics. Model (1) and (4) indicates delegation decisions regarding strategy and financial investments, model (2) and (5) indicates delegation decisions with regard to human resources and marketing and model (3) and (6) indicates delegation decisions regarding planning, architecture and user experience of the software. In model (4) - (6) the control variables (i.e. GROWTH, TENURE and SIZE) are added to the regression. In parentheses are the p-values. * p<0.05, ** p<0.01, *** p<0.001.

Table 11 Skill regressions

Due to the fact that both entrepreneurs as developers filled the questionnaire, in each regression model the dummy for entrepreneur is added. This gives a better fit for the data as some question within the questionnaire are targeted to the entrepreneur.

Discussion and further research

The dataset of the thesis is based on survey data. By making use of survey data, the results are not causal interpretable. Only correlations can be determined. Questionnaires can be viewed as a snapshot of one specific part in time and this is why its not possible to state that when 'A' and 'B' are measured. 'A' has an influence on 'B', but can only show a correlative relationship between 'A' and 'B'. Apart from that, the participant who files the questionnaire does this by answering according to his or her interpretation of reality. This might be different from the real truth. In this line of thinking, it might be the case that participant do not like what happened in reality or want to keep the real truth secret by manipulating the survey and on purpose file wrong answers. This makes it harder or even impossible to measure certain aspects (e.g. the extent to which the agents feels at ease around the principal or corporate culture within an organisation) and can lead an omitted variable bias.

For the survey entrepreneurs or CEOs within the ICT sector needed to participate. I choose for this sector because the company for which I worked had a dataset with ICT entrepreneurs. Although, the sample is representative for the average ICT firm in the Netherlands. This does not mean the results are interpretable for other sectors as well. Apart form the representativeness, it was harder to find other participants. Due to the fact that the participants needed to be from this particular sector. In the end this leaded to a small sample size and a smaller sample size makes it more difficult to observe significant results or might result in negative effects where positive effects are expected.

Within the dataset I acknowledged that some developers instead of entrepreneurs filled the survey, meaning that the results might be biased. If I, for example, review the questions about the degree of decentralization it is possible to see the answers 'I takes all the decisions' and 'Someone else takes all the decisions'. This said it might be the case that some developers, knowing the survey is only for entrepreneurs filled the survey as if they are the entrepreneur. This might be the case that I do not find significant differences in the answers from the entrepreneurs and the developers. I analysed this by including a dummy variable for entrepreneur. Although, some of the developers might have filed the questionnaire as if they where the entrepreneur. This can unfortunately not be captured by my dataset, but could be why the results are insignificant.

data can not be manipulated (on purpose or by coincident) by the participants as the researcher can observe and measure the variables directly. Another interesting option would be to follow different companies over a longer period of time to see their development. Measuring to what extent the company is decentralized and what determinants change if the organisation changes. As mentioned before, my dataset specifically existed out entrepreneurs who work at small and medium sized ICT enterprises and this made me wonder if the same study on big ICT enterprises would result in the same answers?

Conclusions

When looking at the decentralization within a firm the first thing I can conclude from the data is that the degree of delegation is not uniformly distributed among the firm. Decisions with respect to the strategy or investments are more taken by the entrepreneur or CEO of the firm, while decision with respect to the product (i.e. the architecture of the software and the user experience) tend to be delegated more often. Within this thesis the degree of delegation is put in perspective to multiple determinants. Namely, the information asymmetry between the entrepreneur and the developer, whether there is a conflict of interest between the two and the amount of skills within the firm.

When I look at the information asymmetry the organisational architecture theory suggests that an increased gap between the principal and the agent (i.e. in this thesis the entrepreneur and the developer) results in more delegation (Aghion and Tirole, 1997). With the use of my particular dataset I can acknowledge this theory and accept hypothesis 1 'The information asymmetry between the entrepreneur and the developer can be positively correlated with a more decentralized firm.'. Since its possible to see a significant positive relationship between the information asymmetry and the degree of delegation within the firm. For the conflict of interest determinant, I looked at multiple theories. First, the crowding-out theory suggesting that when the agent is monitored (more) this has a negative influence on its motivation (Falk and Kosfelt, 2006). Second, an increase in the reputation of the agent increases the degree of delegation (Graham, et al. 2015). Finally, when there is less alignment between the entrepreneur and the agent, then this results in a more centralized firm (Dominguez Martinez and Sloof, 2017). In my analysis I took these theories together to test whether conflicting interests between the entrepreneur and the developer decreases the degree of decentralization within the firm. According to the theory a positive relationship was expected. The results of my dataset indicate a negative relationship. However, this relationship is not significant. I can thus not accept hypothesis 2 ' When there is more conflict of interests between the entrepreneur and the developer within the firm, this is positively correlated with less decentralization.' since the relationship according to the data is insignificant.