The Complex Tale of HRM System Adoption:

The Complex Tale of HRM System Adoption:

Different Users, Different Reactions, Different Causes S. H. Koop

(s1960865)

Word count (including references and appendices): 21,526

June 2016

Master Thesis programme MSc Change Management (Business Administration) Faculty of Economics and Business

University of Groningen

Supervisor: dr. B. J. M. Emans

Co-assessor: prof. dr. J. Surroca

Foreword

This thesis was not only written as a completion of the master programme MSc Change Management (Business Administration) at the University of Groningen, but also as a completion of my entire academic curriculum. In this thesis I have been able to combine the knowledge and skills that I obtained throughout the years – during the bachelor programme BSc Psychology as well as both master programmes MSc Industrial and Organisational Psychology and MSc Change Management (Business Administration). Being given the opportunity to combine the thesis’ trajectory with an internship at Avebe U.A., I have, moreover, acquired an extra set of practical skills that may be useful in the near future.

I would like to thank Evelien van de Poll (HR Director at Avebe U.A.) for her positive and personal supervision that enabled me to successfully complete my thesis and internship in time. Furthermore, I would like to thank Ben Emans for his guidance throughout the trajectory and commentary on my drafts. Finally, a special thanks goes to Wilco de Boer for supporting me no matter what, and to Dorien Ikink for being a perfect partner in crime once again.

Suzanne Koop

Table of contents

Abstract ... 4

The Complex Tale of HRM System Adoption: Different Users, Different Reactions, Different Causes ... 5

Acceptance and resistance ... 7

System adoption ... 9

Content, context and process ... 10

Implementability levers ... 12

Different groups of users ... 17

The moderating role of social influence ... 20

Method ... 21

Research site ... 21

Participants ... 22

Materials ... 23

Procedure ... 29

Results ... 30

Descriptive Analysis ... 30

System adoption ... 32

Implementability levers ... 32

Different groups of users ... 34

Social influence ... 36

Discussion ... 38

Limitations of the Study and Suggestions for Future Research ... 56

Conclusions ... 57

References ... 58

Appendix A – Items System Adoption and preliminary reliability analysis... 81

Appendix B – Adjusted items Implementability Levers and preliminary reliability analysis ... 83

Appendix C – Items Social Influence and preliminary reliability analysis ... 86

Appendix D – Factor Analysis ... 87

Abstract

Systems that are implemented by the Human Resource Management (HRM) department are often relevant for the entire organization, yet the implementation processes cannot be steered through formal power of line management. Rather, other types of power need to be

employed. Implementation success is conceptualized as system adoption and presented as a two-dimensional construct comprising system acceptance and resistance as two distinct yet related continuums. A multitude of manageable implementability levers within the categories of content, process, and context of a new HRM system (implementation) are hypothesized to positively affect these two dimensions of system adoption. A model is formulated that incorporates these relationships as well as group-level differences between managers and employees, thereby rendering a more holistic view on system adoption. Finally, social influence is incorporated in the model as a moderator of the relations between the

implementability levers and system adoption. The outcomes of the quantitative study show that the implementation of HRM systems is a complex process, in which acceptance and resistance are indeed distinct yet related, in which multiple antecedents apply to the two dimensions, in which group-level differences do indeed exist, and in which social influence should also be taken into account. More specifically, the results indicate that employees are more likely to resist a new system than managers, which may be explained by the notion of hierarchical distance. Additionally, especially embeddedness of a new system appears to be influential in increasing system adoption. HR managers are therefore recommended to enhance system adoption organization-wide by investing in the embeddedness of the system for managerial application and the high quality involvement of managers in the

implementation process, so that managers can in turn enhance system adoption of employees.

Keywords: Human Resource Management; system implementation; system adoption;

implementability levers; managers; employees; social influence

The Complex Tale of HRM System Adoption: Different Users, Different Reactions, Different Causes

"The price of doing the same old thing is far higher than the price of change"

- Bill Clinton

As modern day technology advances, organizations are becoming increasingly

sensitive to new techniques that can be used in their operational management. However, they are confronted with an enormous amount of options to choose from and an even bigger amount of personal preferences of employees to take into account (Gagnon, 2001). A

successful selection and incorporation of new technologies, therefore, is a process that needs thorough planning and close scrutiny in order to conform to as many wishes as possible and/or to have as many employees as possible use the systems effectively (Collins &

Williams, 2014).

In order to guide these complex implementation processes, a multitude of models have been proposed that provide fixed action steps to take and/or factors to adhere to when developing an action plan and when implementing the system (e.g., Emans, Postema, Weering, Peelen, & Boeve, 2011; Grohowski, McGoff, Vogel, Martz, & Nunamaker, 1990;

Venkatesh & Davis, 2000; Yadav, 2015). Moreover, different models exist that are

specifically construed to guide implementation trajectories within different organizational

departments. One especially interesting field for system implementation is Human Resources

Management (HRM). As HRM is a so-called staff-department, its functions are sometimes

somewhat ambiguous and underestimated by other organs within organizations (Wilkenfield,

1986). This can, amongst other reasons, be attributed to the fact that staff is not in line with

the rest of the organizational structure. Rather, they exist to facilitate other parts of the

hierarchy. Because of this they do not hold the same type of power as formal line managers (Agarwal & Vrat, 2015; Guest & King, 2004; Sheehan, De Cieri, Cooper, & Brooks, 2014).

A consequence of this non-line management is that the process of implementing new systems is somewhat different for HRM than it is for formal line management (Daudigeos, 2013).

Still, when HRM does implement new systems, there may be effects for the entire

organization (i.e., all employees). Thorough analysis of this situation is therefore essential for finding means (other than formal power) that may enhance the effectiveness of system

implementations.

The aim of this thesis is to further validate a theoretical model that has been proposed to offer factors that may enhance system implementability (Emans et al., 2011). Moreover, in this thesis I attempt to extend the model in a variety of ways. First, I propose that system adoption is a two- (instead of one-) dimensional construct (thereby validating previous work by van Offenbeek, Boonstra, & Seo, 2013). Second, I suggest that the effects of different implementability factors have different effects for different groups of users of the system (e.g., Oreg, Vakola, & Armenakis, 2011). Finally, I suggest that social influence (i.e., the shaping of perceptions by social interactions) may moderate the effects of different implementability factors on the adoption of systems (e.g., Bouckenooghe, De Clercq, &

Deprez, 2014).

By exploring a more nuanced view of the model, this thesis adds to a growing body of literature that can be used to assist organizations in their daily operations. If different

implementability factors can be used to enhance system adoption within different groups of

users, HRM may be able to tailor their implementation strategies to the different groups in

order to be able to optimise overall system adoption and lower implementation costs at the

same time.

The remainder of this introductory section further explains the different constructs and propositions. The first two subsections introduce the concept of system adoption and the notion that adoption may be a two-dimensional rather than a one-dimensional construct.

Second, a variety of factors (or levers) will be discussed that are hypothesized to affect system adoption. Third, arguments are provided for the suggestion that different types of users may demonstrate different types of user reactions. Finally, the last subsection explains why social influence may alter the aforementioned effects of the implementability factors on system adoption.

Acceptance and resistance

Despite the enormous and ever-growing body of literature regarding change management and all the corresponding practical advices for organizations, many change initiatives still fail to succeed (Burnes & Jackson, 2011), and even though it is sometimes (mistakenly) not regarded as such, a system implementation is also a change initiative (Keen, 1981). Many practitioners and researchers have acknowledged that change is hard and have proposed several ways of enhancing success (e.g., Petter, DeLone, & McLean, 2013; Sharma

& Yetton, 2007). However, before making statements about how to enhance change success, one has to be very specific about what exactly is change success. Only when this is

elaborated upon thoroughly, one can begin to unravel the antecedents of that specific construct of success.

When browsing through the literature on system implementations it becomes evident that researchers employ different constructs of success, leading to ambiguity and different streams of literature within the same topic of interest. Two terms that are especially pronounced are acceptance and resistance. The great majority of articles that have been published in the field of system implementation revolve around either acceptance or

resistance of systems (van Offenbeek et al., 2013). Moreover, when either one of those terms

is selected as the main focus of an article, the other term is often casually mentioned as opposite effect (e.g., Lapointe & Rivard, 2005). In other words, when acceptance is being described, resistance is mentioned as being the opposite of acceptance, and when resistance is being described, acceptance is mentioned as being the opposite of resistance. This would be perfectly logical if the relationship between acceptance and resistance was one with solid foundations. However, there is reason to believe that acceptance is fundamentally something different than resistance (cf. Bhattacherjee & Hikmet, 2007).

One major clue to this perception is that the articles on acceptance are based on distinctly different premises than the articles on resistance. As van Offenbeek, Boonstra and Seo (2013), who reviewed the distinction between acceptance and resistance, state:

“[W]hereas the acceptance lens stresses individual cognition, the resistance lens emphasizes the social and power relationships between groups” (p.437).

One well-known en widely used model in the acceptance literature on system implementation is the Technology Acceptance Model (Davis, 1989). In this model,

acceptance is described as a process in which attitudes towards use are related to intentions to use, and intentions to use are subsequently related to actual use. Acceptance, therefore, ultimately captures the extent to which a system in actually used. Resistance literature, on the other hand, takes a broader perspective, in which varying behaviours are being explained that follow from any change (in this case, a system implementation). To do so it uses theories that describe how individuals engage in social comparison in order to determine whether or not they might benefit from the change (e.g., Equity-Implemenation Model, Joshi, 1991) and adjust their behaviour accordingly in order to aid or stop the change.

Clearly, the two dimensions (i.e., acceptance and resistance) are not necessarily

similar (that is, nothing else than each other’s opposite). Whereas the acceptance dimension

visualizes the extent to which a system is actually used, the resistance dimension visualizes

different types of behaviour that either enforce or thwart a change to some degree. To assume that a low degree of system usage equals preventing change would be a mistake. Evidently, the two might well be related, as someone who wants to prevent the change may also decide not to use the implemented system, but it need not necessarily co-occur. Imagine, for

example, someone who supports a new ICT-system, but does not understand all of its features and therefore only uses is sporadically. Similarly, someone who fanatically opposes the new system may still be forced to use it in order to be able to execute certain tasks.

System adoption

As stated, it might well be that the two dimension are related to one another. More specifically, it might be that the two dimensions together form one construct. This is exactly what van Offenbeek, Boonstra and Seo (2013) proposed in their article on system adoption: a two-dimensional rather than one-dimensional explanation of system adoption, in which acceptance and resistance form two axes that create four quadrants (i.e., four different groups of reactions, see Figure 1).

The two-dimensional perception of system adoption is yet to be validated but seems promising for enhancing the literature on system implementation. By bringing together two distinctly different yet related themes (i.e., acceptance and resistance), two streams of literature that existed in parallel can be combined in order to form a more coherent overview of the explanations for successes and failures that are encountered during system

implementations. This entails that behaviours that are less common (e.g., low actual use, but supportive behaviour), yet also possible, can now also be recognized. Apart from the

theoretical advancement, the new construct also offers a highly practical advantage:

practitioners are enabled to modify their change strategies to be of relevance for different

types of system adoption behaviours. By tailoring change strategies multiple involved parties

can be managed simultaneously, thereby improving chances of implementation success (Milis & Mercken, 2004).

Still, for this practical advantage to be applicable, the antecedents of the two dimensions of system adoption first need to be discovered. As stated by van Offenbeek, Boonstra and Seo (2013), “Understanding the antecedents of user reactions … offers cues on how to manage them” (p.447). Hence, merely being able to observe different types of user reactions does not provide any tangible insight into actually managing a change. Therefore, apart from being validated, the two-dimensional construct of system adoption needs to be incorporated into a more elaborate model that also accounts for system adoption antecedents.

Content, context and process

When discussing antecedents of (the success of) organizational change, researchers often frame their arguments along a framework that was proposed almost 30 years ago (Pettigrew, 1987) and still holds today (e.g., Devos, Buelens, & Bouckenooghe, 2007;

Walker, Armenakis, & Bernerth, 2007). Seemingly simple and yet so comprehensive, Pettigrew’s model has been widely accepted and used in the past 30 years. The model, visualized as a plain triangle, incorporates three dimensions of change that need to be taken into account when considering any organizational transformation (regardless of its scope):

content of the change, context of the change, and the change process. These three dimensions of change can best be explained with three questions: 1) Content: What is being changed?; 2) Context: Under what conditions is it being changed?; and 3) Process: How is it being

changed? (Pettigrew, 1987).

Content describes the specific object of change, such as an HRM system. Typically,

the content dimension comprises several features of the object of change. The context angle

of the triangle is divided into two subsections: inner context on the one hand and outer

context on the other. An analogy with the human body is appropriate: the inner context can

be compared to everything inside the body; the structure of the body, the processes within the body. The outer context can be compared to both the material (such as geographical location) and intangible (such as culture) aspects of the environment surrounding the body. Finally, the process dimension describes all change-related activity (both planned and un-planned) that contributes to the change (both positively and negatively).

Several authors have used the three dimensional-framework in the light of information systems (e.g., Bernroider, Koch, & Stix, 2013; Serafeimidis & Smithson, 1999; Stockdale &

Standing, 2006). They generally agree on the framework’s usefulness, especially because of its multi-layered and interactive nature that matches the complexity of system

implementations perfectly well. However, in these articles the three dimensions have not been explicitly related to variables considering the acceptance or resistance of a system, thereby leaving a gap between theorizing and opportunities for practical recommendations for managing system implementation. Still, there are reasons to believe that the three dimensions are indeed related to system acceptance and system resistance.

For example, Giangreco and Peccei (2005) found that being involved in the change process relates to resistance to change (i.e, process dimension). Likewise, van Dam, Oreg, and Schyns (2008) found several change process characteristics to be related to resistance (i.e., process dimension). Venkatesh et al. (2003), on the other hand, determined a

relationship between facilitating conditions and usage of a new technology (i.e., context dimension). Finally, Davis (1989) recognized long ago that certain system features such as ease of use and perceived usefulness relate to actual system use (i.e., content dimension).

Even though these indications for a relationship between content, context, and process

variables and the two-dimensional construct system adoption are readily available, a holistic

model incorporating all factors remains absent. In this thesis, therefore, such a model is

proposed and investigated. In the next section, a model is described that has been used as a starting point for this research purpose.

Implementability levers

One model that has used the content, context, and process distinction as proposed by Pettigrew (1987) and that is of specific interest for the topic of HRM system implementation is a model by Emans et al. (2011). In this model, a number of so-called levers has been related to the implementability of HRM tools. The model is especially attractive because of its pragmatic nature. Rather than discussing change characteristics that do influence a change but are impossible to alter (and therefore less relevant for change practitioners), the model incorporates factors that are practical and manageable. For that reason, the model by Emans et al. (2011) was selected and used in this thesis.

Presented below are the definitions of different implementability levers that are presented in the model. Note that, given the scope of this thesis, not all levers from the original model have been incorporated in this study. In order to obtain a clear picture of the effects of the levers per cluster (i.e., content, process, and context), at least two levers were selected for each cluster. As the context cluster of the model of Emans et al. (2011) only comprises two levers, selection of context levers was not necessary. For the process and content clusters two and four levers were selected, respectively, that were rated as interesting by one of the HR directors at the research site.

Descriptions are kept to a minimum in an attempt to maintain a tangible piece of information. For more information and elaboration on (the origin of) the levers see Emans et al. (2011).

Process. As stated, process variables are those variables that are related to change-

related activities. For this paper, two process-variables from the model of Emans et al. (2011)

are of specific interest. One of those variables (i.e., publicity) was not originally incorporated

in the model, but was added after qualitative research (interviews with experts, Emans et al., 2011).

Participation. Participation (i.e., taking part in certain actions) is one of those variables that can intuitively easily be related to system adoption. Intuition, however, does not provide a solid foundation for practical advice and change strategies. Many authors have therefore sought to empirically validate the assumption, ultimately leading to the

development of collaborative action plans for change (e.g., Hunton & Price, 1994; Ives &

Olson, 1984). Given its rich background, it is therefore only natural to incorporate the variable participation in a model of system adoption. As defined by Emans et al. (2011), participation is “the degree to which those who have to enact an HRM programme were enabled to contribute to its development or, conversely stated, the degree to which they were confronted with a programme that was entirely developed by others people” (p.9).

Publicity. Publicity was added to the implementability levers by Emans et al. (2011) after interviews with eight HR functionaries, eight line managers and seven employees (from a total of eight different organizations) that had all recently been involved in an HRM

programme implementation. The term publicity is regarded as “[p]rogramme publicity generated by programme successes” (Emans et al., 2011, p.31). It does not necessarily refer to focused promotion or campaigning, but rather to a broader and less formal take on publicity, like lunch-talks between employees or short user stories in the organizational monthly magazine.

Content. Content variables are those variables that revolve around specific features of

the change object. For technologies, these are called system features. System features are

those characteristics of a system that differentiate the system from others, much like facial

features that allow us to distinguish between different faces (note that this does not imply that

features are limited to visible characteristics only).

Flexibility. In their model, Emans et al. (2011) originally proposed a content-variable called programme adaptability (“the degree to which involved organization members, when enacting an HRM programme, have freedom as regards the way they do so, or, conversely stated, the degree to which they are bound by strict rules and procedures”, p.7). However, after their qualitative research and subsequent factor analysis, Emans et al. (2011) concluded that this variable could be split into two different variables (i.e., programme flexibility and programme improvability).

Programme flexibility refers to the “non-existence of protocols” (Emans et al., 2011, p.19). Hence, a system is labelled more or less flexible when there are few or many policies with regard to system use, respectively. Naturally, the use of policies is not without reason.

By using certain rules and procedures, organizational processes can be standardized and formalized more easily. Especially for larger organizations and/or organizations with different geographical locations, formal HRM procedures are needed in order to be able to ensure equal treatment for all employees. These procedures, however, do have a downside:

more rules usually lead to more rigid systems. Rigid systems, in turn, are less user-friendly (as they do not allow for adapting to personal preferences) and can therefore potentially decrease actual system usage.

Improvability. Improvability is a broader take on flexibility. It refers to “room for continuous programme improvements” (Emans et al., 2011, p.19). Whereas the variable described above relates to room for personal adjustments and personalized ways of working with the system, the variable improvability relates to the possibility of enhancing the entire system when general problems or difficulties are encountered. Clearly, improving a system is likely to enhance its actual use.

Simplicity. One of the biggest dilemmas of application development is providing

many options versus keeping it simple. The more options and functionalities a system has, the

higher are the chances that someone will encounter difficulties in using the system. Hence, simplicity really is a very straightforward variable in determining system usage: when a system is simple, it is probably more likely to be used and to be used as intended. As defined by Emans et al. (2011), simplicity is “the degree to which a programme is devoid of elements that are hard to grasp for the actors involved” (p.8).

Embeddedness. Organizations that wish to implement a new system generally have two options: 1) hire an application developer to build a system that does exactly what it was intended for, or 2) search for and apply an existing system that matches the organization’s needs in the best possible way. As the first option is often outrageously expensive,

organizations tend to opt for the cheaper (and faster) second option. A consequence of this, however, is that certain system features may not optimally match with organizational

features, as the system was not built for that organization specifically. These mismatches may cause suboptimal functioning of the system, and may be reason for change recipients not to use the system to its fullest or not at all.

Emans et al. (2011) defined embeddedness as “the degree to which an HRM

programme fits in with existing processes in the organization or, conversely stated, the degree to which it is disconnected to those processes” (p.8).

Context. Apart from the content of a change and the process by which a change takes place, different variables surrounding the change may be able to alter (the degree of) system adoption.

It is this group of variables that highlights the appropriateness of the model of Emans

et al. (2011) for HRM system implementation specifically. As stated, HRM departments do

not hold formal line-power, and need to rely on forces other than power when implementing

new systems. Two things that appear influential are the extent to which the HRM department

shows so-called coworkership and the extent to which the HRM department is accessible for help and advice (Emans et al., 2011).

HRM coworkership. An HRM department can exert influence over the context of a change in order to steer the change in the right direction. Clearly, this is easier said than done, as there will be other influences that may counter all efforts (e.g., status quo bias, Polites &

Karahanna, 2012). One effort that is both within the HRM department’s scope, and that is unlikely to trigger opposition, is decreasing the burden of using a new system by making its use as simple as possible (i.e., facilitate its use). The HRM department can, for example, decide to offer tutorials regarding use of the system, or to allow for ‘testing’ time in

employees’ schedules. As defined by Emans et al. (2011), HRM coworkership is “the degree to which the HRM department relieves the work load or otherwise facilitates the task of organization members whose task is to enact an HRM programme” (p.12).

HRM accessibility. Another possibility for the HRM department in exerting influence over the change context is to simply be readily available for help and advice when employees encounter difficulties while using the system. This implies that the HRM department needs to make sure there is someone available at all times that knows as much as possible about the essentials and functioning of the new system. This is also easier said than done, as those HR functionaries may need extra training and guidance themselves before being able to help others. Emans et al. (2011) define HRM accessibility as “the degree to which the HRM department can be contacted for help and advice each time those who have to enact an HRM programme are in need of help and advice” (p.12).

The definitions and arguments above all indicate positive effects of the implementability levers on system adoption (for an overview of the levers and their

definitions, see Table 1). A distinction can be made, additionally, for levers that are likely to

affect system acceptance (i.e., system use) and levers that are likely to affect system support.

Whereas system usage may be mainly dependent on the (features of the) system itself as well as factors that facilitate usage, system support may be more likely to be affected by features of the implementation process. In other words, content and context levers appear to be specifically aimed at decreasing the burden of using a system, whereas process levers appear to be aimed at ensuring a positive perception of the new system. In line with these

propositions, the following hypothesis is proposed:

H1: Each of the implementability levers included in Table 1 is hypothesized to

positively affect system adoption. More specifically, it is hypothesized that the content levers (i.e., flexibility, improvability, simplicity, and embeddedness) and the context levers (i.e., HRM coworkership and HRM accessibility) are primarily responsible for increasing system usage, whereas the process levers (i.e., participation and publicity) are primarily responsible for increasing system support.

Different groups of users

Described above are eight levers that are hypothesized to alter either the acceptance or the resistance dimension of system adoption. It may be overly simple, however, to assume that these eight levers have exactly similar effects on every individual that is involved in a change situation (Joshi, 1991; Oreg et al., 2011; Srite & Karahanna, 2006; Venkatesh et al., 2003). Even though all levers may have a positive effect on system adoption for all

individuals, the sizes of the effects may differ.

In theory, to know which levers affect which person the most would be of great value, because it would result in the possibility to implement individually tailored change strategies.

In practice, construing individually tailored change strategies is of course highly unfeasible.

Moreover, given the enormous diversity of mankind it would be ridiculous to aim for

identifying all personal factors that cause differences in change perceptions. Zooming out, however, provides more attainable possibilities. Rather than assessing individual differences, it may be worthwhile to assess differences at group level, so that implementation strategies may be adjusted to match the different needs of the different groups. Bouckenooghe (2012) similarly argued for analysing commitment to change at unit rather than individual level, given that the majority of change trajectories are also aimed at units or organizations as a whole (Burke & Litwin, 1992).

Other authors have also recognized this need to distinguish between different groups.

Especially relevant is one article by Valero and Sanmartin (1999), who categorized different groups of users of information systems. They highlighted the importance of doing so by referring to the possibility to design different options within a system to suit the different knowledge bases of users. Their assertion was that different categories of users within an organization have different roles within that organization and corresponding different types of knowledge about that organization. Indeed, their results indicated that a distinction could be made between so-called receivers of a system (in their case, students) and those users that could use more system functionalities (in their case, teaching staff and information services).

Hence, there is reason to assume that a distinction between different groups of users can be made. These groups of users may be classified according to their organizational roles.

Those users with more complex organizational roles are likely to have more complex system user roles. To have a more complex user role implies more exposure to the content, context and process of a system (implementation). The implementability levers may therefore have greater influence on those complex users than on end-users of the system.

Oreg et al. (2011), who conducted a 60-year review of articles on reactions to change,

showed that several job characteristics are related to differences in change perceptions and

reactions. Amongst others, change readiness was reported to be higher for individuals with

more complex jobs (Cunningham et al., 2002; Eby, Adams, Russell, & Gaby, 2000) and for individuals that experienced more autonomy (Weber & Weber, 2001).

When considering HRM system implementation specifically, a logical categorization of user roles could be managers versus employees. Per definition, systems that are

implemented by the HR department have to do (to at least some extent) with personnel. There are, in turn, two broad categories of personnel: managerial and professional employees (Barker, 2010, as in Krausert, 2014). In this thesis, managerial employees will simply be referred to as managers and professional employees will simply be referred to as employees.

A categorization in which managers are distinguished from employees is common in post-industrial organizations (Watson, 2011) and is often used in empirical research (e.g., Krausert, 2014; Skakon, Kristensen, Christensen, Lund & Labriola, 2011; Voerman & Van Veldhoven, 2007). In line with the abovementioned job characteristics, and as compared to employees, managers can (on average) be stated to experience more autonomy (Krausert, 2014) and hold more complex jobs (see, for characteristics of complex jobs, Cummings &

Oldham, 1997). Additionally, HRM systems are amongst other reasons implemented to facilitate and encourage manager-employee interactions (devolution of the HR function, cf.

Sheehan, 2005). Managers interact with multiple employees, whereas employees (in most cases) only have to report to one manager. Managers, therefore, per definition have to use such HRM systems more extensively than employees.

Adding this to a combination of 1) the notion that implementability levers may have

greater influence on complex users than on end-users of a system and 2) the findings that

employees with more complex jobs and autonomy report greater readiness for change, in this

thesis I propose that the effects of the implementability levers on system adoption will be

larger for managers than for employees.

H2: It is hypothesized that different groups of users will display different reactions towards the implementation of a new system. More specifically, it is hypothesized that the effect sizes of the implementability levers on system adoption will be larger for managers than for employees.

The moderating role of social influence

The sections above argue how different change content, context, and process variables may alter the degree of system adoption (two-dimensionally conceptualized) for different groups of system users. By incorporating group-level differences in this theoretical

proposition, a presumably overly simple model has been extended towards a more coherent one. Still, the model remains rather static, neglecting one important aspect of organizational life: interaction.

Being social creatures, people have the need to interact with others (Baumeister &

Leary, 1995). Especially when situations are ambiguous, people seek help in processing and forming an opinion about what is going on (Festinger, 1954). This socially constructed process of absorbing, digesting, expressing, and adjusting environmental information is called sense making (Salancik & Pfeffer, 1978).

Sense making also occurs among employees when organizations adopt new

procedures and systems, rendering change attitude a socially formed construct (Ford, Ford, &

D'Amelio, 2008). Whereas early in the change process individual’s attitudes may primarily

be shaped by content, context, and process features of the change, later in the change process

change perceptions are also shaped by social interactions. These social interactions cause a

convergence of attitudes towards the change (Leonardi & Barley, 2010) and may therefore be

influential in determining the degree of system adoption.

Kozlowski and Klein (2000) argued that ignoring this social aspect of forming attitudes towards change would lead to inadequate models, providing distorted views of reality. Despite this, the effect of social influence in the context of system implementation has not been tested explicitly up until now. In this thesis, therefore, social influence is also

incorporated in the theoretical model, be it in an exploratory fashion.

H3: It is hypothesized that social influence will have an effect on the relations between the implementability levers and system adoption. As this is an exploratory rather than confirmatory part of this thesis, neither effect size nor direction of the effect is hypothesized.

Method

The aim of this study was 1) to validate the two-dimensional construct (HRM) system adoption (i.e., acceptance and resistance as separate dimensions), 2) to assess the effects of content, context, and process variables on that two-dimensional construct (for two different groups of users, namely employees and managers), and 3) to explore the moderating effect of social influence on those relationships. A quantitative approach (in this case, an online questionnaire) was deemed most appropriate for this research purpose.

Research site

The research site at which the online questionnaire was distributed was Avebe U.A., an internationally oriented cooperation that manufactures starch products from potatoes. Its headquarters are situated in Veendam, The Netherlands, and its factories are located in The Netherlands, Germany, and Sweden.

Three years ago, the HRM department (headquarters, Veendam) implemented a new

procedure for managing employee performance at all Dutch locations (in all, 1023

employees). The new procedure encompassed three employee-manager meetings per year:

one meeting to develop goals for the employee that related to the organizational strategy, a second meeting to discuss the progress of attaining those goals and adjusting them if needed, and a third meeting to evaluate the year and attainment of the goals.

One year after this HR-cycle implementation, the HRM department rolled out a new ICT-system (SuccesFactors) that was designed to facilitate the HR-cycle. SuccesFactors encompasses several functionalities to relieve HR managers, managers, and employees from some of the administrative burden that accompanied the HR-cycle. Some of the most

important functionalities: meetings can be digitally requested and approved, goals can be entered and progress of goals tracked, and minutes of the meetings can be digitally

documented, saved, and checked by both manager and employee. Every employee can view and approve his or her own personal details, whereas every manager can view and edit the personal details of all of his or her employees. Moreover, HR managers can export overviews of employees, check the quantity and quality of the meetings, and calibrate assessment of employee performance organization-wide. Use of SuccesFactors is mandatory for every employee, manager, and HR manager.

Proper use, however, does not necessarily follow from obliging use of the system. The HRM department received multiple complaints about SuccesFactors and heard rumours about managers not using the system as recommended. They therefore felt a need to conduct a thorough evaluation of the system and its usage, which fitted perfectly well with the researcher’s objective to investigate system adoption and its antecedents.

Participants

In total, 553 (out of 1023) Dutch employees of Avebe U.A. filled in (at least some

part of) the questionnaire via the online link they received by e-mail (e-mails were sent by the

researcher) or the (identical) link that was posted on the organization’s intranet. Every

employee at Avebe U.A. has his or her own personal e-mail account that needs to be used at least once a month to register working hours, it was therefore expected that the majority of employees would encounter the link that was sent by e-mail. In order to remind employees to check their inboxes, posters were put up on production sites and managers were asked to motivate employees to fill in the questionnaire.

After inspection of the raw dataset, 291 questionnaires were deemed appropriate for analysis. Incomplete questionnaires (n = 207), questionnaires that were ended because the participant indicated to have never used the system (n = 36), questionnaires that showed signs of so-called satisficing behaviour (e.g., non-differentiation, Barge & Gehlbach, 2012, n = 2), and questionnaires with additional comments about the quality of the answers (n = 17) were excluded from subsequent analyses.

235 of the participants were men, the remaining 56 were women. Their ages ranged from 19 to 65 (M = 51.39, SD = 9.20). Educational levels ranged from secondary education (n = 9) to intermediate vocational education (n = 126), higher vocational education (n = 94), and university (n = 51) – the remaining 11 participants selected other as answer.

Organizational tenure had a minimum of one year and a maximum of 49 years (M = 22.68, SD = 13.57) and finally, of all participants, 55 responded as manager and the remaining 236 as employee.

Materials

An online questionnaire was used to collect as much data as possible in a total of three

weeks. The questions that were used in the questionnaire were reviewed a couple of times

(two times by three individual HR managers and one time by two members of the Works

Council) in order to ensure appropriateness for the research site. Moreover, the questionnaire

was pilot tested by two employees that had no prior knowledge about the subject and/or

questions used.

The final questionnaire comprised questions concerning background information (including system user role), implementability levers, social influence, and system adoption.

As the survey was distributed among Dutch employees of Avebe U.A., the questionnaire was constructed in Dutch. The original scales for social influence and system adoption were construed in English and therefore needed to be translated to Dutch in order to be relevant for this study. After initial translation, the items were peer reviewed by a fellow student as well as this thesis’ supervisor to make sure item content was unaltered.

Before the data analysis, the variable scales were checked for reliability (Cronbach’s Alpha) and validity (Factor Analysis). For scales that were adjusted after the Factor Analysis, additional reliability checks were performed. Described below are the final scales as used in the analysis (number of items and Cronbach’s Alpha for each final scale). Appendices A, B, and C provide the original items of the scales as well as the translated and/or adjusted items as used in the questionnaire. Cronbach’s Alphas for each initial scale are also provided in the appendices. A description of the procedure of the subsequent Factor Analysis as well as its outcomes are provided in appendix D.

System adoption. The concept of system adoption as a two-dimensional construct is a fairly new concept. In other words, recognizing acceptance and resistance as two separate dimension rather than two ends on a single dimension is a fairly new idea. A consequence of this is that no scales have yet been construed that assess either acceptance or resistance. A multitude of scales that assess acceptance and resistance simultaneously does exist. However, these scales are not relevant for testing the two-dimensionality of system adoption. For this particular study, therefore, new scales have been construed that do assess either acceptance or resistance. The scales were construed using qualitative research outcomes (Offenbeek,

Boonstra, and Seo, 2013) as guidelines and while taking into account the specific situation at

the research site.

Acceptance. Acceptance of the new system was assessed by measuring system usage.

A scale was construed that did not assess if a respondent used the system, but rather the extent to which he or she used the system as intended. This method of assessment and a similar way of scale construction can also be found in the article of Bouma and Emans (2005).

In order to be able to know what exactly was the intended use of the system, an HR manager of the organization was asked to provide a list of system actions for employees (i.e., end-users) on the one hand, and for managers (i.e., complex users) on the other hand. These actions were subsequently transformed to measurable items (by the researcher).

For both employees (α = .84, e.g., “I use SuccesFactors to approve the notes of the performance meetings”) and managers (α = .71, e.g., “I use SuccesFactors to track the personal goals of my employees”) the final scales comprised three statements. Respondents could indicate the degree to which they agreed with each statement on a Likert-scale ranging from 1 (totally disagree) to 7 (totally agree).

Resistance. Resistance was assessed by measuring resistant versus supporting behaviours and attitude towards the system. The items that assessed resistance behaviours were based on qualitative research results from van Offenbeek, Boonstra, and Seo (2013).

The items were construed as statements. Respondents could answer each item by indicating the degree to which they agreed with it on a Likert-scale ranging from 1 (totally disagree) to 7 (totally agree). An example of an item that was used to assess resistance behaviours is: “I complain about the programme to my colleagues”.

The items that assessed system attitude were retrieved from Malhotra and Galletta

(1999) and were translated to Dutch and adjusted for the specific change situation (i.e.,

programme was changed into SuccesFactors). The items for measuring system attitude

deviated from most other items used in the questionnaire in the sense that, this time,

respondents were not asked to indicate the degree to which they agreed to the statements.

Rather, they were asked to indicate on a Likert-scale their opinion about using the new system (e.g., extremely wise, quite wise, slightly wise, neither wise nor foolish, slightly foolish, quite foolish, extremely foolish).

After the Factor Analysis (see Appendix D), the first two behavioural resistance items were excluded from further analysis because they had primary factor loadings on a factor structure other than resistance. All other items for resistance (both behavioural and

attitudinal) had primary loadings higher than the minimum criterion of .30 on the resistance factor structure and where therefore grouped to form one resistance scale. The final scale consisted of eight statements with a reliability score of α = .80.

Implementability levers. Scales for assessing the implementability levers were retrieved from Emans et al. (2011). As the management of Avebe U.A. requested to restrict the length of the survey as much as possible, not all items from the original list of questions were used in this study’s questionnaire. Only those questions that were deemed relevant for this study were included. Subsequently, they were altered in order to be of relevance for the specific situation at Avebe U.A. and to revolve around the implementation of SuccesFactors specifically. Some of the questions were adjusted once more or even excluded from the questionnaire after organizational reviews of the draft version or after remarks from the people who pilot tested the survey. Especially the questions containing double negatives were rated as being difficult.

The Factor Analysis for the implementability levers was conducted in two steps (see

Appendix D). First, all items for all levers were included in one analysis in order to find

factor structures according to the content, context, and process categories. Leaving out factor

loadings below .30, three factor structures appeared that were for the major part in line with

the three expected categories. One item for embeddedness was excluded from further analysis

given its relatively low factor loading (as compared to the other items). The same item was a reversed statement in the questionnaire, which may be a potential explanation for this result.

The results for publicity were somewhat ambiguous. In order to examine the validity of this scale in more detail, the publicity items were added to the second step of the Factor Analysis nonetheless.

In the second step, three separate Factor Analyses were executed (one for each

category of implementability levers) in order to examine the factorability of the levers within the categories. This time, the results for the process category showed a clear distinction between the publicity and participation items, thereby providing a sufficient indication for scale validity of the publicity scale. All other results indicated scale validity similarly, thereby confirming that the items could be converted into the expected scales.

In these final scales, 20 items assessed seven different implementability levers (flexibility was excluded from further analysis due to its low reliability score): improvability (two items, α = .77), simplicity (two items, α = .89), embeddedness (four items, α = .73), participation (three items, α = .89), publicity (two items, α = .67), HRM coworkership (three items, α = .90), and HRM accessibility (four items, α = .97).

The items were construed as statements (e.g., one of the improvability items, “The programme is continuously being improved based on practical experiences of users”).

Respondents could indicate for each statement the degree to which they agreed on a Likert scale ranging from 1 (totally disagree) to 7 (totally agree).

Social influence. Social influence was assessed by measuring two constructs with

existing scales (Venkatesh & Davis, 2000): Colleague Opinions and Subjective Norm. The

items were construed as statements. An example of an item of the Colleague Opinions scale

is: “The majority of my colleagues think that the change towards the new programme is a

good idea”. An example of a Subjective Norm item is: “People that are important to me

think that I should use the programme”. Respondents could indicate for each statement the degree to which they agreed on a Likert scale ranging from 1 (totally disagree) to 7 (totally agree).

All items were translated to Dutch and adjusted to the specific change situation (i.e., programme was changed to SuccesFactors – see, for both the original and transformed scales, appendix C). The Factor Analysis (see Appendix D) showed that the Subjective Norm and the Colleague Opinion items loaded onto the same factor structure, indicating that the items could be grouped as intended. The resulting Social Influence scale consisted of five statements with a reliability score of α = .87.

Background information. A number of questions considering background information (gender, age, educational level, and organizational tenure) were added to the questionnaire in order to be able to give a clear description of the set of participants, and in order to be able to include covariates in the data-analyses if needed. The number of

background questions was kept to a minimum, so that respondent identification through answer deduction would be difficult, if not impossible. This was done per request of the management of Avebe U.A., as anonymity is a sensitive issue within the organization.

One question of special relevance that was included in this question section was an

item that asked the respondent to indicate his or her type of user role with regard to the

system (“I use the system as manager” or “I use the system as employee”). The answer to this

question, in turn, determined which System Acceptance items were shown to the respondent

(Skip Logic function in Qualtrics). A third answer possibility to this question was “Up until

today, I have never used the system”. Participants that clicked on this answer were redirected

to the end of the survey.

Procedure

All participants were communicated with both directly and indirectly, through use of personal e-mails, news messages on the organization’s intranet, and messages on the

announcement screens at the factories. During the period of data collection (the survey was active for three weeks), the respondents were reminded of the study a couple of times both through messages by e-mail and messages on the organization’s intranet.

A general link directing to the questionnaire was e-mailed by the researcher to all Dutch employees of Avebe U.A., accompanied by a short message that explained the

research purpose. The questionnaire started off with a foreword in which mean response time was indicated (ten to 15 minutes), and in which was stated that the respondent could stop participating at any time. Moreover, anonymity of the results was guaranteed and contact information of the researcher was provided. Finally, the respondents were informed that clicking on the ‘next’ button implied conscious agreement to participate in the study.

Questions concerning the following subjects were presented in the following order:

background information, implementability levers, system adoption. After those questions, the respondents were provided with the possibility to leave any additional comments about either the questionnaire or the ICT-system SuccesFactors in an open text field.

In order to prevent ambiguity about the research topic (SuccesFactors), the questions about SuccesFactors were introduced with a short explanation. Also the respondents were told that answers were about their personal opinion and could therefore not be right or wrong.

After answering all questions, the participants were informed they had reached the end of the survey, thanked for their participation, and again provided with contact

information of the researcher.

Results Descriptive Analysis

Regression diagnostics were examined in order to check the dataset for outliers and to check the assumptions for regression analysis. Based on their resistance scores, two persons were indicated as outliers and were therefore excluded from further analysis (cases that exceeded the predicted value with more than three standard deviations were considered outliers). VIF- and Durbin-Watson-scores revealed no violations of collinearity or

autocorrelation. Visual inspection of the residual plots did not reveal any major violations of normality, homoscedasticity, or linearity, but did not assure non-violation either. Additional tests (Kolmogorov-Smirnov and Shapiro-Wilk) showed a violation of the assumption of normality of variances for both dependent variables (i.e., resistance and acceptance).

Moreover, the Brown-Forsythe test indicated a violation of the assumption of

homoscedasticity for acceptance (this specific test was selected because it uses the median instead of mean as statistic and is therefore less sensitive to non-normality than Levene’s test, Brown & Forsythe, 1974).

Despite these violations of normality (acceptance and resistance) and

homoscedasticity (acceptance), the data were used in their original form for the subsequent analyses. The consequence of this decision is that the outcomes of the analyses need to be interpreted with greater caution as compared to a situation with non-violation of the assumptions.

In order to check the dataset for covariates, correlations were calculated for the demographic variables and acceptance and resistance (see Table 2). From this data it can be concluded that age, level of education, and organizational tenure may act as covariates for acceptance as well as resistance and are therefore added to the analyses described below.

Another correlation matrix was created in order to scan the dataset for a first indication of

(significant) relations between the implementability levers, social influence, acceptance, and resistance (Table 3). Means and standard deviations of the scale variables are provided in the table as well. Finally, yet another correlation matrix was created, but this time the dataset was divided by user group (i.e., employees and managers), in order to scan the dataset for a first indication of differences between the two groups (see Table 4 for the correlations for employees and Table 5 for the correlations for managers).

One striking observation from the second correlation matrix (that holds all

implementability levers, social influence and both acceptance and resistance for the two user groups combined) is that all (but one) correlations are significant at the .01 level. This implies that, as expected, all variables are interconnected to some extent. From that perspective these results are encouraging for all that is to come (the more detailed regression analyses). Still, the matrix provides correlations that are statistical rather than causal relations. Conclusions regarding effects of one variable on another are therefore premature. Furthermore, 21 out of the 45 correlations can be considered fairly low (below .40), despite their significance.

The correlation matrices for employees (Table 4) and managers (Table 5) appear to be quite similar. Most correlations in the matrices are significant at the .01 level, which is in line with the outcomes described above. This indicates that for employees as well as managers the levers, the dependent variables, and social influences are interconnected to some extent. Even though most of the correlation sizes are also somewhat similar for employees and managers, there are a couple of differences that should be mentioned. For managers, the correlations of HR accessibility with the other variables are not as significant as for employees. For

employees, there is one lever (participation) that does not significantly correlate with either

acceptance or resistance (reversed). For managers, the correlations between participation and

acceptance and resistance (reversed) are significant.

The following sections describe the more detailed analyses that were performed in order to test the three hypotheses as described in the introductory section. The sections respectively present the results of 1) system adoption as a two-dimensional construct, 2) the relationships between the implementability levers and the two dimensions of system

adoption, 3) the differences in system adoption and implementability levers for employees versus managers, and 4) the moderating role of social influence.

System adoption

One of the aims of this thesis was to confirm and extend the notion that system acceptance and resistance are two distinct dimensions that together form the concept of system adoption. According to van Offenbeek, Boonstra, and Seo (2013) one simple way of assessing this notion is to plot the user reactions to the dimensions in a graph of which one axis represents acceptance and the other axis represents resistance (or, more specifically, reversed resistance). If the dimensions would depict the same underlying construct, user reactions would fall in either the low acceptance/low reversed resistance or the high acceptance/high reversed resistance quadrant. If the dimensions have different underlying constructs, user reactions in the other quadrants would also be possible (i.e., low

acceptance/high reversed resistance or high acceptance/low reversed resistance).

A visual inspection of the dimensions of system adoption was hence carried out accordingly (see Figure 2). From this graph it can be cautiously concluded that acceptance and resistance are indeed two separate yet related dimensions. The correlation between the two constructs (.48, see also Table 3) similarly indicates that the two dimensions do not entirely coincide with each other. For this reason, subsequent analyses to test the effects of the implementability levers and social influence were executed for acceptance and resistance separately.

Implementability levers

It was hypothesized that 1) all implementability levers would positively affect system adoption, and 2) that the content (flexibility, improvability, simplicity, and embeddedness) and context (HR coworkership and HR accessibility) levers would be primarily responsible for increasing system usage (i.e., acceptance) while the process levers (participation and publicity) would be primarily responsible for increasing system support (i.e., reversed

resistance). Linear regression analyses were executed to test these relationships. Age, level of education, and organizational tenure were added as covariates given their significant

correlations with the dependent variables (for correlations, see Table 2). Flexibility was not included in the analysis due to its low scale reliability.

For acceptance, positive significant results were found for embeddedness (β = .41, t = 4.94, p = .05) and publicity (β = .21, t = 2.73, p = .01), a marginally significant positive result was found for HR coworkership (β = .16, t = 1.64, p = .10), and negative significant results were found for improvability (β = -.13, t = -2.00, p = .05) and participation (β = -.26, t = - 4.12, p = .00). The results for simplicity and HR accessibility were not significant. Overall, the model tested significant, F(10,280) = 12.64, p = .00, and explained 31% of the variance of acceptance.

For resistance (reversed), positive significant results were found for simplicity (β

= .18, t = 2.99, p = .00), embeddedness (β = .38, t = 5.03, p = .00), and HR coworkership (β

= .21, t = 2.28, p = .02), whereas negative significant results were found for improvability (β

= -.16, t = -2.59, p = .01) and participation (β = -.15, t = -2.63, p = .01). The results for publicity and HR accessibility were not significant. Overall, the model tested significant, F(10,280) = 10.04, p = .00, and explained 41% of the variance of resistance.

As not all levers are positively related to system adoption and the results do not

indicate any pattern along the three implementability groups (i.e., content, process, and

context), these outcomes are not in line with the hypothesized effects.

Different groups of users

Two different groups of users of the new system were included in the study (employees and managers). The two groups use different system functionalities and have different responsibilities and may therefore differ in their degree of system adoption

similarly. It was hence hypothesized that 1) different groups of users may respond differently to the introduction of a new system, and 2) that the effect sizes of the implementability levers on system adoption will be larger for managers than for employees.

System adoption. In order to test the first part of this proposition, two t-tests were executed to compare the means of acceptance and resistance (reversed), respectively, for the two groups of users. The means and standard deviations for acceptance and reversed

resistance for both user groups, and the outcomes of the t-tests are provided in Table 6. These results indicate that there are indeed differences in reactions, but for the reversed resistance dimension only. Managers scored significantly higher on resistance (reversed, t = 4.44, p

= .00), suggesting that they engage in more supportive (or less resistant behaviours) than employees do. No differences were found for acceptance.

Implementability levers. Another series of t-tests was executed to compare the mean scores of the different implementability levers for employees on the one hand and managers on the other hand. The means and standard deviations of all levers for both user groups and the outcomes of the t-tests are provided in Table 6. The results from this series of t-tests show that managers rated embeddedness, participation (marginally), HR coworkership, and HR accessibility significantly higher than employees, thereby confirming that levers are not necessarily assessed similarly by different groups of users. There were no significant differences in rates for improvability, simplicity, and publicity.

Additionally, four linear regression analyses were executed to examine the relations

between the implementability levers and the acceptance and resistance (reversed) dimensions

for employees and managers, thereby elaborating on the results that were found in the preliminary correlation analyses (see Table 4 for employee correlations and Table 5 for manager correlations). In all analyses, age, level of education, and organizational tenure were added as covariates. For an overview of the outcomes, see Table 7.

A number of results deserve extra attention. First of all, it is noteworthy that for all models a significant proportion of the variance of either acceptance or resistance (reversed) is explained by the implementability levers. Secondly, there is a clear distinction between the manager- and the employee models. Whereas the models for managers show less significant results for the implementability levers than the models for the employees, the manager models do explain a larger proportion of variance for both acceptance and resistance

(reversed). Thirdly, one implementability lever that seems especially important when a new system is implemented is embeddedness. For both managers and employees embeddedness is significantly and positively related to both acceptance and resistance (reversed). Finally, despite the proposition that all levers would positively affect system adoption, but in accordance with the results described in the previous section, negative results have been found for improvability and participation for both acceptance and resistance (reversed), be it only for employees.

The results described are partially in line with H2, in the sense that there are indeed

differences in reactions to a new system implementation when comparing managers and

employees. Moreover, the implementability levers did explain a larger proportion of variance

of both acceptance and reversed resistance for managers than for employees, indicating that

the levers do indeed affect manager system adoption to a greater extent than employee system

adoption. One unforeseen result, that is not necessarily in line with the hypothesis, however,

is that employee system adoption is related to more aspects of the implementation (i.e.,

implementability levers), in which some aspects even negatively relate to system adoption

(i.e., improvability and participation). Generally speaking, it seems that the implementability levers have a more diffused relation with system adoption for employees and a more

concentrated relation with system adoption for managers.

Social influence

An additional aim of this thesis was to explore the effects of social influence when considering system adoption as well as several implementability levers as antecedents of system adoption. It was hypothesized that social influence would moderate the relations between the levers and system adoption in some way. Neither effect size nor direction were proposed, given the exploratory nature of the hypothesis.

Again, a total of four linear regression analyses were executed for acceptance and reversed resistance, split by user group (i.e., employees and managers). In order to be able to test the moderating role of social influence, each analysis was extended with an extra step in which interaction variables were added to the model. These interaction variables were computed beforehand by multiplying the independent variables (normalized) with social influence (normalized; one interaction variable for each independent variable). As with the previous analyses in de sections above, age, level of education, and organizational tenure were added as covariates.

Instead of using the individual implementability levers as independent variables, this time the levers were grouped according to the content, process, and context dimensions. This procedure was opted for in order to make the analyses less extensive and less difficult to interpret, and was justified by the outcomes of the Factor Analysis (see Methods section or Appendix D for details) that showed that the levers rated perfectly within the expected groups (content, process, and context). Moreover, this procedure matched with the aim of the

analysis: to explore the effects and to provide some first indications of the effects that need

later be examined more thoroughly.