1
Amsterdam Business School
Degree of control tightness: the influence of customer
involvement and professional service intensity
Name: Suzanne Lemaire Student number: 11425296
Thesis supervisor: H. Kloosterman, MSc Date: June 25, 2018
Word count: 13.450
MSc Accountancy & Control, specialization Accountancy Faculty of Economics and Business, University of Amsterdam
2
Statement of Originality
This document is written by student Suzanne Lemaire who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3
Abstract
This master’s thesis uses a survey method to investigate the influence of customer involvement on the design of management control systems (MCSs) in professional service firms (PSFs). This paper uses a framework which varies among the degree of professional service intensity (PSI). MCS was conceptualized in four sets of controls (result, action, personnel and cultural), distinguishing between tight and loose control. Previous studies have focused on central characteristics to explain PSFs. However, due to the explanation of PSFs, customer involvement is a critical omission. Using the framework, I try to empirically test whether customer involvement leads to a tight or loose MCS and try to find evidence how this relation is affected by PSI. Data consist of employees of several PSFs in different industries and with different educational backgrounds. Using multiple regression analyses, the results show (1) a significant positive relationship between customer involvement and implicit personnel control tightness, (2) a significant positive relation between customer involvement and explicit result control and between customer involvement and explicit action control, and (3) no significant evidence that PSI significantly affects the relation between customer involvement and explicit control tightness.
Keywords: professional service firm; management control system; customer involvement; control tightness; professional service intensity; agency theory
4 Contents 1 Introduction ... 7 2 Theory ... 9 2.1 PSI ... 9 2.1.1 Characteristics ... 10 2.2 Customer involvement ... 10 2.3 Agency Theory ... 12
2.4 Management Control System ... 12
2.5 Control tightness ... 15 3 Hypothesis development ... 16 4 Research methodology ... 19 4.1 Empirical approach ... 19 4.2 Sample ... 19 4.3 Research question ... 21 4.4 Variable measurement ... 21
4.4.1 Independent variable – Customer Involvement ... 21
4.4.2 Dependent variable – control tightness ... 22
4.4.3 Moderating variable – PSI ... 25
4.4.4 Control variables ... 28 4.5 Analysis ... 30 5 Results ... 31 5.1 Descriptive statistics ... 31 5.2 Correlation Analysis ... 31 5.3 Testing H1 ... 34 5.4 Testing H2 ... 35 5.5 Testing H3 ... 36
5
5.5.1 Normality ... 36
5.5.2 Results ... 37
5.6 Additional analysis – H3 ... 40
6 Discussion and conclusions ... 42
6.1 Discussion of the Findings ... 42
6.2 Limitations ... 43
6.3 Implications for Future Research ... 44
References ... 46
Appendix A. Survey questionnaire ... 51
Appendix B. Control tightness variables ... 63
Appendix C. Size ... 66
6
List of tables
Table 2.1 – Taxonomy PSFs……….. 9
Table 4.1 – Sample section………20
Table 4.2 – Characteristics respondents………20
Table 4.3 – Factor analysis for CI………..22
Table 4.4 – Factor analysis for CT………24
Table 4.5 – Respondents specified per occupation………26
Table 4.6 – Respondents specified per PSF type with their particular characteristics………...27
Table 4.7 – Factor analysis for PSI………28
Table 5.1 – Descriptive statistics………...31
Table 5.2 – Pearson Correlations………..33
Table 5.3 – Outcome regression analysis H1………34
Table 5.4 – Outcome regression analysis H2………36
Table 5.5 – Normality test……….36
Table 5.6 – Outcome regression analysis H3………39
Table 5.7 – Results additional test low PSI – high PSI………..41
List of figures Figure 1 – Hypothesis 1……….16
Figure 2 – Hypothesis 2……….17
Figure 3 – Hypothesis 3……….18
7
1 Introduction
In recent years Professional Service Firms (PSFs) have become prominent in economies all over the world (DeLong & Nanda, 2003, p. ix). They pose unique challenges related to coordinating, often globally, the activities of many highly autonomous professionals. Without PSFs, businesses as we know them would come to a complete standstill (Sharma, 1997). PSFs’ professionals provide expert advice to customers, who do not have the expertise to solve a particular problem. Consequently, the professional has power over the customer, because the customer is not qualified to evaluate the services needed (Mills & Margulies, 1980). To create value for customers, it is necessary to make sure that professionals are not only internally focused but also customer focused. However, to meet customer’s needs, the professional also depends on the information that the customer shares. Both, the power of the professional and the dependency of the customer on the professional make it difficult to develop a management control system (MCS) in PSFs.
The reliance on customer involvement is what makes PSFs different from other types of firms, as they need to employ a high percentage of highly educated people, and depend on the information that customers share in order to be of value to customers (Løwendahl, Revang , & Fosstenløkken, 2001). Consequently, the relationship between professionals and their customers is more important in PSFs compared to other types of firms (e.g. mass service firms).
Customer involvement is of importance for PSFs because its customers possess detailed information that is crucial to the accomplishment of certain tasks (Mills & Margulies, 1980). Without being properly informed the professional will not be able to carry out the required tasks. Furthermore, feedback from customers can be an important way for firms to monitor the professionals’ behavior, which makes customer involvement an additional control mechanism (Rucci, Kirn, & Quinn, 1998).
Reviewing the existing literature reveals a research gap regarding customer involvement as a means to determine the degree of control tightness. Previous studies have focused specifically on the issue of MCSs in PSFs by comparing PSFs with mass service firms (Auzair & Langfield-Smith, 2005). However, the degree of professional service intensity (PSI) varies across PSFs and may generate different organizational outcomes (Von Nordenflycht, 2010). Interestingly, little research has focused on the importance and impact of customer involvement in PSFs.
For example, Sharma (1997) concluded that professionals have power over the customers, because the customers limitedly understand the specialized services and execution
8
of associated tasks. In other words, knowledge asymmetry arises between professionals and customers. In the presence of knowledge asymmetry, the customer must be able to rely on the findings of the professional. To control and incentivize the professional, firms make use of management controls. Management controls are tools to ensure that a firm’s strategy, goals and objectives will be achieved (Ouchi, 1979), which leads to a PSF’s ability to create value for customers. The degree to which controls are integrated into PSFs can be explicit or implicit. Implicit control tightness is achieved when the level of tolerance, i.e. how flexible the rules are used, is minimized, while, explicit control tightness is achieved by creating more controls, i.e. more rules and procedures.
This paper investigates to what extent customer involvement influences the degree of control tightness and whether PSI plays a role in the development of MCS in PSFs. More precisely, my research question is:
To what extent is the degree of control tightness within PSFs determined by customer involvement and how is this affected by PSI?
This study contains three contributions to the existing literature. First, it provides novel insights into the effect of customer involvement into MCS, thereby addressing the lack of attention towards customer involvement and the design of a MCS. Second, this study contributes to the literature on control tightness, by distinguishing explicit and implicit control tightness. Finally, while most studies focus on a central characteristic and apply the results to the entire sector of PSFs (Von Nordenflycht, 2010), this study uses data from PSFs with different characteristics, by using the taxonomy developed by Von Nordenflycht (2010). This framework helps understanding the implications of knowledge intensity, capital intensity and professionalization, which are the three key characteristics of PSFs.
The paper is structured in the following manner: section two provides the theoretical framework to study PSI, customer involvement, MCS and control tightness, which is needed to examine the developed hypotheses in section three. Section four describes the research method and explains the variables and models used for this study. Thereafter, the results of the regression analysis are provided in section four. Finally, section five contains a comprehensive discussion of the results, the study’s limitations, and implications for further research.
9
2 Theory
In this section, I review the existing literature on PSF, agency theory, MCS and CT. The first part of this section summarizes research to date on PSFs, customer involvement in PSFs, and on the relationship between professionals and customers. Thereafter, the relevant literature on management control systems is summarized and the concept of control tightness is defined.
2.1 PSI
In previous studies, the term PSF is often undefined or defined indirectly, which has led to an inconsistent definition. In this thesis, I will use the taxonomy developed by Von Nordenflycht (2010) to identify a PSF. He identifies three important characteristics of PSFs: (1) knowledge intensity, (2) low capital intensity and (3) professionalized workforce, which are the most comprehensive characteristics in the degree of PSI. Knowledge intensity indicates the degree of organizational output which relies on the complex knowledge of a professional (Starbuck, 1992). Low capital intensity indicates that the firm is more labor intensive and less capital incentivized. Thus, the organizational output relies on professionals’ knowledge and not on the output obtained through machines. Professionalized workforce indicates how the specific knowledge of a professional and its application is regulated and controlled. Based on these three characteristics, Von Nordenflycht (2010) proposes a taxonomy for four types of PSFs that can be found in table 2.1.
Table 2.1: Taxonomy PSFs Knowledge intensity Capital intensity Professionalized workforce
Category (with examples)
Technology Developers High High Low
Biotech R&D labs
Neo-PSFs High Low Low
Consulting Advertising
Professional Campuses High High High
Hospitals
Classic PSFs High Low High
Law Accounting
10
2.1.1 Characteristics
The way PSFs are designed and operate will result in some specific challenges and opportunities. Examples of these are the preference of the professionals to operate autonomously, the challenge for customers to measure the impact of the professional and the opportunity of no investor protection.
Regarding the first, professionals prefer to operate autonomously (Raelin, 1989), since they are shaped by their discipline and culture in carrying out their technical responsibilities as member of a PSF. Consequently, it is difficult to steer professionals to do things they do not want to do (Raelin, 1989). This autonomy becomes problematic when professionals demand, and often achieve, control not only over their professional activities, but also over the purpose thereof (Barley & Tolbert, 1991). To better satisfy the professionals’ preference for autonomy, PSFs may display more autonomy for professionals (Von Nordenflycht, 2010).
Løwendahl (2001) looks at the challenge of opaque quality. Opaque quality arises when it is difficult for customers to measure the quality of the professionals’ output. This challenge gives rise to the need for controls in order to signal quality. Von Nordenflycht (2010) distinguishes such controls in four types: (1) bonding, (2) reputation, (3) appearance and (4) ethical codes. These control mechanisms help to signal quality output where quality is opaque. For example, ethical codes can help to protect customers’ interest. Since ethical codes are one of the core features of most professions, the professionalization of a profession may be a response to opaque quality (Von Nordenflycht, 2010). Furthermore, developing and maintaining a reputation is very important. Building a good reputation among customers is likely to guarantee quality. Satisfied customers tend to return for future business and at times assist in recruiting new customers by word-of-mouth (Rucci, Kirn, & Quinn, 1998).
Low capital intensity leads to no investor protections, which generates an opportunity (Von Nordenflycht, 2010). When a firm is less capital incentivized, it reduces the need for raising investment funds and thereby reduces the need to act in a way that protects outside investors (Von Nordenflycht, 2010). This allows firms to adopt more autonomy, which may lead to professionals who are satisfied.
2.2 Customer involvement
Von Nordenflycht (2010) does not mention customer involvement in his taxonomy. However, PSFs apply their complex knowledge to the customer’s specific situation, rather than producing
11
standardized and repetitive tasks for the mass-consumer sectors (Maister, 1993). Furthermore, face-to-face interaction with customers is an important part of most PSFs (Maister, 1993). Thus, both taking into account the customers’ specific situation and face-to-face interaction with customers, create value for PSFs. Therefore, customer involvement is a critical gap in Von Nordenflycht’s (2010) framework for defining PSFs
In the literature, customer involvement is referred to as customer contact, customer presence, customer participation, customer interaction, customer influence, and customer co-production. Auh et al. (2007) define co-production as constructive customer participation in the service creation and delivery process and clarify that it requires meaningful, cooperative contribution to the service process. Dabholkar (2015, p. 484) uses the term customer participation as the degree to which the customer is involved in producing and delivering the service. In general, all of these terms refer to the degree to which the customer is involved in the service provision.
Tasks with a high degree of customer involvement are relatively unpredictable. In order to coordinate service provision, there is a need for interaction and close cooperation between the service supplier and the customer (Løwendahl B. , 2005). Furthermore, customer involvement creates a certain level of uncertainty in terms of what the delivered service is going to be in detail. Heterogeneity in services makes it difficult to generalize and standardize outputs for multiple customers (Løwendahl B. , 2005).
By contrast, customer involvement may also result in certain benefits. It helps PSFs to gain access to critical knowledge-based resources (Lengnick-Hall, 1996) and is therefore the fundamental source of competitive advantages that add value to firms (Lengnick-Hall, 1996). Because customers are the ones using a firm’s services, their experience can inform firms on how to improve their offerings and processes (Fornell & Wernerfelt, 1987). In addition, customer involvement is seen as an important success factor for innovation. The degree of customer involvement declares the differences between successes and failures in service innovations (Anning-Dorson, 2016).
To analyze customer involvement, I look at customer involvement from a reliance point of view; the degree to which the help of a customer is required in order for the PSF to reach a good result. Active customer involvement involves complex interdependencies and is difficult to coordinate (Mills & Morris, 1986), which may impact the design of the MCS. To deal with the increased uncertainty, additional controls can be implemented in the MCS. However, to deal with the customer’s needs, professionals are allowed to use creativity and to work outside the scope of their job. This may require more flexibility.
12 2.3 Agency Theory
Agency theory investigates relationships between the principles and agents with different goals and division of labor. There are two major types of agency theory: (1) the traditional form and (2) the customer-professional agency theory. The traditional theory is used for years in management accounting. The assumption of this theory is that the interests of the principal and the agent are different. Within this context, the principal wishes to maximize the profit, while the agent will choose actions to maximize his/her earnings, which might be at the expense of the principal (Zhang & Zenios, 2008). Additionally, the agent is more knowledgeable about his/her profession, and therefore possesses more information than the principal. The greater the difference in information between the principal and the agent, the higher the level of information asymmetry (Abernethy, Bouwens, & van Lent, 2004). To align the interests of the principal and the agent, and to reduce the information asymmetry, controls are of great necessity. Controls can influence the agent’s behavior, since they will be rewarded and penalized based on their decisions (Abernethy, Bouwens, & van Lent, 2010).
Sharma (1997) examines the customer-professional agency theory. He focuses on the relationship between professionals of service firms and their customers. Customers hire external professional services to address the issues they do not hold the specific expertise on. These situations can be viewed as principal-agent-exchange (Eisenhardt, 1989). Professionals have power over customers by means of their expertise, functional indispensability and intrinsic ambiguity associated with the services they provide (Sharma A. , 1997, p. 768). Hence, knowledge asymmetry increases because customers do not know how professionals do their job. Knowing that customers only limitedly understand the specialized services and execution of associated tasks, the knowledge asymmetry will be higher in PSFs than in non PSFs.
In this thesis, I specifically focus on customer involvement, and therefore I use the customer-professional agency theory.
2.4 Management Control System
Merchant and Van der Stede (2007, p. 5) define management control as all devices and systems that managers use to ensure that the behavior and decisions of their employees are consistent with the respective firm’s objectives and strategies. The devices and systems are distinguished among four types of control: action controls, result controls, cultural controls and personnel controls.
13
Action controls and result controls refer to hard controls. Hard controls are more explicit, formal and visible. These controls can help to guide the professional’s behavior through defined policies and procedures. Action controls are implemented to ensure that employees act in the firm’s best interest. Therefore, they are the most direct controls (Merchant & Van der Stede, 2012). It influences employees’ behavior in a direct manner by prescribing the actions they should take. Action controls can only be effectively applied if the actions of the employees can be monitored (Merchant & Van der Stede, 2012). Result controls control the behavior of professionals. This type of control can influence the actions of or decisions made by professionals (Merchant & Van der Stede, 2012). Result controls are only effective when the professionals can influence the results that they are held accountable for and when the firm is able to effectively measure the results (Merchant & Van der Stede, 2012).
Cultural and personnel controls refer to soft controls. Soft controls are founded in the culture of a firm, which consists of mechanisms that influence the motivation, loyalty, integrity, inspiration and personal values of the professionals. Cultural controls are designed to encourage mutual monitoring. It is built on shared traditions, norms, beliefs, values, ideologies, attitudes and ways of behaving. The norms are embodied in written and unwritten rules that govern professionals’ behavior (Merchant & Van der Stede, 2012). The controls become tighter when a strong culture is present (Merchant & Van der Stede, 2012). Personnel controls build on the professionals’ natural tendencies to control or motivate themselves, and involve self-monitoring. Personnel controls are effectively implemented when the right people are selected and employed (Merchant & Van der Stede, 2012).
Several factors and distinctive features can influence the control process, such as complexity, degree of formal responsibility and style of control. Therefore, different steps need to be taken to design and implement a concrete MCS (Amigoni, 1978). An important feature of PSF relevant to the design of MCS is customer involvement. A high degree of customer involvement offers greater flexibility, which indicates that MCS has to allow more variability and less programmable controls (Auzair & Langfield-Smith, 2005). Furthermore, when there is a high degree of flexibility, tasks are not easy to define, which can result in uncertainty. This notion is supported by Abernethy et al. (1995) who believe that the unpredictability of complex work requires less obtrusive forms of management control in PSF.
However, contradictory findings were suggested by Whitley (1999) who reported that professionals, who cope with the unpredictability of complex work, are likely to be tightly
14
monitored and controlled. In general, trust in their competence and commitment is unlikely to be high because customers do not know how professionals do their job.
To control the professional’s behavior, Sharma (1997) makes the following proposition:
• The customer should have a high degree of involvement in the coproduction of service products;
• Community reputation and controls by peers should be used to audit the agent.
• The professional is likely to restrain their misaligned behavior by using behavior based controls, because customers can easily share information about their experiences with particular professionals;
• Professional firms should hire professional superordinate supervisors to audit the customer;
• Engage in long-term relationship for the same or linked products or services.
A higher degree of customer involvement leads to less uncertainty. It creates a situation in which it is less likely that the customer holds unique information of which the professional is uninformed, or a situation in which the customer does not know what the professional is doing. Hence, in this paper customer involvement is referred to as an extra control mechanism. Seeking feedback from customers may be an important way for firms to monitor professionals’ behavior. Customers who are satisfied tend to return for future business and sometimes assist in recruiting new customers by word-of-mouth (Rucci, Kirn, & Quinn, 1998). Therefore, customer involvement replaces behavior controls.
As already stated, Sharma’s proposition also includes two types of soft controls, namely personnel controls and cultural controls (Merchant & Van der Stede, 2012). Personnel controls are used to find the right people to do a specific job. Cultural controls are based on the relationship with the firm and the professionals. Both play a role in the involvement of customers because soft controls allow professionals to determine how to apply procedures. This notion is supported by Alvesson (2004) who believes that PSFs use more flexibility because of the high degree of customer involvement.
Prior studies suggest that service quality measurements in PSFs are unstructured and focused on personnel and cultural controls. Fitzgerald et al. (1991) found that PSFs have a high degree of flexibility and task uncertainty. This indicates that an MCS that allows quick response is more critical in PSFs. This notion is supported by Ouchi (1979) who found that informal
15
controls are highly relevant in less controllable and unsecure circumstances. Focusing on only informal controls can be a limitation to the research of MCS in PSFs. However, none of these studies focus on the relationship of all four types of control and how they influence each other. This study focuses on a range of PSFs and contributes to the lack of knowledge that exists on the interaction between the four types of control and on the influence of customer involvement on the design of MCS in PSFs.
2.5 Control tightness
In order to fully understand the design of the MCS in PSFs, it is important to focus on not only the four types of control but also on the tightness of these controls. All types of control can be applied to a tight or loose MCS, whereby not all alternatives are equally effective (Van Der Stede, 2001).
Control tightness is not systematically defined in literature (Van Der Stede, 2001). The existence of many definitions of control make it difficult to define a control system (Fisher, 1998). Therefore, few authors choose to define tight and loose control as a whole.
The degree to which controls are integrated into the firm can be explicit or implicit. In this study, implicit control tightness is based on the level of tolerance, how flexible the rules are used. This degree of control tightness is achieved when the tolerance for deviation from the controls is minimized. On the contrary, explicit control tightness represents the extent or scope of the MCS. In this case, tightness is achieved by creating more controls, more rules and procedures that represent explicit control tightness.
Based on previous literature, the form of control that a firm has to choose varies with the nature of the task (Burns & Stalker, 1961). When tasks are highly uncertain, which often occurs in PSFs, tightening the controls would work counterproductive. Tightening the controls may prohibit creativeness and could result in dysfunctional behavior (Auzair & Langfield-Smith, 2005). Thus, to cope with this uncertainty some controls are required to be loose.
16
3 Hypothesis development
Based on the different theories and concepts reviewed in the previous chapter, this chapter presents my hypotheses.
Professionals prefer to operate autonomously (Raelin, 1989). This makes authority more problematic in PSFs. In contrast to usual business relationships, the relation between professionals and customers is often loyal and very strong, resulting in customers following the professionals if they leave the firm (Greenwood, Li, Prakash, & Deephouse, 2005). This relationship requires time and effort from professionals to achieve the demands of customers, which increases their workload (Hsieh, Yen, & Chin, 2004). If professionals are not content with this increase, they can easily change firms since their skills are transferable. Therefore, professionals are in a very strong bargaining position.
In the presence of autonomy and having a strong bargaining position, I assume that firms subject professionals to less tight controls and give them more discretion. This leads to my first hypothesis (figure 1):
H1: Customer involvement is positively associated with greater implicit control tightness
The quality of a professional is complex to evaluate. Problems within PSFs arise due to the knowledge asymmetry between the professional and the customer. The customers do not know how the professionals perform their tasks, which makes it difficult to evaluate the quality of the service (Greenwood, Li, Prakash, & Deephouse, 2005). Additionally, PSFs are forced to attract and retain professionals because of their complex knowledge. This complex knowledge must be integrated and managed to create value (Hitt, Biermant, Shimizu, & Kochhar, 2001). Furthermore, a high degree of customer involvement is characterized by complexity, uncertainty and interdependency. This is caused by the requirements of interaction between the professional and the customer (Løwendahl B. , 2005). Consequently, this can impact the individual performance of the professional because it makes it unpredictable and difficult to
Customer Involvement
Figure 1: Hypothesis 1
17
coordinate. Hence, there is a need for more frequent, detailed and timely monitoring allowing less flexibility, which leads to my second hypothesis (figure 2):
H2: Customer involvement will increase the number of controls in the MCS
The relation mentioned here (hypothesis 2) implicates that due to the high degree of customer involvement, the PSFs will monitor more frequently, more detailed and more timely. Nevertheless, firms with a high degree of PSI will trust professionals in their knowledge and capabilities to perform tasks without being regularly supervised (Von Nordenflycht, 2010). They are often highly educated people who are expected to contribute to the firm’s reputation. This leads to a competitive advantage because of the knowledge asymmetry experienced by customers (Hitt, Biermant, Shimizu, & Kochhar, 2001). Therefore, reputation serves as a social signal to customers (Greenwood, Li, Prakash, & Deephouse, 2005). Furthermore, one of the key characteristics of PSFs is low capital (Von Nordenflycht, 2010). Low capital intensity leads to no investor protections, which generates an opportunity. When a firm is less capital incentivized, it reduces the need for raising investments funds and thereby reduces the need to act in a way that protects outside investors (Von Nordenflycht, 2010). This allows firms to adopt more autonomy, i.e. less controls, which may lead to satisfied professionals. Therefore, my third hypothesis is as follows (figure 3):
Customer Involvement
Figure 2: Hypothesis 2
18
H3: PSI will negatively affect the relationship between customer involvement and the number of controls in MCS.
Customer Involvement
Figure 3: Hypothesis 3
Explicit control tightness
19
4 Research methodology
4.1 Empirical approach
This study is motivated by a desire to increase our understanding of the degree of control tightness in PSF, in the presence of customer involvement in different occupations. Since there is no public data available to test the hypothesis, a survey has been used in order to gather the specific data.
4.2 Sample
For this study, the hypotheses were tested with a survey among mid-level employees of medium to large professional service firms. The employees met the following criteria: (1) the employee has worked in the field for more than three years (but probably less than ten years); (2) the employee is not an owner/partner or board member of the organization. In other words, the employee had to be subject to the management account and control system rather than having designed it. (3) The employee works in a medium/large size organization (>50 employees). To gather this sample, I joined a project of Helena Kloosterman. Together with other students I had the opportunity to use her survey on the condition that we would gather at least ten respondents. The final database consists of 730 employees, which was collected by other students and myself.
From the final database, a total of 447 (61,2%) responses were usable. Of the 730 responses, 92 employees did not finish the survey. In addition, the first question contained a control question, asking if the respondent has read the above definition and remarks, which 21 employees did not. Their answers were untrustworthy since the possibility existed that employees, who did not read the definition and remarks, did not fully understand the survey. Consequently, this could have resulted in bias and therefore their answers were removed. As mentioned before, one of the criteria was that the employee has worked in the field for more than three years. A total of 170 employees had less than 3 years of work experience or did not meet this question and were thus deleted from the sample.
20 Table 4.1: Sample selection
Criteria #Frequency % Percentage
Completed PSF-surveys 2015 to 2018 730 100,0%
Unfinished surveys -92 12,6%
638
Have you read the above statements and remarks? -21 2,9% 617
Less than 3 years' work experience or did not meet this
question -170 23,3%
Total respondents 447 61,2%
According to Field (2013), a good sample size has a minimum of 300 samples and a poor sample size contains less than 100 samples. My sample size contained 447 samples and exceeded the above minimum of 300 samples. Therefore, it can be regarded adequate. Table 4.2 provides an overview of the respondents’ characteristics.
Table 4.2: Characteristics respondents
% Percentage Mean Std. Deviation # Number
Gender
Male 67,8%
Female 32,2%
Age (in years) 37 8,99
Youngest 24 Oldest 66 Education 4 years 15,9% 5 years 13,2% 6 years 10,9% 7 years 7,7% 8 years 7,3% 9 years 3,8% 10 years or more 41,2%
21 4.3 Research question
The following research question will be answered in this thesis
To what extent is the degree of control tightness within PSFs determined by customer involvement and how is this affected by PSI?
This study contributes to the existing literature since it can provide new insights in the relation of customer involvement between different occupations and the degree of control tightness. Customer involvement creates a dependency and uncertainty within firms which can be the reason why PSFs attach more or less importance to the degree of control tightness. I expect that when customer involvement and PSI are taken together, it weakens the explicit use of controls in MCS.
4.4 Variable measurement
4.4.1 Independent variable – Customer Involvement
The independent variable in this research is customer involvement (CI). The survey contained one topic with several sub-questions that measured how employees work with clients. In addition, it measured how often an employee needs to coordinate the activities with the customer, if they have to work in close collaboration with their customer and if they are depending on the customer to provide them with information during their work. Together these sub-questions provided insight in how the employee works with customers. The employee had to answer six questions on a 5-point Likert scale. The construct measuring CI were adapted from Homburg and Stebel (2009).
To verify the scale constructions, factor analysis (principal component analysis) was performed. A factor analysis is a correlation matrix that looks for variables that are highly correlated. Items that load on an underlying factor will correlate very high with each other (Field, 2013). In this setting, five items were loaded on one factor. To improve reliability one item had to be deleted (0.559). Deleting this question let to an increased Cronbach’s Alpha from 0.874 to 0.895. Nevertheless, this increase was negligible and both values reflect a good degree of reliability (Field, 2009). Therefore, this question was not deleted. Table 4.3 shows the result of the factor analysis. The Kaiser-Meyer-Okin measure was 0,791, which is seen as middling, and indicates that the factor analysis has an acceptable correlation (Field, 2013). Additionally, the Bartlett test of Sphericity was significant (χ2 = 7897,417 and p = 0.000). The total variance explained by the factor CI is 13,76%.
22 Table 4.3: Factor analysis for CI
Items CI
During my work, I depend a lot on client to provide required
data, information, materials, etc. 0,855
I often need to coordinate my activities with the client during
the performance of my main tasks 0,808
In our work, we also able to perform our tasks successfully
without the cooperation of our clients (or their employees) ** 0,559 In my organization, we must work in close collaboration with
our client in order to ensure a successful service outcome. 0,787 In order to do my work (properly), I depend a lot on the client
to provide me with data, information and materials. 0,865 I often have to wait for client input before I can move on to the
next step of my work. 0,795
Variance explained 13,76%
Cronbach's alpha 0,874
Cronbach's alpha* 0,895
* Cronbach's Alpha after deleting items ** Deleting this item improves reliability
The results in table 4.3 show that the items are adequate for factor analysis. I centered the variables for further analysis.
4.4.2 Dependent variable – control tightness
The dependent variable in this research is control tightness (CT). A part of the survey is divided into four management control variables. The variables are action control, result control, personnel control and cultural control. CT can be defined as the degree of flexibility in the control system. Based on two indicators in the control system, namely implicit CT (IMP. CT) and explicit CT (EXP. CT), I designed CT as a reflective construct that consist of four different measures: result control, action control, personnel control and cultural control. Most of the constructs used assertions adapted from previous literature, while others were new. Appendix B provides an overview of the different constructs specified according to the types and literature.
To determine the reflective constructs, a factor analysis was performed. Based on pervious literature, I expected to extract eight factors. Hence, I extracted a fixed number of eight factors. In addition, I selected Varimax rotation suppressed small coefficients below 0.4.
23
The reliability test, showed that a total of 59 respondents (12,3%) had some missing values. This is above the threshold for missing values, which is 10% (Field, 2013). Therefore, I selected “exclude cases pairwise”.
The results of the factor analysis are shown in table 4.4. The Kaiser-Meyer-Okin measure was 0.796, which is seen as middling. According to Field (2013), a measure of 0.796 indicates that the factor analysis has an acceptable correlation. Additionally, the Bartlett test of Sphericity was significant (χ2 = 3918,904 and p-value = 0.000). This means that the items that load on one underlying factor are correlating with each other (Field, 2013). The lowest correlated value in the factor analysis was 0.423. By deleting this item, the Cronbach’s alpha did not improve, therefore I did not delete it from my factor analysis. The total variance explained by the factors was 58,25% and consisted of EXP. CULT. CT (15,8%), EXP. RES. CT (10,85%), EXP. ACT. CT (7,10%), IMP. ACT. CT (5,94%), EXP. PERS. CT (5,48%), IMP PERS. CT (5,08%), IMP. CULT. CT (4,31%) and IMP. RES. CT (3,69%).
The Cronbach’s alpha for IMP. RES. CT was 0.385, which is far below the acceptable level of reliability. Hence, IMP. RES. CT was removed from the analysis.
24 Table 4.4: Factor analysis for CT
Items EXP. CULT. CT EXP. RES. CT EXP. ACT. CT IMP. ACT. CT EXP. PERS. CT IMP. PERS. CT IMP. CULT. CT IMP. RES. CT Q5_1 0,710 Q5_4 0,717 Q5_6 0,755 Q5_7 0,740 Q5_9rev 0,631 Q5_10rev 0,554 Q5_11rev 0,614 Q5_12 0,666 Q4_1 0,768 Q4_2 0,745 Q4_4 0,681 Q4_8** 0,423 0,523 Q4_10rev 0,769 Q4_12rev 0,826 Q4_13rev 0,784 Q3_1 0,813 Q3_2 0,808 Q3_3** 0,690 Q3_5 0,457 Q3_6 0,787 Q3_7rev 0,517 Q3_8 0,743 Q3_11rev 0,594 Q10_2 0,815 Q10_3 0,547 Q10_5rev 0,819 Q10_6 0,603 Q10_7 0,702 Q10_4 0,763 Q10_8 0,748 Q10_9 0,659 Variance explained 15,80% 10,85% 7,10% 5,94% 5,48% 5,08% 4,31% 3,69% Cronbach's alpha 0,797 0,807 0,746 0,768 0,742 0,624 0,650 0,385 Cronbach's alpha * 0,806 0,747 0,770
* Cronbach's Alpha after deleting items ** Deleting this item improves reliability
25 4.4.3 Moderating variable – PSI
The moderating variable is PSI. PSI was measured using different questions based on a 5-point Likert-scale. These sub-questions provided information about knowledge intensity, capital intensity and professionalized workforce. The definition and the construct measuring capital intensity are both adapted from Subramaniam and Youndt (2005). The other two (knowledge intensity and professionalized workforce) were based on new items, developed for the purpose of the project.
According to Von Nordenflycht (2010), PSI is based on the presence of characteristics. The distinctive characteristics are the following: knowledge intensity, low capital intensity and a professionalized workforce. However, in 2015 he also identified CI as a shared distinctive characteristic of PSFs. He argued that PSFs apply their knowledge base to the specific circumstance of a given client (von Nordenflycht, et al., 2015). Von Nordenflycht’s (2010) distinguishes four types of PSFs: Technology Developers, Professional Campuses, Classic PSFs and Neo PSFs. Each type representing a particular combination of the characteristics. The respondents specified per occupation are depicted in table 4.5.
26 Table 4.5: Respondents specified per occupation
Occupation # Frequency % Percent # PSF type
Accounting 66 14,8% 1 Advertising 8 1,8% 3 Architecture 2 0,4% 1 Biotechnology 14 3,1% 4 Consulting Engineering 1 0,2% 1 Consulting IT 32 7,2% 3 Consulting HR 15 3,4% 3 Consulting Management/Strategic 32 7,2% 3 Consulting Technology 3 0,7% 3 Engineering 21 4,7% 1 Financial advising 5 1,1% 3 Graphic design 2 0,4% 1 Insurance brokerage 1 0,2% 1 Investment banking 8 1,8% 1 Investment management 8 1,8% 1 Law 24 5,4% 1 Marketing/public relations 10 2,2% 3
Media production (film, TV, music) 10 2,2% 3
Medicine/Physician practices 42 9,4% 2 Pharmaceutical 10 2,2% 4 Project management 26 5,8% 3 Real estate 7 1,6% 3 Recruiting executive 5 1,1% 3 Research/R&D 27 6,0% 4
Risk management services 12 2,7% 3
Software development 8 1,8% 3
Talent management/agency 3 0,7% 3
Other 45 10,01 5
Total 447 100,0%
1 = Classic PSFs or Regulated PSFs 3 = Neo-PSFs 5 = Other 2 = Professional Campuses 4 = Technology Developers
Under PSF types the occupations are distributed amongst the four categories of Von Nordenflycht. By interpreting Von Nordenflycht’s (2010) framework, I included occupations that he does not mention in his taxonomy. The occupation that could not be allocated to the category were labelled as “Other”. In table 4.6, the respondents are specified per PSF type with their particular characteristics. As mentioned before, “Other” cannot be allocated to one of the four categories and were therefore removed from this table.
27 Table 4.6: Respondents specified per PSF type with their particular characteristics
I modeled PSI as a formative construct that is defined by three characteristics, namely knowledge intensity, capital intensity and professionalized workforce. For example, the second row meets all three characteristics, such as hospitals. These firms are more capital intensive, which often stems from a specialized physical infrastructure (Von Nordenflycht, 2010). Moving downward, the firms in the third row differ from the firms in the second row by being less capital intensive and by having non-professionalized workforces. Thus, for each category, the predicted intensity is indicated.
Examination of knowledge intensity (5 items), capital intensity (3 items), and professionalized workforce (4 items) on reliability shows that they were adequate for factor analysis. The Cronbach’s alpha in table 4.7 presents that the all items score above the threshold reliability of 0.700.
To verify the reflective scale constructions, factor analysis was performed. The Kaiser-Meyer-Okin measure was 0.751, which is seen as meritorious, and indicates that the factor analysis has an acceptable correlation (Field, 2013). Additionally, the Bartlett test of Sphericity was significant (χ2 = 1648.180 and p = 0.000). The total variance explained is 37,34% and consists of knowledge intensity (22,77%), professionalized workforce (20,65%) and capital intensity (12,59%).
28 Table 4.7: Factor analysis for PSI
Items Knowledge Intensity Professionalized workforce Capital Intensity Q13_1 0,698 Q13_3 0,790 Q13_4 0,656 Q13_5 0,804 Q12_5rev Q12_8 0,840 Q12_9rev 0,822 Q12_11rev 0,772 Q11_1 0,768 Q11_2 0,752 Q11_3 0,810 Q11_4 0,758 Q11_5 0,738 Q32_1 Q32_2 Q32_3rev** Q32_4 Q32_5 Q32_6 Variance Explained 22,77% 20,65% 12,59% Cronbach's Alpha 0,824 0,736 0,751 Cronbach's alpha *
* Cronbach's Alpha after deleting items ** Deleting this item improves reliability
4.4.4 Control variables
Control variables were included because they may give rise to alternative explanations for the degree of CT. As control variables I took into account the factors that could influence the degree of CT by CI. Since all three hypotheses measure the relationship between CI and the numbers of controls in the MCS, the control variables for the three hypotheses were the same. The first control variable was organizational size (SIZE). Khandwalla (1972) found that large organizations made greater use of sophisticated controls due to an increase in information. An increase in information is important to control the behavior of employees. Therefore, these
29
organizations adopt a MCS that tend to use more formal, bureaucratic controls. This notion is supported by Sharma (2002) who find that an increase in size changes the MCS design. In the survey, respondents were asked how many people are employed by their entire company. They could give their answer based on four options (see Appendix C. Size for an overview).
The second control variable was environmental uncertainty (ENV). I controlled for ENV since it plays a role in the development of a MCS. Gordon et al. (1984) argue that ENV is a very important contingency variable because the higher the perceived ENV, the greater the need for external and nonfinancial information. To test for ENV, respondents were asked how intense price compensation, competition for manpower and bidding for new contracts/clients is in their industry. Furthermore, one question was asked about the number of products and/or services within the industry in the last five years. The last questions addressed the predictability of clients’ tastes and competitors’ activities within their industry. The questions were adapted from Gordon and Narayanan (1984).
The third control variable was reputation (REP). Advantages of economy of scale are that the firm’s reputation becomes known to a bigger audience, consequently marketing will become less crucial because the reputation itself attracts customers (Greenwood, Li, Prakash, & Deephouse, 2005). By adding REP as control variable, I excluded the possibility that firms with a good REP and a strong brand name affected the design of the MCS (Greenwood, Li, Prakash, & Deephouse, 2005). Professionals prefer to act on their expertise without feeling the burden of being controlled. REP allows PSFs to adopt more autonomy for professionals, therefore it influences the design of MCS. The survey asked respondents to indicate how their organization is viewed by the broader public. There are four questions, adapted from Combs and Ketchen (1999), on a five-point Likert scale about the perceived strength of the brand name, respect and REP with regards to the organization.
SIZE has a reliability score of 0.713, ENV has a score of 0.638 and REP has a score of 0.811. Both items, SIZE and REP score over 0.700, which is considered as a scale of high reliability (Field, 2013). Although ENV does not meet the criterion of 0.700, it is still over 0.500 and therefore accepted as indicating a moderately reliable scale (Hinton, McMurray, & Brownlow, 2004).
30 4.5 Analysis
This study tested the theoretically assumed relationship between the independent variable CI, the dependent variable CT and the moderator PSI, which may affect the relation between CI and EXP. CT
First, I conducted a linear regression to test hypothesis 1 and hypothesis 2. Hypothesis 1 assumed that IMP. CT is higher by more CI. Hypothesis 2 assumed that EXP. CT is higher by more CI. The estimated regression models for these hypotheses are:
H1 – Model
𝐼𝑀𝑃. 𝐶𝑇 = 𝛽0+ 𝛽1𝐶𝐼 + 𝛽2𝑆𝑖𝑧𝑒𝑖+ 𝛽3𝐸𝑁𝑉𝑖+ 𝛽4𝑅𝐸𝑃𝑖+ 𝜀𝑖 H2 – Model
𝐸𝑋𝑃. 𝐶𝑇 = 𝛽0+ 𝛽1𝐶𝐼 + 𝛽2𝑆𝑖𝑧𝑒𝑖+ 𝛽3𝐸𝑁𝑉𝑖+ 𝛽4𝑅𝐸𝑃𝑖+ 𝜀𝑖
Finally, I tested hypothesis 3. Hypothesis three assumed that PSI moderates the relation between CI and EXP. CT. In order to test hypothesis 3, I performed a linear regression. This third linear regression was the main regression. The outcomes of this regression provided an answer to the research question. The main hypothesis examined whether the moderator weak the original relation. This entailed that I assumed that PSI negatively affects the relationship between CI and EXP. CT. The moderating effect was tested with a newly created variable (PSI*CI). The estimate regression models for this hypothesis were:
H3 – Model 1 𝐸𝑋𝑃. 𝐶𝑇 = 𝛽0+ 𝛽1𝑆𝑖𝑧𝑒𝑖+ 𝛽2𝐸𝑁𝑉𝑖+ 𝛽3𝑅𝐸𝑃𝑖+ 𝜀𝑖 H3 – Model 2 𝐸𝑋𝑃. 𝐶𝑇 = 𝛽0+ 𝛽1𝑃𝑆𝐼 + 𝛽2𝐶𝐼 + 𝛽3𝑆𝑖𝑧𝑒𝑖+ 𝛽4𝐸𝑁𝑉𝑖+ 𝛽5𝑅𝐸𝑃𝑖+ 𝜀𝑖 H3 – Model 3 𝐸𝑋𝑃. 𝐶𝑇 = 𝛽0+ 𝛽1𝑃𝑆𝐼 + 𝛽2𝐶𝐼 + 𝛽3(𝑃𝑆𝐼 ∗ 𝐶𝐼) + 𝛽4𝑆𝑖𝑧𝑒𝑖+ 𝛽4𝐸𝑁𝑉𝑖+ 𝛽5𝑅𝐸𝑃𝑖+ 𝜀𝑖
The residual (ε) is the difference between the predicted and the observed value which cannot be explained by the model. Both the dependent variable and the independent variable were measured on an ordinal scale (the five-point Likert scale).
To create an interaction term in model 3, PSI was multiplied by CI, which could lead to multicollinearity. To avoid this the interaction effect was centered.
31
5 Results
5.1 Descriptive statistics
In this section the descriptive statistics of my variables are included in table 5.1. For all variables, the standard deviation is low. The data points are close to the mean indicating the consistent behavior of my dataset. Thus, the dataset is tight or little dispersed for all variables. Additionally, the means for all the variables are slightly above the average of the range. Table 5.1: Descriptive statistics
Variables Range Mean Minimum Maximum Std. Dev
Dependent variables IMP. ACT. CT 1-5 2.90 1.00 5.00 0.90 IMP. PERS. CT 1-5 3.12 1.00 5.00 0.76 IMP. CULT. CT 1-5 3.45 1.25 5.00 0.74 EXP. RES. CT 1-5 2.77 1.00 5.00 0.87 EXP. ACT. CT 1-5 3.09 1.00 5.00 0.85 EXP. PERS. CT 1-5 3.23 1.00 5.00 0.78 EXP. CULT. CT 1-5 3.52 1.00 5.00 0.91 Independent variables CI 1-5 3,72 1.00 5.00 0,89 PSI 1-5 3,53 2.21 5.00 0,46 Control variables SIZE 1-5 2.74 1.00 4.00 0.94 ENV 1-5 3.15 1.00 4.83 0.64 REP 1-5 4.12 1.50 5.00 0.71 5.2 Correlation Analysis
Table 5.2 shows the Pearson correlation between dependent, independent and control variables. It is interesting to see that there is a significant relationship between PSI and all dependent variables, while for CI there is only a significant link to EXP. RES. CT (r = 0.167, p < 0.01). Additionally, PSI has a positive, significant relationship with all of the control variables, except ENV.
The dependent variables are almost all positively correlated. Of these, only the dependent variable IMP. ACT. CT has negative significant correlations, while the other
32
dependent variables all correlate positively. There is a moderate regression between EXP. RES. CT and EXP. ACT. CT (r = 0.462, p <0.01).
The control variables have a significant relationship with more than half of the CT variables. The dependent variable EXP. PERS. CT is even positively and significantly correlated with all the control variables. The strongest correlation is between EXP. CULT. CT and REP (r = 0.379, p < 0.01).
This result provides a first indication of the existence of multicollinearity. The data shows multicollinearity when the independent variables and the control variables are too much correlated with each other (Field, 2009). Table 5.2 shows that there are no values above 0.800 between the independent variables, which might implicate that there is no multicollinearity. Multicollinearity was tested in more detail with tolerance and VIP. When conducting the collinearity statistics, the tolerance was between 0.950 and 0.978 and the VIP was just above 1. The tolerance value should be higher than 0.1 and preferably higher than 0.2. The VIF value should be lower than 10 (Field, 2009). As is clear from the correlation table, and the tolerance and VIP, multicollinearity is not a problem to my test.
33
Table 5.2: Pearson Correlations
IMP. ACT. CT IMP. PERS. CT IMP. CULT. CT EXP. RES. CT EXP. ACT. CT EXP. PERS. CT EXP. CULT. CT
CI PSI SIZE ENV REP
IMP. ACT. CT IMP. PERS. CT 0.020 IMP. CULT. CT -0.177 ** 0.141 ** EXP. RES. CT 0.009 0.020 0.098 * EXP. ACT. CT 0.130 ** 0.063 0.052 0.462 ** EXP. PERS. CT -0.032 0.127 ** 0.220 ** 0.263 ** 0.210 ** EXP. CULT. CT -0.158 ** 0.091 0.460 ** 0.205 ** 0.143 ** 0.362 ** CI -0.009 0.078 0.057 0.167 ** 0.082 0.060 -0.011 PSI -0.092 * 0.305 ** 0.227 ** 0.187 ** 0.216 ** 0.184 ** 0.133 * 0.654 ** SIZE 0.145 ** 0.142 ** 0.028 0.144 ** 0.201 ** 0.170 ** 0.020 0.150 ** ENV -0.079 -0.123 ** 0.057 0.182 ** -0.006 0.147 ** 0.071 0.150 ** 0.025 0.021 REP -0.060 0.108 * 0.209 ** 0.048 0.172 ** 0.160 ** 0.379 ** 0.009 0.221 ** 0.167 ** 0.110 *
Pearson correlations are the first line numbers, Spearman’s correlations are on the second line. *. Correlation is significant at the 0.05 level (2-tailed).
34 5.3 Testing H1
Hypothesis 1 predicted that IMP. CT is higher with more CI. A linear regression was conducted to establish whether the relation between IMP. CT and CI is significant.
Table 5.3 shows the results from the linear regression analysis. Since the independent variable is centered, the unstandardized beta-coefficients are given. These numbers represent the effect of increasing one unit of the independent variable with the value of the dependent variables IMP. ACT. CT, IMP. PERS. CT and IMP. CULT. CT. Assessment of the Shapiro-Wilk’s test (p-value = 0.377) shows the independent variable CI has an approximately norm distribution. Furthermore, the F-value is significant for all variables, which indicates that the model as a whole explains a significant amount of the variance in the IMP. CT.
Table 5.3: Outcome regression analysis H1
IMP. ACT. CT IMP. PERS. CT IMP. CULT. CT Variable Independent variable CI -0.002 0.082 ** 0.042 Control variable SIZE 0.156 *** 0.101 *** -0.010 ENV -0.103 -0.178 *** 0,031 REP -0.100 * 0.109 ** 0.215 ***
Model fit indicates
R² 0.034 0.055 0.048 Adjusted R² 0.025 0.046 0.039 F change 3.974 6.477 5.597 P-value 0.004 0.000 0.000 * significant at p < 0.1 ** significant at p < 0.05 *** significant at p < 0.01
The adjusted R² provides an indication of the variance in the dependent variable explained by the independent variables and the control variables. For IMP. ACT. CT, the variance explained is 2,5%, for IMP. PERS. CT this number is 4,6% and for IMP. CULT. CT the variance explained is 3,9%. The results of the control variables show that SIZE has a significant effect on IMP. ACT. CT and IMP. PERS. CT, but not on IMP. CULT. CT. REP and
35
SIZE are statistically significant for two variables, whereas ENV is only significant for IMP. PERS. CT.
For IMP. PERS. CT, the results show that CI is statistically significant, with β = 0.084 and p = 0.036. On the contrary, for IMP. ACT. CT and IMP. CULT. CT the results are not statistically significant, with p-value = 0.992 and p-value = 0.275. Therefore, hypothesis 1 is only supported for IMP. PERS. CT. A higher CI will allow for more tolerance in deviation in IMP. PERS. CT.
5.4 Testing H2
Table 5.4 shows the result for hypothesis 2, which predicted that EXP. CT is higher by more IC. The results suggest a significant model fit (p = 0.000), which explains 6,4% of the variance in EXP. RES. CT, 5,9% of the variance in EXP. ACT. CT, 5,6% of the variance in EXP. PERS. CT and 14,9% of the variance in EXP. CULT. CT. The results of the control variables show that SIZE has a significant, positive relationship with all EXP. CT. ENV is not significant for EXP. ACT. CT and EXP. CULT. CT, while REP is only insignificant for EXP. RES. CT.
There is a significant relation between EXP. RES. CT and CI (β = 0.138 and p = 0.002), and between EXP. ACT. CT and CI (β = 0.080 and p = 0.072). This indicates that the number of rules and procedures followed by professionals increase when CI increases. Therefore, hypothesis 2 is supported for EXP. RES. CT and EXP. ACT. CT.
36 Table 5.4: Outcome regression analysis H2
EXP. RES. CT EXP. ACT. EXP. PERS. EXP. CULT. Variables CT CT CT Independent variable CI 0.138 *** 0.080 * 0.033 -0.021 Control variable SIZE 0.127 *** 0.160 *** 0.119 *** 0.106 ** ENV 0.213 *** -0.051 0.151 *** 0.046 REP 0.008 0.175 *** 0.132 ** 0.455 ***
Model fit indicates
R² 0.072 0.067 0.064 0.156 Adjusted R² 0.064 0.059 0.056 0.149 F change 8.769 8.155 7.677 20.679 P-value 0.000 0.000 0.000 0.000 * significant at p < 0.1 ** significant at p < 0.05 *** significant at p < 0.01 5.5 Testing H3 5.5.1 Normality
Before analyzing my hypothesis, I had to test normality. I had to assess whether both the independent variables CI and PSI have an approximately normal distribution. Table 5.5 shows the results of the test for normality. CI has a Z-score for skewness of -7.02 and a Z-score for kurtosis of 1.10. PSI has a Z-score for skewness of 0.20 and a Z-score for kurtosis of 0.46.
Table 5.5: Normality test
Skewness Kurtosis Shapiro - Wilk
Independent variables Statistic SE Statistic SE Statistic Sig
CI -0.793 0.113 0.274 0.226 0.996 0.337
PSI 0.022 0.112 0.047 0.224 0.997 0.615
Furthermore, the Shapiro-Wilk’s test is insignificant for both the CI (p-value = 0.377) and PSI (p-value = 0.615). Thus, these findings suggest that the assumptions of normality are met and further analysis can be done.
37
5.5.2 Results
The third and main hypothesis described the interaction effect of CI and PSI on EXP. CT. It assumed that both independent variables together (PSI and CI) weaken the independent relation of the variables with the outcome variable. Thus, it assumed that the involvement of PSI negatively affects the relationship between CI and the number of controls in de MCS.
Table 5.6 shows the outcome of the linear regression analysis. Under model 1, the control variables are tested. The F value for model 1 is significant, with p-value = 0.000. This means that the control variables influence the dependent variables. Additionally, the adjusted R² showed that the total variance explained by the factors EXP. RES. CT (4,7%), EXP. ACT. CT (5,4%), EXP. PERS. CT (5,7%) and EXP. CULT. CT (15,0%) is 30,8%. The results show that most of the control variables are statistically significant, except for the relation between ENV and EXP. ACT. CT (β = -0.034 and p = 0.577), the relation between ENV and EXP. CULT. CT (β = 0.043 and p = 0.483), and for the relation between REP and EXP. RES. CT (β = 0.004 and p = 0.937).
Under model 2, the dependent and independent variables are tested. The second hypothesis assumed that EXP. CT is higher by more CI. In table 5.4, CI is significant for EXP. RES. CT and EXP. ACT. CT. However, in model 2, CI is not significant for all CT. Therefore, in model 2, this hypothesis is not supported. The adjusted R² is 37,4%, which is explained by the factors EXP. RES. CT (7,0%), EXP. ACT. CT (7,8%), EXP. PERS. CT (7,5%) and EXP. CULT. CT (15,1%). The ANOVA analysis indicates that model 2 as a whole is significant (p = 0.000).
The interaction effect between PSI and CI is tested in model 3. The results show that this interaction is not statistically significant, with β = -0.033 and p-value = 0.720 for EXP. RES. CT, β = 0.130 and p-value = 0.148 for EXP. ACT. CT, β = 0.017 and p-value = 0.832 for EXP. PERS. CT, and β = 0.137 and p-value = 0.135 for EXP. CULT. CT. Therefore, hypothesis 3 is not supported. The effect of the independent variable CI on the dependent variable EXP. CT does not change as a consequence of the interaction between PSI and CI. Furthermore, the scores show that the control variables are still not significant between ENV and EXP. ACT. CT, between ENV and EXP. CULT. CT, and between REP and EXP. RES. CT. Additionally, under model 1, REP has a statistically significant effect on EXP. PERS. CT (β = 0.129 and p = 0.011), while under model 3 this effect is not statistically significant (β = 0.082 p = 0.113). The ANOVA analysis indicates that model 3 is significant (p = 0.000). The adjusted R² shows
38
that the total variance explained by the factors EXP. RES. CT (6,8%), EXP. ACT. CT (8,0%), EXP. PERS. CT (7,3%) and EXP. CULT. CT (15,3%), is 37,4%
39 Table 5.6: Outcome regression analysis H3
* significant at p < 0.1 ** significant at p < 0.05 *** significant at p < 0.01
40
Although the interaction between PSI and CI is not statistically significant, figure 4 shows that there is an interaction, except for EXP. PERS. CT. The findings show that PSFs increase their reliance on EXP. RES. CT, EXP. ACT. CT and EXP. CULT. CT together with an increase in PSI, when CI is high. In contrast to EXP. ACT. CT and EXP. CULT. CT, PSFs also increase their reliance on EXP. RES. CT together with a decrease in PSI, when CI is high. Thus, the amount of rules and procedures followed by professionals are tighter when CI increases, irrespective of the level of PSI.
According to Faraway (2014), an insignificant interaction does not necessarily means that an interaction effect does not exist in the population. Therefore, an additional test was conducted.
Figure 4: Interaction effect
5.6 Additional analysis – H3
For this test, the sample was divided into two subsamples based on the median of PSI, which is 3.53. The results below the median were shown as low PSI and the results above the median were shown as high PSI. Table 5.7 provides the outcome of the additional analysis.
Low CI High CI EX P . RES UL T CT Low PSI High PSI Low CI High CI EX P . A CT ION CT Low PSI High PSI Low CI High CI EX P . P ERS ON NEL CT Low PSI High PSI Low CI High CI EX P . CUL T URA L CT Low PSI High PSI
41 Table 5.7: Results additional test low PSI - high PSI
Hypothesis 3 assumed that PSI negatively affects the relationship between CI and EXP. CT. Table 5.7 shows that there are statistical significant relations among the results for low PSI (β = 0.158 and p = 0.035 between CI and EXP. RES. CT and β = -0.162 and p = 0.043 between CI and EXP. CULT. CT), while the results for high PSI are statistically insignificant.
I expected that an increase in PSI would give professionals more autonomy of their work. An increase in PSI would enable PSFs to attract and retain professionals, since they prefer to act on their expertise without feeling the burden of being controlled. Given the results, the relation between CI and EXP. CT is not affected by a high PSI. These results indicate that the amount of rules and procedures followed by professionals are not affected by high PSI when CI increases. Therefore, hypothesis 3 is rejected.
42
6 Discussion and conclusions
This section describes in depth the findings of the study and the theoretical implications. First, the main findings of this research are discussed, followed by an overview of the research’ limitations. Finally, suggestions for further research are described.
6.1 Discussion of the Findings
The objective of this thesis was to analyze the impact of CI on the degree of CT within PSFs, and how this is affected by PSI. A survey was filled out by employees working at operational and low-management levels at PSFs. The data consisted of answers derived from employees of several PSFs working in different industries and with different educational backgrounds.
The study examined the expectations that need to be met in order to affect the degree of CT within PSFs. Using the consumer-professional agency theory, I considered the degree of CT to be effective when the variables CI and PSI were present. As expected, results show a positive significant relation between CI and IMP. PERS. CT. This finding underlines the importance of the hiring process in a PSF, since PSFs need to constantly employ professionals with a high level of knowledge. According to this result, deviation from human resource standards is tolerated when CI increases. A possible explanation for this is that PSFs are forced to attract and retain professionals because of their complex knowledge. Without professionals, there is no ability for PSFs to create value for customers and, therefore, the respective firm’s objectives will not be achieved. However, results do not show a significant relation between IMP. ACT. CT and CI, and between IMP. CULT. CT and CI. This implies that, in the presence of CI, PSFs do not allow any deviation from standard processes, procedures, rules and routines, and display no low value congruence. In line with this, it is possible that PSFs rely on these controls to manage the behavior of professionals, which leads to a PSF’s ability to create value for customers.
The study also examined the relation between CI and EXP. CT. Results show a positive significant relation between CI and EXP. ACT. CT, and between CI and EXP. RES. CT. This observation was important for this study since my goal was to analyze the interaction effect of PSI on the relation between CI and EXP. CT. This did not weaken the relation of the explicit use of controls in MCS in the presence of PSI. This implies that in the presence of PSI, a PSF does not rely on loose controls to manage CI. An additional test partially underlines this finding. According to these results, to manage CI the degree of EXP. CT is not affected when
