An empirical investigation of the service-dominant logic: the matter of office buildings

MSc Business Administration: Change Management

Faculty of Economics and Business

Master Thesis

An empirical investigation of the service-dominant

logic: the matter of office buildings

Willemijn Aukema | S2480328 |

w.m.aukema@student.rug.nl

January 2019

Supervisor: Prof. dr. ir. D.J. (David) Langley

Co-assessor: Prof. dr. A. (Albert) Boonstra

2

Abstract

Management research is experiencing a vital shift in perspective from the conventional goods-dominant logic toward the service-dominant logic. However, empirical scrutiny of service-dominant logic in itself and more specifically the environment that best facilitates value co-creation is needed. The overarching purpose of this study is to further advance and refine the understanding of the service-dominant logic through theory testing in the context of office buildings. This will be achieved through a dual-objective research design whereby measurement scales are developed for service-dominant design and value co-creation behavior and subsequently their relationship tested. As no proper scales yet exist, the development thereof precedes testing the relationship between the theoretical constructs that have been extensively discussed by scholars. The dual-contribution this paper thereby makes is first transforming the abstract theoretical concepts into measurable constructs. A subsequent contribution is made as empirical evidence is found supporting a positive relation between the constructs. Overall, this research has laid the groundwork for future research to improve the measurement scales scientific rigor and continue the pursuit toward empirical validation of service-dominant logic.

Keywords: Service-dominant design · value co-creation behavior · office context · scale development ·

3

Table of Content

A change in the institutional logic ... 7

Principles of service-dominant logic ... 8

Conceptualizing value co-creation. ... 9

The ideal: a service ecosystem. ... 10

Framing the context of a service ecosystem. ... 10

Different roles of similar actors ... 11

Method ... 13

Research context ... 13

Traditional office concept. ... 13

Open-plan office concept. ... 14

Society-inspired office concept. ... 14

Scale development ... 15

Stage 1: item generation. ... 15

Independent variable. ... 16

Dependent variable. ... 18

Participation behavior. ... 18

Citizenship behavior. ... 19

Control variables. ... 19

Stage 2: content adequacy assessment. ... 20

Stage 3: questionnaire administration. ... 21

Data collection procedure ... 23

Preliminary data analysis: Scale development ... 23

Normality. ... 24 Normality statistically. ... 24 Normality visually. ... 25 Univariate Outliers. ... 25 Multivariate Outliers. ... 25 Results ... 27

Stage 4: Factor analyses ... 27

Exploratory factor analysis. ... 27

4

Dependent variable: personal interaction. ... 28

Confirmatory factor analysis. ... 29

Dependent variable constructs. ... 29

Stage 5: Reliability and validity analysis ... 30

Testing the relationship: descriptive statistics ... 31

Hypothesis tests: hierarchical multiple regression ... 33

Discussion and Conclusion ... 36

Theoretical contributions ... 39

Managerial implications ... 40

Limitations and future research... 41

References ... 43

Appendices ... 47

Appendix 1 – Questionnaire (with modifications from content adequacy assessment) ... 47

Appendix 2 – Proportion of interjudge agreement ... 50

Appendix 3 – Normality statistically per item ... 51

Appendix 4 – Frequency distributions per item ... 52

5

Introduction

For as long as management scientists can remember goods have functioned as the fundaments of economic exchange (Vargo & Lusch, 2004). However, over the past several decades, management research is experiencing a vital shift in perspective from the conventional goods-dominant (G-D) logic toward the service-dominant (S-D) logic (Vargo & Lusch, 2017; 2008; 2004; Lusch & Nambisan, 2015; Chandler & Vargo, 2011). S-D logic builds on the foundation that the activities emanating from specialized knowledge and skills (rather than the unit-of-output, i.e. goods) represent the source of value and thus the purpose of exchange (Vargo & Lusch, 2017). Formulated differently: people don’t value objects, people value performances (Vargo & Lusch, 2008). The ideal is a service ecosystem representing a system of actors amongst whom resource integration and service exchange continuously takes place contributing to the co-creation of value (Lusch & Nambisan, 2015; Chandler & Vargo, 2011). Over the years many theoretical contributions have been made to the theory of S-D logic. Scholars throughout the world have theorized how S-D logic relates to value co-creation. However, barely any empirical research is conducted providing evidence for the theoretical relations suggested. Surprisingly, academic research fails to empirically confirm the potential of service-dominant principles and how they relate to optimal value co-creation behavior of relevant actors (Yi & Gong, 2013). Value co-creation behavior represents the behavioral manifestations of actors for the purpose of value co-creation. Following this gap in service-dominant reasoning, the research question of this study therefore reads: How do service offerings characterized by

service-dominance relate to value co-creation behavior? The overarching objective of this study is to

further advance and refine understanding of the service-dominant logic through theory testing in the context of office buildings (Vargo & Lusch, 2017).

6

of the S-D logic to practice. The initial objective therefore was to test the relationship between service-dominant design of office buildings and the value co-creation behavior of the related actors. But, despite the existence of many high quality publications, a lack of focus on empirical scrutiny has resulted in the absence of adequate measurement scales. Surprisingly, service-dominant logic scholars withhold themselves from empirical validation of the theory. Hence, the primary objective has shifted toward the development of measurement scales capturing the theoretical essence of the relevant variables. Consequently, to answer the research question, first the following sub-research questions must be studied:

How can we develop a valid and reliable measurement scale for service-dominant design? and How can we develop a valid and reliable measurement scale for value co-creation behavior?. For developing

appropriate scales, the framework of Hinkin (1995, 1997) is applied. Only when proper scales exist can the relationship between service-dominant design and how it facilitates the context for optimal value co-creation be tested (Vargo & Lusch, 2017; Yi & Gong, 2013). Thereby answering the following sub-research question: How does service-dominant design relate to value co-creation behavior of actors in the office

context? In testing this relationship, another proposition central to S-D logic is taken into account.

Theoretically seen, optimal value co-creation follows after 11 foundational premises are met. In line with Vargo and Lusch’ (2017) call for more zooming in, this research zooms in on one foundational premise. Foundational premise 6 states: ‘value is co-created by multiple actors, always including the beneficiary’, which disassociates actors from the traditional, predesigned roles and sees every actor in the service ecosystem as fulfilling the role of co-creator as well as beneficiary of that value (Vargo & Lusch, 2016). The choice for this focus follows from Vargo & Lusch’ reasoning (2016) that this premise is central to any meaningful conceptualization of S-D logic. Premise 6 is included in the analysis as control variable for the value co-creation behavior of different actor roles.

7

S-D logic to the organization context. Overall, the dual-objective approach contributes to the advancement of service-dominant logic toward a robust theory.

The change in institutional logic not only represents a prominent change for scholars but greatly affects organizations too. Fundamental changes in business models and strategies may result from managerial application of the service-dominant logic. How value is created and assessed greatly differs from the conventional wisdom. A managerial contribution is made in terms of clarification and applicability of the S-D logic. This empirical study provides practitioners with a first set of tools to put the S-D logic into practice. Managers are provided with both empirical evidence as well as measurement scales to make substantiated decisions regarding the office building design and measure the value co-creation behavior of relevant actors. This study has bridged the gap between theory and practice whereby the established groundwork enables managers to make better substantiated decisions in the office building context.

Service-dominant logic represents a promising lens to consider the value potential of resources and operations (Vargo & Lusch, 2017; 2008; 2004; Lusch & Nambisan, 2015; Chandler & Vargo, 2011). Before developing reliable and valid scales for service-dominant design and value co-creation behavior, it is required to gain a thorough understanding of the literature. After shortly exploring the theoretical background of the conventional goods-dominant logic, we zoom in on the service-dominant logic. Elaborating on value co-creation, the service ecosystem, the context and the actors will provide adequate understanding of the theory leading us to the hypothesis of this research.

Literature Review

A change in the institutional logic

8

Table 1. The elements of the goods-dominant and service-dominant logicª

G-D logic S-D logic

Primary unit of exchange Tangible goods Experience offering

Role of goods Goods as the end product Goods as the conveyor of service

Role of customer Loyalty Engagement

Determination of value Value is determined by producer. Value is co-created by all actors. Source of economic growth Competition Collaboration

ª Adapted from Vargo & Lusch (2004) complemented with other publications (Vargo & Lusch, 2017; 2016; 2008; Lusch & Nambisan, 2015; Chandler & Vargo, 2011).

of operant resources. Hereby, wealth is established through ‘the application and exchange of specialized knowledge and skills’ (Vargo & Lusch, 2004, p. 7). The change in institutional logic is a prominent change for management scholars but also for organizations and the way they conduct business. It changes the fundaments of how scholars assess the economy and the fundaments of organizations’ business models. The many papers published in high quality journals suggest that the S-D logic holds meaningful potential (Vargo & Lusch, 2016; 2004; Lusch & Nambisan, 2015; Barrett, Davidson, Prabhu & Vargo, 2015). To increase understanding and applicability of this new perspective, a thorough review of this fundamentally different logic is required.

Principles of service-dominant logic

9

takes place, and more specifically the context that best facilitates value co-creation. Lastly, foundational premise 6 is discussed along with the different actor roles that prevail in the process of value co-creation.

Conceptualizing value co-creation. Under the conventional G-D logic, value is created when

people or organizations (through the use of resources) transform and integrate resources by (ex) changing its form, time, place, and possession (Lusch & Nambisan, 2015). They thereby enhance a source of utility which is labeled as value-in-exchange. Following the S-D logic, value is not embedded within a firm’s offering but rather value occurs when the offering is useful to another actor. This is better known as

value-in-use (Vargo & Lusch, 2004; Lusch & Nambisan, 2015; Chandler & Vargo, 2011). Following this

formulation, value co-creation is then the result of interaction, i.e. a mental, physical or virtual contact. This describes how an actor creates opportunities to engage with other actors and they cohesively advance the potential for value creation (Gronroos & Voima, 2013). In that, value is no longer a unilateral provision of one actor to a utilizer of that value. Rather, it should be approached from an activity-based view with an emphasis on resource-integration behavior of all actors. As with the generic framework, the abstract definition of value co-creation lacks applicability for empirical research. The priority of this research lies at conceptualizing service-dominant variables, and therefore the focus of this paper will be on the micro-level construct: value co-creation behavior (VCB). Following Vargo and Lusch’ (2017) argumentation of alternately zooming in and out, the VCB construct will allow zooming in on specific actors and their behavior for the purpose of value co-creation (Yi & Gong, 2013). Essentially, value co-creation behavior refers to actors’ contribution to and participation in the process of value co-creation. Particularly, VCB offers an in-depth conceptualization of the different behaviors that may occur when value co-creating. Hence, VCB is defined as the behavioral manifestations of actors that result in participation in and further

progressing of activities that advance the potential for value creation in a service ecosystem context (Lusch

10

be applicable for every actor in general. Before going into depth on the actor dimension, it is further elaborated how the network of interactions between actors and resources functions through the exploration of the service ecosystem concept.

The ideal: a service ecosystem. Rather than perceiving the interactions as conventional networks

of resources, people, and product flows, S-D logic proposes the idea of a service ecosystem (Vargo & Lusch, 2017). This theoretical ideal embodies a self-contained, self-adjusting system of actors amongst which service exchange and resource integration continuously takes place contributing to the co-creation of value (Frow, McColl-Kennedy, Hilton, Davidson, Payne and Brozovic, 2014; Lusch & Nambisan, 2015; Chandler & Vargo, 2011). In the network literature, ecosystems are described as entities (organizations and individuals) that coevolve their capabilities and roles to improve effectiveness and survive (Lusch & Nambisan, 2015). They are capable of adapting and evolving in response to environmental changes but when changes are too great ecosystems may collapse (Frow et al., 2014). However, as most S-D literature has focused on the meta-level of theoretical abstraction, the idea of a service ecosystem hasn’t yet been subjected to any micro-level empirical verification. Consequently, the concept indirectly suggests a certain imperishable nature or at least disregards any susceptibility of the system (Vargo & Lusch, 2017). It assumes the service ecosystems to be capable of spontaneously sensing and responding to changes (Chandler & Vargo, 2011). This system view differs from the more conventional network view in that each act of resource integration changes the nature of the service ecosystem to some degree and thus the context for the next instance (Frow et al., 2014). For this line of reasoning to hold, S-D logic acknowledges the need for flexibility (Edvardsson, Tronvoll & Gruber, 2011; Lusch & Nambisan, 2015; Vargo & Lusch, 2016). Flexibility in how the ecosystem is organized and the context is structured. Yet, the shortage of empirical confirmation in S-D logic makes that theoretical dimensions as such await to be tested. According to Vargo & Lusch (2017), when researching value co-creation, the service ecosystem forms the unit of analysis. However, in that same paper they call for more micro-level research lending itself to ‘direct testing, verification and application’ (2017, p. 50). As mentioned before, the context plays a significant role in a service ecosystem (Lusch & Nambisan, 2015; Chandler & Vargo, 2011; Frow et al., 2014). Hence, the focus of this paper will be on the physical context and how that relates to value co-creation behavior.

Framing the context of a service ecosystem. The notion of context is important as it enables us to

11

contexts either allow resources to be drawn upon or act as deterrents. This would mean that resources (including actors) have a bigger potential to contribute to value co-creation in one context over the other. However, the S-D logic only discusses the context dimension in an abstract manner and fails to elaborate on characteristics of an ideal context for value co-creation behavior. Academics have succeeded in creating awareness that context is important to consider but fail to specifically evaluate which aspects are important to consider. Nonetheless, based on the abstract theorizing and the general foundational premises, certain assumptions can be made about which characteristics should be important when discussing a S-D context. For a physical context to be conducive to value co-creation behavior, the following characteristics are considered (Lusch & Nambisan, 2015, Chandler & Vargo, 2011; Vargo & Lusch, 2004). The physical context must provide access to all relevant resources potentially needed for service exchange to occur. The more constrained the physical context, the more constrained the service exchange amongst actors. Actors are continuously letting go of old and establishing new connections with other actors and resources (Lusch & Nambisan, 2015). This means that contexts are always in flux which requires a dynamic character. Although seemingly obvious, it must be understood that the physical context cannot deliver value but can offer a value proposition as invitation for actors to engage with the resources presented (Lusch & Nambisan, 2015). When the contexts facilitates the structure for rapid exchange of resources amongst actors, value co-creation experiences arise. As such, the actors must experience the physical context as flexible in that it allows them to capitalize on different actors’ expertise and skills and thereby create synergies. This flexible character enables actors with different roles in the system to all participate in value co-creation behavior.

Different roles of similar actors. One of the most central propositions according to Vargo & Lusch

12

ecosystem as co-creator and beneficiary of value. Although, never has it been tested if every actor in a context characterized by S-D logic actually behaves as such. Theoretically, in a service ecosystem context it is assumed that every actor shows value co-creation behavior. This generic actor designation should however not be confused with the proposition that all actors are identical or show the same level of value co-creation behavior (Vargo & Lusch, 2016; Yi & Gong, 2013).

Every actor’s specialization makes that they behave according to a certain role. Despite that every actor in the ecosystem has a general potential to co-create and experience value they do so coming from different positions. S-D logic broadly describe three different roles an actor can have depending on the nature of the service exchange and the context in which this takes place (Lusch & Nambisan, 2015). These encompass the role of ideator, designer and intermediary. According to S-D logic, every actor, may it be the CEO or the company cafeteria-employee, is of importance. The ideator role describes the actors which bring in knowledge to the firm and integrate that with existing offerings. The designer role covers the actors that carry the capability to mix and match existing components to configure and stimulate service exchange. These actors oversee what resources are present in an organization and based on that decide how the context should be configured. Lastly the intermediary role describes the actors that facilitate and orchestrate the service ecosystem thereby assisting the flow of exchange. Even though different actor roles are discussed in literature (Lusch & Nambisan, 2015), Vargo & Lusch (2016) make no specific distinction in how actors in a service ecosystem behave. They all contribute to the value co-creation and thus we hypothesize:

Hypothesis 1: Service offerings that more closely follow service-dominant principles positively relate to value co-creation behavior of actors.

13

Method

Research context

Assessing the physical context for service-dominant logic was found both relevant and important as discussed in the literature review. For this empirical research the focus was laid on the office context in the Netherlands. The choice for the office environment was in line with Vargo & Lusch’ call for more micro-level research (2017). The office often functions as the context for business research (Gibson, 2003; Appel-Meulenbroek, Groenen & Janssen, 2011; Zalesny & Farace, 1987). Despite trends of telecommuting, a vast amount of business still takes place in office buildings (Kreiner, Hollensbe & Sheep, 2009; JLL report 2017; 2018). Organizational literature has demonstrated that different office designs stimulate different behaviors (Zalesny & Farace, 1987; Gibson, 2003; Appel-Meulenbroek et al., 2011). Not only does research support the office as a suitable context for management research. Also, international research statistics show that the rental growth rate for offices has increased every year over the past ten years and is expected to continue growing (JLL report, 2018). Specifically, the Netherlands formed a suitable context as the value of Dutch office buildings has experienced the largest growth in comparison to other leading European capital cities (JLL report, 2018; Van Bockxmeer, 2018). This indicated that the corporate real estate sector is thriving and supported the choice to conduct research in this context. Additionally, the Dutch context proved suitable for conducting management research as being the third most chosen non-north American country for data collection in the ‘Academy of Management Journal’-publications (Kirkman & Law, 2005). In the dictionary, the office is defined as ‘a room used as a place for commercial, professional, or

bureaucratic work’ or ‘the local center of business used to provide a particular service’ (Oxford Dictionary,

2018). Aspects of these definitions correspond to the S-D logic in that the office frames the previously discussed service exchange into a physical context. Narrowing down the abstract concept of ‘context’ (p. 11 & 12) to the office context, allowed us to place parameters on our empirical analysis which strengthened our findings. The office context has lent itself for direct testing, verification and application. Fluctuating occupancy and rental rates, increasing pressures for efficient resource usage and overall changes in the way we do business have led to the rise or perseverance of many different office concepts (Vargo & Lusch, 2004; JLL report, 2018; 2017; Van Bockxmeer, 2018). The challenging principles underlying these concepts further motivate research in this context. Ambiguity prevails about how different office concepts facilitate value co-creation behavior. Consequently, it is most logical to shortly consider the different concepts that are currently dominant in the economy. Table 2 presents the generalized characteristics of each concept retrieved from practitioner-focused publications and facilities journals.

Traditional office concept. Originating from a hierarchical background, the most traditional office

14

and title. Often job title goes along with a certain amount of square meters of which you have the right to claim. Every actor in this office design is provided with a personal workspace which fosters physical and personal boundaries. It results in high levels of perceived privacy and control over work. The furnish and design symbolically conveys information about organizational status which regulates the interpersonal behavior taking place across the hierarchical levels.

Table 2. The different characteristics of the prevailing office concepts ª

Traditional Open-plan Society inspired

Hierarchical Equality Society

Closed layout Open layout Activity-based layout

Static Flexible Fluid

Physical boundaries Physical proximity Physical flexibility Scarce digital engagement Little digital engagement Virtual platform Monetary exchange Monetary and social exchange Social capital exchange Lengthy financial contracts Financial contracts Social contracts

e.g. law firms, automotive

industry, public sector e.g. FMCG sector, retail industry e.g. Seats2Meet

ª Appel-Meulenbroek et al., 2011; Gibson, 2003; De Been & Beijer, 2014; Zalesny & Farace, 1987; JLL report 2017; 2018.

Open-plan office concept. The literal application of open-office plans has existed ever since the

industrial era. However, the open-plan office concept discussed in this research covers the transition from the traditional design to a less static one. The underlying drivers are increased flexibility and communication while lowering operating costs simultaneously. Despite inconsistent support for these claims, increased physical proximity is thought to enhance working conditions, trust and facilitate a favorable organizational climate with strong, positive interpersonal relations (Zalesny & Farace, 1987). Nevertheless, the ideals are not always rewarding as research reveals examples where work satisfaction, motivation and involvement decreased (De Been & Beijer, 2014; Kreiner et al., 2009; Zalesny & Farace, 1987). For some stakeholders the transformation toward an open-plan office results in loss of status and diminished authority while others interpret it as increased democratic distribution of symbols.

Society-inspired office concept. Still very new, the final office concept considered is one

15

exchange of social capital more than monetary assets. Of the three concepts this design is by far the most flexible as it consists of a network of office locations owned by different stakeholders rather than one owner holding a portfolio of office real estate. Every owner in abundance of office space can join this movement. As opposed to the other two concepts, the society-inspired concept isn’t yet visible on great scale. The organization Seats2Meet is one of the driving forces behind this concept and represents one of the data sources for this research. As the specifics of the research context have been discussed, the established background now permits scale development for service-dominant design and value co-creation behavior in the office context.

Scale development

To adequately measure the abstract concepts under examination it was necessary to follow principles suitable for organizational research. Only then could both validity and reliability of the results be guaranteed and the data appropriately be interpreted. The Hinkin framework (1995; 1997) satisfied these criteria. Originally founded in the psychological research domain, the framework describes multiple stages that ensure valid and reliable development of construct scales. Wherever the framework lacked applicability for the scope and context of this research, other literature on scale development has been exploited to complement the Hinkin framework (De Jong & Elfring, 2010; Rich, LePine & Crawford, 2010; Marsh, Hau, Balla & Grayson, 1998; Rust & Cooil, 1994; Fleiss & Cohen, 1973; Landis & Koch, 1977; Schriesheim, Powers, Scandura, Gardiner & Lankau, 1993). The initial step conducted in the scale development process was item generation for each construct. Subsequently, multiple (statistical) assessments both before and after factor analyses were performed to ensure reliability and validity of the scales.

Stage 1: item generation. Two approaches lead to item development, deductive (logical

partitioning) and inductive (grouping) generation. The deductive approach requires thorough understanding and defining of the concept and its constructs through literature, which guides the

16

2010; Marsh et al., 1998). Taking into account possible deletion of items, at least five items were generated when developing a new construct or transforming an existing construct. Those constructs left untouched sufficed with the existing three (or more) validated items. It was important to limit the length of the questionnaire so as not to tire out the respondents which might have resulted in premature

withdrawal. An understanding of the relevant literature has resulted in the following conceptualizations, constructs and items.

Independent variable. Extensive literature review has brought us to the following definition of

service-dominant design: a contextual design representing service-dominant principles in its functional

features and relational aspects (Lusch & Nambisan, 2015; Chandler & Vargo, 2011; Frow et al., 2014).

Service-dominant design was thereby divided into two constructs: the functional design construct (Chandler & Vargo, 2011; Zalesny & Farace, 1987) and the relational design construct (Vargo & Lusch, 2016; Lusch & Nambisan, 2015). Before going into depth on both constructs, one main characteristic to determine the extent of service-dominant design is discussed: flexibility (Edvardsson et al., 2011; Lusch & Nambisan, 2015; Vargo & Lusch, 2016). Since every actor is unremittingly exchanging and integrating resources, change takes place continuously affecting what is demanded from the context (Frow et al., 2014). This calls for a certain fluidity. Narrowing it down to the physical context of an office building, this implies that office concepts that are flexible would better facilitate value co-creation over more static concepts. A flexible design allows different configurations of actors to participate in the office, adapt to new stimuli, and create new value opportunities (Lusch & Nambisan, 2015). The broader objective of a flexible design is to ensure that the overall configuration is easily adaptable to anticipate necessary changes. Moreover, that it facilitates all the possible, relevant activities that lead to value co-creation. This because value lies in macro-activities (e.g. meetings, presentations, operational processes) but according to S-D logic also in micro-activities (e.g. entering the building, making a phone call, having lunch with colleagues or even the cleaning of the workplaces) occurring in an organization (Gibson, 2003). This general flexibility principle has guided the item generation of the constructs.

The functional construct was defined as follows: the tacit functionalities of the office context

supporting all relevant activities and interaction among actors in the office. Functional flexibility means

that the office is capable of supporting whatever activity needed for different actors to co-create value (Edvardsson et al., 2011). The functional features form a crucial construct as they frame the behavior, interaction and activities taking place in the organization (Chandler & Vargo, 2011; Zalesny & Farace, 1987). Office designs that remove traditional, functional barriers which hinder the flow of work thereby maximize interaction and communication among actors which is in accordance with S-D logic. This was measured by the items: ‘the office is functionally designed in such a way that it facilitates any relevant

17

interact freely, the office layout should be open and free in usage (Vargo & Lusch, 2017; Zalesny & Farace, 1987). Physical proximity between actors in the office provides opportunity to develop interpersonal relations. This results in individuals having a more complete view of their contribution to the general flow of work, a crucial aspect in the theory of S-D logic. This was measured by the following items: ‘the office

is an open environment’, ‘the way the office is designed stimulates interaction between those others present’

and ‘all workspaces are free to be used by whomever wants to do so’.

The relational construct describes how the office building as a resource is not contracted in a static manner but rather expands and contracts in response to human actions (Constantin & Lusch, 1994; Chandler & Vargo, 2011). Offices designed according to S-D principles bring flexibility to the connection between any actor and the office. With connection is meant: the extent to which the office design gives the actor implicit space to make use of the resource as he or she behooves. The focus is not so much on the actual relationship between the office and the actor but more on if the design facilitates service-dominant (i.e., flexible) usage. As such the relational construct was defined as: the implicit impression the office context

gives its users in facilitating them to make flexible usage of the office and its services. An office design

giving any actor the impression that it can be used in a flexible manner suitable for the actor’s need is service-dominant. This was measured through five items capturing flexibility: ‘the office is designed such

that I am flexible in …: when, where, what, how and with whom actors work (Lusch & Nambisan, 2015;

Vargo & Lusch, 2016). Where the functional construct items cover the tacit aspects of an office, the relational construct conveys the implicit, perceived connection between each actor and the office environment (appendix 1).

18

Dependent variable. The dependent variable is value co-creation behavior based on Yi & Gong’s

(2013) VCB scale. As mentioned before, the value co-creation behavior concept is defined as: the

behavioral manifestations of actors that result in participation in and further progressing of activities that advance the potential for value creation in a service ecosystem context (Lusch & Nambisan, 2015; Yi &

Gong, 2013; Laud & Karpen, 2017; Gronroos & Voima, 2013). The original scale comprises of two dimensions (participation and citizenship behavior) each existing out of four constructs. Information seeking, information sharing, responsible behavior, and personal interaction represent participation behavior. Feedback, advocacy, helping, and tolerance cover the citizenship behavior. However, this original scale is focused on customers solely which is a conventional role division that is abandoned in the most S-D logic papers (Vargo & Lusch, 2004; 2008; 2016; 2017; Lusch & Nambisan, 2015). Thus, applicability for this research required some adaptation to measure the VCB of different actors in the office building context. This means that some constructs have gone through little reformulation, some required serious transformation and others have been deleted entirely. Those constructs that went through minor formulation changes only, sufficed with three validated items while the constructs that went through total reformulation required at least four items. For this specific context, the responsible behavior construct was irrelevant as this describes customers recognizing their duties as partial employee (Yi & Gong, 2013). In the office environment it is assumed that every actor recognizes certain responsibilities as they are all employees of an organization and therefore this research omits this construct. Also, the tolerance construct was deleted as this is based on the traditional customer-produce role division which does not correspond with the papers’ theorizing.

Participation behavior. Information increases actors’ understanding and allows them to integrate

in the activities in the office. When scoring high on information seeking, an actor is better able to have influence in the process of value co-creation. Yi & Gong’s (2013) items capture the essence of this construct and required only reformulation to present tense (appendix 1). Reformulation to present tense was necessary for all other items as well. To successfully co-create, actors should also provide information that can be used in the process of value co-creation. The original items for information sharing are designed around the customer role discussed earlier on and therefore required more deductive adaptations. The following items captured the key of information sharing behavior: ‘I explain to others what I am working on’, ‘I give

others information on what I am working on’, ‘I make an effort answering the questions of others’ and ‘I share any information which is helpful to others’. All the previously mentioned items were preceded by ‘In the office building…’ to ensure correct understanding of the context. The final construct measuring

19

others’, ‘I choose to interact with others’, ‘I search for the best ways to interact and ‘I pay attention to the interactions of others’.

Citizenship behavior. This dimension of value co-creation behavior captures the manifestations that

provide extraordinary value in the process of co-creation (Yi & Gong, 2013). It conveys the extra-role behavior the voluntary gestures of actors that further advance successful value co-creation. The primary construct, feedback, is a construct that improves value co-creation but is not fundamental to VCB. Feedback regards solicited and unsolicited information that actors provide to each other helping to improve the behavior and interactions of others. Here, the original items required little adaptation to be applicable for this research. The same applied for the construct advocacy which directly measured the extent to which actors recommend the business, in this case the office, to others. Advocacy indicates allegiance to the office beyond personal interest. Actors showing helping behavior extend empathy to others and thereby increase interaction (Yi & Gong, 2013). Once again, the original items were very suitable. The last construct of citizenship behavior is helping. Helping refers to spontaneous behavior aimed at assisting others. The items appropriately capture the construct and have therefore not been transformed. Respondents valued the statements according to a 7-point Likert-scale where 1 = never, 2 = less than once a month, 3 = once a month, 4 = once a week, 5 = daily, 6 = more than once a day, and 7 = hourly.

Control variables. The scale development process required data collection in order to run statistical

20

included as a control variable (appendix 1). Through controlling for the different roles, it can be analyzed if an actor’s role has a significant, relative effect on the independent and dependent variables.

Besides role type, other control variables were taken into account for this research. Many years of work experience in an office might result in a rusty, possibly more traditional, perception and behavior. Also, firm size potentially influences an organization’s ability to change in a flexible manner (Vaccaro, Jansen, van den Bosch & Volberda, 2010). Therefore, internal validity was enhanced by including gender, age, office experience (in terms of length) and organization size of which the latter two are measured in a nominal manner. Office experience is categorized as minor office experience (less than 3 months), little experience (3 months – 3 years), moderate experience (3 – 15 years) or rich experience (over 15 years). Organization size is categorized between micro (1-10 employees), small (11-50 employees), medium (51-250 employees) and large (more than (51-250 employees) (European Commission, 2003). Including these control variables mathematically accounts for the effect of these variables from the relationship between the independent and dependent variable. For the statistical analysis, dichotomous, dummy variables were created for office experience, organization size and actor role. When creating dummy variables this must be done for k – 1, as one category is represented by the 0 of the other dummy variables. Every last category of our control categories was covered by the 0 and functioned as the reference category (rich office experience, large organization size and facilities role). The results of the relations between the dummy variable and the dependent variables must then be interpreted relative to the reference category.

Lastly, solely results were included of respondents working in an office in the Netherlands. The following step assured content adequacy of the items generated prior to the distribution of the questionnaire. This has enhanced construct validity as it provides the opportunity to pre-test the concepts.

Stage 2: content adequacy assessment. Several methods exist to conduct content adequacy

21

The proportion of interjudge agreement of the content adequacy assessment was derived from 130 pairwise agreements out of a total of 192 possible agreements resulting in .68 (appendix 2). Rust and Cooil (1994) provide convenient reference tables enabling researchers to directly compute their PRL rather than having to work with the fairly cumbersome formulas. As the tables are applicable for content assessments examining adequacy for two to five categories, a challenge was faced in the assessment of eight categories. Following the formula, the more categories are included, the higher the PRL score for a similar number of judges and proportion of interjudge agreement. Looking at table 6 in the paper of Rust & Cooil (1994, p. 12) for five categories, it can be seen that an interjudge agreement of .68 and four judges gives a PRL reliability measure of .96. This is regarded as a rather confident measure for the judgements of the four judges. But, as eight rather than five categories were assessed, another statistical measure was also considered for assessing the reliability of agreement, namely Fleiss’ kappa (Fleiss & Cohen, 1973; Landis & Koch, 1977). The Fleiss kappa is suitable for the measure of agreement between more than two judges and calculates the degree of agreement over that what would be expected by chance. The Fleiss kappa is .60 (SE = .07), which according to Landis and Koch (1977) corresponds to moderate agreement. Following both measures, the statistical reliability of interjudge agreement was concluded to be sufficient. Though, it has appeared that items tap into constructs other than the ones intended. Five items correspond both to the intended construct as to an unintended construct. As removal of items jeopardizes a sufficient number of items per construct, the formulation of the dubious items was critically reassessed and where necessary minor modifications were made (Schriesheim et al., 1993) as can be seen in appendix 1. Item 17 which intended to measure information sharing behavior through: ‘In the office building, I make an effort

answering the questions of others’ was categorized under helping behavior by three out of four judges.

Reformulation resulted in ‘In the office building, I provide necessary information so that others can perform

their jobs’. Also, item 31 measuring helping behavior: ‘In the office building, I teach others to use the office correctly’ was categorized under information sharing behavior twice, feedback behavior once and once

under helping behavior as intended. As the original formulation of item 17 better correlated with the helping construct item 31 was modified into: ‘In the office building, I make an effort helping others’. This technique for content adequacy assessment does not guarantee a content valid scale but does ensure that the items represent a sound measure of the concepts under examination. It reduces the need for scale modification later on in the scale development process.

Stage 3: questionnaire administration. Under questionnaire administration, Hinkin (1995, 1997)

22

turn assures that the factor loadings accurately reflect the true population values. Therefore, the bigger the sample size, the bigger the likelihood of attaining statistical significance. One of Hinkin’s articles (1995) describes how the sample size must vary between a 1:4 and 1:10 ratio for items to responses, for the 32 items of this research this would mean between 128 and 320 responses. His other article (1997) suggests that 100 observations is sufficient for confirmatory factor analysis while exploratory requires 150 responses. As both exploratory and confirmatory factor analyses were conducted, the aim was to collect 150 responses at the least. Before continuing with the data collection procedure the finalized items and corresponding abbreviations are summarized in table 3.

Table 3. Service-dominant design and value co-creation behavior items Abbreviation Item

Functional design

IVFD1 The office is functionally designed in such a way that it facilitates any relevant activity needed.

IVFD2 The majority of the office space can be used for multiple purposes.

IVFD3 The office design is an open environment.

IVFD4 The way the office is designed stimulates contact between those others present.

IVFD5 All workspaces are free to be used by whomever wants to do so.

Relational design

IVRD1 The office is designed such that I am flexible in when I work.

IVRD2 The office is designed such that I am flexible in where I work.

IVRD3 The office is designed such that I am flexible in what work I conduct.

IVRD4 The office is designed such that I am flexible in how I work.

IVRD5 The office is designed such that I am flexible in who I work with.

Information seeking

DVPBISE1 In the office building, I ask others for information.

DVPBISE2 In the office building, I search for information.

DVPBISE3 In the office building, I pay attention to others.

Information sharing

DVPBISH1 In the office building, I explain to others what I am working on.

DVPBISH2 In the office building, I give others information on what I am working on.

DVPBISH3 In the office building, I provide necessary information so that others can perform their jobs.

DVPBISH4 In the office building, I share any information which is helpful to others.

Personal interaction

DVPBPI1 In the office building, I am open to interact with others.

DVPBPI2 In the office building, I spend time working with others.

DVPBPI3 In the office building, I choose to interact with others.

DVPBPI4 In the office building, I search for the best ways to interact with others.

DVPBPI5 In the office building, I pay attention to the interactions of others.

Feedback

DVCBF1 In the office building, if I have a useful idea, I let others know.

DVCBF2 In the office building, when I receive help from others, I comment about it.

23

The subsequent steps of the scale development process will be discussed under the ‘Results’ section as these require (preliminary) statistical analysis of the collected data. Before doing so, the data collection procedure is described and a preliminary analysis is conducted.

Data collection procedure

The developed scales were evaluated and tested through cross-sectional survey data collection. When the scales are proven valid, the relationships between the variables can be tested making use of that same data. Survey research is commonly applied in the field of management (Scandura & Williams, 2000; Pinsonneault & Kraemer, 1993; Hinkin, 1995) as it enhances external validity, i.e. generalizability of the results across the population of interest. The population of interest were individuals who have working experience in an office located in the Netherlands. To ensure variation in responses and a sufficient number of responses for statistical analysis, eight office locations across varying domains (office community Seats2Meet, consultancy, telecommunication, automotive) were approached and willing to distribute the survey. Moreover, personal network and social media (LinkedIn, Facebook and WhatsApp) allowed for further distribution among relevant respondents. The questionnaire was distributed through a hyperlink redirecting to the digital questionnaire and paper versions. An exact estimation of the responses rate lacks as the choice has been made to make use of multiple communication channels (which do not all permit tracking) to ensure sufficient responses. Lastly, double responses from the same respondent were prevented through a functionality of the survey tool (Qualtrics). A total of 295 finalized responses was collected. After deleting unfitting responses (i.e. concerning foreign office experience) and respondents who answered to have no current or previous work experience, a total of 276 responses remained.

Preliminary data analysis: Scale development

Preliminary data analysis ensures validity of subsequent rigorous statistical analyses (Hinkin, 1997; Mat Roni, 2014). First, a check for monotone responses (i.e. that have zero variance) revealed that all responses contained variance. Only finalized responses were included which assured that no values are missing. Respondents’ average age was 34 years (SD = 10.87) and 51% were female. Not surprisingly, the majority of the respondents filled in the survey from a ideator (i.e. regular user) perspective (78%), followed by the designer (general management) perspective (13%) and the intermediary (facilities role) perspective (9%).

Advocacy

DVCBA1 I say positive things about the office to others.

DVCBA2 I say positive things about the office to others.

DVCBA3 I encourage others to use the office.

Helping

DVCBH1 In the office building, I assist others if they need my help.

DVCBH2 In the office building, I help others if they seem to have problems.

DVCBH3 In the office building, I make an effort helping others.

24

This was to be expected as the ratio regular users to management or facilities employees is conventionally high. Regarding office experience, the minority of the respondents stated to have minor office experience (< 3 months; 10%), almost a quarter to have either little experience (3 months – 3 years; 28%) or rich experience (>15 years; 22%) whilst the majority had moderate experience (3 – 15 years; 40%). A vast majority specified to work in a large organization (>250 employees; 63%) whilst the remaining worked in a small organization (11-50; 13%), micro (1-10: 13%) or medium organization (51-250; 11%). Figure 1 graphically presents the percentage of responses coming from different industries.

Figure 1. Percentage of responses per industry

Normality. Normality is one of the most critical probability statistics that needs to be assessed

before conducting further analyses (Hair, Black, Babin, Anderson, & Tatham, 2006). It assesses the random error of a sample in the relationship between independent variables and the dependent variable. The distribution of this error should be normal, i.e. approximately, symmetrically distributed. Deviations from normality render statistical accuracy. Normality was checked for in two ways: relying on statistical tests or visual inspection (LePine, Zhang, Crawford & Rich, 2016; Dineen, Noe, Shaw, Duffy & Wiethoff, 2007). Statistical tests have the advantage of making an objective judgement of normality, but are disadvantaged by sometimes not being sensitive enough to low sample sizes or overly sensitive to large sample sizes. Graphical interpretation has the advantage of allowing adequate judgement to assess normality in situations when numerical tests might be over or under sensitive, but graphical methods do lack objectivity.

Normality statistically. There are multiple statistical tests for normality that were conduct in the

predictive analytics-software SPSS 25. Considering the skewness and kurtosis, the statistics divided by its standard error results in the Z-value which should range within ±2.58 (p = .01) for the data to be considered approximately normally distributed (LePine et al., 2016; Hair et al., 2006). Following this rule of thumb the only items that showed a statistic in accordance with the range for both skewness and kurtosis were the three items for the dependent variable construct ‘advocacy’ (DVCBA1-3). All other items scored too high for skewness Z-values (appendix 3). Regarding the kurtosis Z-values, only the independent variable items (except IVFD4) were considered normal. The Kolmogorov-Smirnov and Shapiro-Wilk test were not

25

significant for any item meaning that the null hypothesis (stating the data is normally distributed) was rejected. These statistics hint toward a non-normal distribution but before concluding so, visual normality of the data was considered.

Normality visually. To complement the statistical normality analysis, the histogram, normal Q-Q

plot and boxplot of each item were interpreted (Dineen et al., 2007). Noticeable in the histograms are the strong peaks around value 6 and some smaller peaks around value 2 resulting in slightly skewed graphs (appendix 4). In line with our expectations, some people experience their office as having a strong functional and relational design while another majority ranks their office much lower on these service-dominant principles. The peaks might hint toward a structural measurement issue or that the data contains two subpopulations. The former can be disregarded as the histograms of the dependent variables are approximately normal. Considering the latter, as the nature of possible subpopulations is unclear, we argue that the histograms are sufficiently, approximately normal to continue the analysis. The Q-Q plot of some items do not entirely follow the normal distribution which might hint toward outliers. This was the case for information seeking item 3 (DVPBISE3), personal interaction item 2 (DVPBPI2), 3 (DVPBPI3) and 4 (DVPBPI4). Personal interaction item 1 (DVPBPI1) was deviating most from a normal distribution when considering the Q-Q plot. Lastly, the boxplot generated for each item separately showed those responses considered as outlier according to the statistical algorithm. A more thorough analysis of these data points strengthens the normality analysis.

Univariate Outliers. The boxplots indicated some potential outliers (Raaijmakers, Vermeulen,

Meeus & Zietsma, 2015). Appendix 5 shows the frequency of univariate outliers per response. For example, response 207 was categorized as univariate for 10 out of 32 items. Close examination of these responses, however, gave little reason to believe something is wrong with the data. Besides the outliers being relatively low scores, there was little reason to question the response being the actual opinion of the respondent. Nevertheless, before drawing definite conclusions a multivariate outlier analysis was conducted to enrich the outlier analysis. Where the univariate analysis identifies extreme values on single items, a multivariate analysis reveals a combination of unusual scores on multiple items.

Multivariate Outliers. To identify multivariate outliers the Mahalanobis Distance values were

26

never above ‘4’). What was noticeable about response 206 and 242 is the high variation within each response (ranging from ‘1’ to ‘7’). Research science has no consistent recommendation for dealing with outliers. It is difficult to determine if the outliers are actual outliers (e.g. human mistakes) or represent deviant opinions. Respondents with a deviating opinion are also part of society and their input should not be excluded from analyses. Many articles recommend removing outliers, others suggest transforming outliers into approximate normal distributions while others question the statistical algorithm to be suitable for identifying outliers (Wendler & Gröttrup, 2016; Rousseeuw & Hubert, 2011; Raaijmakers et al., 2015; Steensma & Corley, 2001; Mat Roni, 2014). As strict scientific guidelines lack, the subsequent statistical analyses (up until latent variable composition) have been conducted for four different datasets. These included the following: one including all responses (N = 276), one excluding the univariate outlier responses (N = 269), one excluding the multivariate outlier responses (N = 265) and lastly one excluding univariate as multivariate outlier responses (N = 261).

Table 4. Univariate and/or multivariate outlier responses Response label p-value MD Univariate outliers ≥ 5ª

134 .00000 207 (10) 193 .00000 241 (9) 9 .00002 243 (9) 242 .00004 242 (6) 87 .00014 108 (5) 185 .00017 128 (5) 259 .00024 206 (5) 108 .00032 40 .00055 141 .00057 206 .00068

a Outlier frequency between parentheses.

27

multivariate outliers are identified as such. However, after close examination of the dataset, there was no ground to belief that data was entered incorrectly. The final possible cause describes how the sample data deviates from exact mathematical abstractions. Data of the real world are approximations and thus never exact like mathematics. Hence, we decide to continue with the dataset including all responses (N = 276). Now that the preliminary data analysis is completed, the following chapter will discuss the results.

Results

Stage 4: Factor analyses

Factor analyses evaluate how adequate items represent what they are intended to represent. Upon completion of the data collection and prior to conducting the factor analyses, the inter-item correlation matrix for each construct of the two variables was checked. Those items correlating less than .4 with all other items may be deleted from the analysis (Hinkin, 1997). However, no such items were detected and therefore all items are suitable for the factor analyses. Further descriptive statistics will be discussed when the scale development process is finalized and latent variables are composed. Two types of factor analyses have been conducted in this research. Exploratory factor analysis (EFA) assesses the extent to which a set of items represents a particular theoretical construct. Hereby the possible relation between every item and construct is tested in an exploratory manner. This suits the scale development of the two independent variable constructs which have no foundation in existing scales (Barrick, Thurgood, Smith & Courtright, 2015; Rich et al., 2010). Also one dependent variable construct (personal interaction) was examined through EFA as it was entirely transformed compared to its origin. As the dependent variable constructs were based on an existing validated scale (Yi & Gong, 2013) and the foundation of the framework was left fairly similar, confirmatory factor analysis (CFA) was most suitable for this construct (Owens & Hekman, 2016; LePine et al., 2016; Steensma & Corley, 2000; 2001; Hinkin, 1997). The original scale was substantially changed but these changes solely conveyed generalization, context-specific and verb tense transformations. Hence, CFA was applied to confirm the strength of the relationship and assess how well the items represent the construct through a goodness-of-fit test.

Exploratory factor analysis.

Independent variable: functional and relational design. Since this research is the first to develop

a scale for service-dominant design and has no existing scales to rely on, the factor approach was exploratory. Multiple extraction methods exist when conducting exploratory factor analysis in the IBM SPSS Statistics 25 program. The objective was not to simply reduce the data but rather identify appropriate latent constructs and thus a principal axis factoring was conducted. The rotation method was oblique (direct

oblimin method with delta equals 0) since there was no scientifically validated ground to assume no

Kaiser-Meyer-28

Olkin measures (measures the amount of variance) were above .6 and if the Bartlett’s test values were significant (< .001) which was the case. The conventional Kaiser’s eigenvalue-greater-than-one rule is criticized by some for its inaccuracy (Ruscio & Roche, 2012; Courtney & Gordon, 2013). Therefore the choice was made to rely on Cattel’s Scree test for determining the number of factors to retain. Despite the potential for interjudge reliability bias, this test is more accurate (41,7%) than Kaiser’s test (0.0) and better corresponds to the scope of this research (Courtney & Gordon, 2013). Visual inspection of the scree plot suggested one factor at best or a marginal two-factor solution (Barrick et al., 2015). Given the conceptual overlap between the functional and relational constructs, loading of the items onto the other factor was to be expected and suggests some evidence for convergent validity. Following the pattern matrix, two factors with an acceptable explained variance (52%) are retained as each item loads onto one factor with more than .4 without any cross-loading (Lattin, Carroll & Green, 2003; Hinkin, 1997; Rich et al., 2010). The factor loadings can be found in table 5 below. The table shows that two out of five functional design items load onto the relational design construct. When consulting the items IVFD1 (‘the office is functionally designed

in such a way that it facilitates any relevant activity needed’) and IVFD2 (‘the majority of the office space can be used for multiple purposes’) some conceptual overlap is identified. It could be argued that both items

also reflect a relational aspect between the functional design and office use which could explain the factor loading onto the relational rather than the functional construct. As the threshold of three items is still met for the functional design construct the latent variables were composed based on highest factor loadings for each item meaning that IVFD1 and IVFD2 are composed under the relational design construct.

Table 5. Factor loadings independent variable ª Item Relational design Functional design IVRD4 .855 IVRD3 .845 IVRD1 .645 IVRD5 .632 IVFD1 .514 IVRD2 .469 IVFD2 .437 IVFD3 .821 IVFD4 .611 IVFD5 .504

ª Significance < .001. N = 276, rotation converged in 8 iterations.

Dependent variable: personal interaction. The dependent construct ‘personal interaction’ was

29

described above, a principal axis factory was conducted with oblique (direct oblimin method) rotation method. Here too were the KMO and Bartlett statistics satisfactory. The EFA results are presented in table 6 show that all items load onto one construct with values greater than .74 (explained variance = 64%). As this corresponds to statistical standards the construct was included in the CFA with the other dependent variable constructs.

Table 6. Factor loadings dependent variable ‘Personal Interaction’ ª

ª Significance < .001. N = 276, rotation converged in 6 iterations.

Confirmatory factor analysis.

Dependent variable constructs. To ensure construct distinctiveness among the six dependent variable

constructs, confirmatory factor analysis using IBM SPSS Amos 25 program were conducted (Owens & Hekman, 2016). The program runs a maximum likelihood solution on a specified independence model (an a priori sample variance-covariance matrix) in comparison with the observed (default) model. The overall chi-square was statistically significant (X 2 _{= 443,15; df = 194; p < .01) which is undesirable as the null}

hypothesis (stating the observed model fits the data) is then rejected. Despite its insignificance, the chi-square does fit within the conventional range of three times the degrees of freedom which is appropriate. Moreover, as the chi-square absolute model fit is sensitive to sample size and non-normality it is relevant to consider other descriptive fit statistics too (Lattin et al., 2003; Hinkin, 1997). Considering the root mean square error of approximation (RMSEA = .068), the adjusted goodness of fit index (AGFI = .829) and the comparative fit index (CFI = .949) the model indicates a significantly better fit than the alternative model. It can therefore be claimed that the observed model is substantially less false than the baseline model, i.e. the independence model (LePine et al., 2016; Steensma & Corley, 2001; Patel et al., 2013; Hinkin, 1995; 1997). Table 7 shows the relative contribution of each item to its corresponding construct with the lowest item valued .652. In light of the significant chi-square absolute model fit, the developed model consisting of six DV constructs may be rejected. However, as the parameter estimates are statistically significant and the descriptive fit statistics are satisfactory, the model is sufficiently appropriate to compose latent variables and continue the statistical analyses.

Item Personal Interaction DVPBPI3 .890

30 Table 7. Standard regression weights ª

Item Estimate DVPBISE1 .785 DVPBISE2 .646 DVPBISE3 .661 DVPBISH1 .858 DVPBISH2 .886 DVPBISH3 .771 DVPBISH4 .729 DVPBPI1 .747 DVPBPI2 .803 DVPBPI3 .878 DVPBPI4 .785 DVPBPI5 .900 DVCBF1 .829 DVCBF2 .873 DVCBF3 .891 DVCBA1 .840 DVCBA2 .971 DVCBA3 .896 DVCBH1 .888 DVCBH2 .934 DVCBH3 .930 DVCBH4 .890 ª Significance < .001. N = 276.

Stage 5: Reliability and validity analysis

service-31

dominant logic field is insufficiently advanced. Discriminant validity was however considered through the average variance extracted (AVE). Listed in table 8, the AVEs are all except one above the .50 threshold and thus indicate discriminant validity (Steensma & Corley, 2000; Patel et al., 2013).

Table 8. Reliability and validity analysis

Construct Cronbach's alpha Average variance explained

1 Functional design .77 .43 2 Relational design .87 .53 3 Information seeking .73 .51 4 Information sharing .88 .66 5 Personal interaction .90 .68 6 Feedback .90 .75 7 Advocacy .93 .82 8 Helping .95 .83

This means that the degree to which the measures do not correlate with measures they theoretically should not correlate with, is satisfactory. Except for the functional design construct (AVE = .43), which now consists of three items as the other two items (IVFD1 & IVFD2) initially theorized under the functional construct had stronger loadings on the relational design construct. The choices made for the latent variable composition may have resulted in a lower AVE by the construct ‘losing’ two items it was intended to hold. Hence, another technique, whereby a comparison was made between the AVE and the shared variance (i.e. squared correlation estimate) between each pair of constructs was conducted. This proved the AVE greater than the shared variance for all constructs demonstrating sufficient discriminant validity for all eight constructs (Steensma & Corley, 2000; Hair et al., 2006).

Having followed all the possible stages of the Hinkin framework, eight substantive scales have now been developed, of which two independent constructs (functional and relational design) and six dependent constructs (information seeking, information sharing, personal interaction, feedback, advocacy and helping). Now, the focus turns to the second objective of this research paper, testing the theoretical relationships between generated constructs.

Testing the relationship: descriptive statistics

32 Table 9. Descriptive statistics and bivariate correlations ª

ª N = 276.

* _{Correlation is significant at the .05 level (2-tailed).} ** _{Correlation is significant at the .01 level (2-taile}

33

included as a dummy variable. So for office experience, relative to a rich office experience (more than 15 years), for organization size to a large organization (more than 250 employees) and for actor role relative to the facilities role. For example, the designer (general management) role has a relatively higher correlation than the intermediary (facilities) role with the variables relational design and advocacy behavior (p < .01).

When reviewing the other correlations, it can be perceived that quite some correlations were found to be significant (table 9). For the ease of reading, four quadrants were separated. One showing the correlation amongst the control variables, one amongst all variables and one amongst the independent (IV) and dependent (DV) variables. Noticeable in the second quadrant is that except for one correlation, none of the estimates between office experience and the IVs and DVs correlate significantly. This might contradict our theorizing that more office experience has an influence on perception of service-dominant design and value co-creation behavior. In the correlation amongst the IVs and DVs, all estimates are positive and almost all significant. Only information seeking does not significantly correlate with functional design and advocacy behavior. The strong correlation amongst the IVs and amongst the DVs was to be expected due to their conceptual overlap (Godart et al., 2015). When independent variables significantly correlate with each other this might indicate multicollinearity (King et al., 2011; Patel et al., 2013; Owens & Hekman, 2016). Multicollinearity indicates that there are very high correlations between the independent variables. When multicollinearity is present in the data, the confidence intervals of the coefficients might become wide and the statistics small. This could disturb the subsequent analysis interpretations checks for multicollinearity (VIF < 10, Tolerance > .1) indicate that this was not present and that the results are robust (LePine et al., 2016; Owens & Hekman, 2016).

Overall, the correlations between the IVs and DVs are difficult to interpret. As the Pearson correlation does not take into account which variables are the IVs and which is the DV, the direction of the relationship can’t be interpreted. The hierarchical multiple regression provides results that tell more about the direction of the relation between service dominant office design and value co-creation behavior and the relative differences between the different actor roles.

Hypothesis tests: hierarchical multiple regression

The hypothesis of this research reads: Service offerings that more closely follow service-dominant

principles positively relate to value co-creation behavior of actors. We assessed the hypothesis through the

independent variable service-dominant design which is measured by two constructs: functional and

relational design, and the dependent variable value co-creation behavior (VCB) measured through six

constructs: information seeking, information sharing, personal interaction, feedback, advocacy and helping behavior. In our analysis we include five control variables: gender, age, office experience (three dummies),

organization size (three dummies) and actor role (two dummies). To test the hypothesis, a hierarchical

34 Table 10. Results of hierarchical regression analysis ª

ª Standardized regression coefficients are displayed; N = 276. ᵇ df = 12.

* Correlation is significant at the .05 level (1-tailed). ** Correlation is significant at the .01 level (1-tailed.