A Field Study of the Effect of Interpersonal Trust on [PDF]

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

RESEARCH ARTICLE

A FIELD STUDY OF THE EFFECT OF INTERPERSONAL TRUST ON VIRTUAL COLLABORATIVE RELATIONSHIP PERFORMANCE1 By: David L. Paul Department of Information Technology and Electronic Commerce Daniels College of Business University of Denver 2101 S. University Boulevard Denver, CO 80208 U.S.A. [email protected] Reuben R. McDaniel, Jr. Department of Management Science and Information Systems McCombs School of Business University of Texas at Austin Austin, TX 78712 U.S.A. [email protected]

Abstract This article examines the relationship between interpersonal trust and virtual collaborative relationship (VCR) performance. Findings from a

1

Michael D. Myers was the accepting senior editor for this paper.

study of 10 operational telemedicine projects in health care delivery systems are presented. The results presented here confirm, extend, and apparently contradict prior studies of interpersonal trust. Four types of interpersonal trust—calculative, competence, relational, and integrated—are identified and operationalized as a single construct. We found support for an association between calculative, competence, and relational interpersonal trust and performance. Our finding of a positive association between integrated interpersonal trust and performance not only yields the strongest support for a relationship between trust and VCR performance but also contradicts prior research. Our findings indicate that the different types of trust are interrelated in that positive assessments of all three types of trust are necessary if VCRs are to have strongly positive performance. The study also established that if any one type of trust is negative, then it is very likely that VCR performance will not be positive. Our findings indicate that integrated types of interpersonal trust are interdependent, and the various patterns of interaction among them are such that they are mutually reinforcing. These interrelationships and interdependencies of the different types of interpersonal trust must be taken into account by researchers as they attempt to understand the impact of trust on virtual collaborative relationship performance.

MIS Quarterly Vol. 28 No. 2, pp. 183-227/June 2004

183

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

Keywords: Interpersonal trust, collaboration, virtual teams, telemedicine

Introduction One should expect trust to be increasingly in demand as a means of enduring the complexity of the future which technology will generate (Luhmann 1979, p. 16). In a complex world, trust is a necessity. Trust effectively and efficiently reduces complexity by enabling parties with different knowledge bases and experiences to collaborate (Gefen 2000; Luhmann 1979; Lewis and Weigert 1985). Such collaborative relationships can be either virtual or face-to-face. In virtual collaborative relationships, technology is often considered the Achilles’ heel; it is likely that trust is the key issue. This research examines the relationship between interpersonal trust and virtual collaborative relationship (VCR) performance. Findings resulting from data collected in a field study of VCRs in health care delivery, specifically 10 operational telemedicine projects, are presented. VCRs in health care delivery have characteristics that make them exemplary subjects for studying trust in a context where trust is very important, yet they often operate in a context hostile to developing and maintaining trust. This study makes several contributions to research on interpersonal trust in VCRs. First, the formation of collaborative relationships, be they virtual or face-to-face, does not guarantee they will be effectively utilized. Collaborative relationships are inherently social constructions, and their success or failure may be due to the social context in which they exist, and not the quality of the relationships themselves. This study extends research on the relationship between interpersonal trust and performance. Partial support for a relationship between performance and affectbased interpersonal trust has been found (McAllister 1995), but no relationship between

184

MIS Quarterly Vol. 28 No. 2/June 2004

interpersonal trust and performance has also been found (Zaheer et al. 1998). Prior research found that antecedents to trust were a good predictor of trust development in virtual teams (Jarvenpaa et al. 1998), while another study found a relationship between communication processes and swift trust development (Jarvenpaa and Leidner 1999). Second, this study helps us to understand why technically sufficient information systems may not be adopted. Deploying information technology systems is often predicated on the assumption that the value of an information system’s information quality and increased information processing capabilities is sufficient to justify system use. However, this assumption does not always hold, and decision makers often rely on sources they trust the most—regardless of the timeliness, quantity, and quality of information from these sources (Mishra 1996; Staw et al. 1981). There are many reasons why systems are not adopted (Collins and Bicknell 1998). This study attempts to differentiate between complexity reduction efforts that fail due to insufficient information processing capabilities, such as inadequate hardware or software capabilities, and others that fail due to a lack of trust within the relationship. Third, this study introduces a methodology, facet theory, previously not utilized in information systems research. Facet theory originated in psychology and is a systematic approach to facilitating and integrating research construction, design, and data analysis of complex social systems. It utilizes multidimensional data analysis guided by a theoretical framework (Borg and Shye 1995; Guttman and Greenbaum 1998; Shye 1998). Facet theory has the potential to address many of the concerns and challenges information systems researchers face in performing field research. Fourth, this research makes a significant contribution to practice by demonstrating that organizations must address interpersonal trust factors if they want to reap the benefits of newer work relationships. It also suggests new considerations in the ongoing concerns about the more extensive adoption of virtual relationships, specifically telemedicine, in the effective and efficient deployment of health care resources.

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

Interpersonal Trust Trust in an organizational setting is an effective enabler of complexity reduction—especially when important decisions and new technology are concerned (Gefen 2000; Lewis and Weigert 1985; Ring and Van de Ven 1994). Trust plays a key role as a foundation for effective collaboration (Kramer 1999; Mayer et al. 1995; Rousseau et al. 1998; Whitener 1998) and is the salient factor in determining the effectiveness of many relationships (Gefen 2000; George and Jones 1998; Newell and Swan 2000; Sako 1998; Zand 1972). While trust may reduce transactions costs (Williamson 1981, 1985, 1993), its main impact on collaborative relationship performance is facilitating the learning and innovation (Goshal and Moran 1996; Newall and Swan 2000; Sako 1998) needed to address the ambiguity and unstructured nature of wicked decision problems (Mason and Mitroff 1973). A direct link between trust and collaborative relationship performance exists; once the need for collaboration is established, trust becomes the salient factor in determining performance. Trust is particularly important in newer organizational forms such as virtual collaborative relationships (McKnight et al. 1998; Meyerson et al. 1996; Newall and Swan 2000; Ring 1996). In the virtual world, trust is a way to “manage people whom you do not see” (Handy 1995, p. 41). Yet trust in virtual teams is difficult to build, and it has been argued that face-to-face contact is irreplaceable for building trust (Nohria and Eccles 1992). The replacement of technology for collocation undermines the emotional relationship aspects of trust. Collocation reinforces social similarity, shared values, and expectations, and increases the immediacy of threats from failing to meet commitments (Jarvenpaa and Leidner 1999; Latane et al. 1995; Sako 1998). Trust is especially important in health care delivery because health care providers rely on collaboration as a primary means of complexity reduction. Health care delivery is a collaborative activity whose quality, efficiency, and responsiveness is enhanced by the use of inter-

disciplinary teams (IOM 1996a). Traditionally, health care providers have reduced complexity by collaborating with other health care knowledge workers (IOM 1996a; Silberman 1992). Such complexity results from the co-morbidity of patient conditions that requires providers to simultaneously deal with multiple problems that interact with each other and often preclude assignment of causal relationships with any certainty (IOM 1996b). Patient symptoms and test results are subject to multiple plausible, but conflicting, interpretations (IOM 1996b). Multiple treatment courses with multiple, contradictory, and often uncertain outcomes are available (Silberman 1992), and patient and family members vary in their treatment and quality of life preferences. While trust is critically important in health care delivery relationships, the health care delivery environment does not facilitate the creation and maintenance of trust (IOM 1996b; OTA 1995;). Collaborative consultations can expose one to malpractice liabilities due to the actions of others. A health care provider engaging in collaborative consultations risks losing patients to the other party, and must trust that the other party will not attempt to steal patients. Health care delivery involves multidisciplinary teams, where team members have to rely on others whose training and perspective are different from their own, and where there are significantly different power and status relationships. Trust in such differentiated teams can be especially difficult to create and maintain (Luhmann 1979). Indeed, it can be argued that lack of trust is the natural state of the health care delivery environment. The importance of and difficulty in creating and maintaining trust is especially in force in telemedicine. Telemedicine is the process of two or more geographically separated health care providers collaborating via information technology to provide value-added health care delivery (IOM 1996b). The standard measures of health care delivery—and measures of the effectiveness of complexity reduction efforts—are access, cost, and quality (IOM 1990, 1993, 1996a), and it is claimed that telemedicine can increase the access to and quality of health care delivery while simul-

MIS Quarterly Vol. 28 No. 2/June 2004

185

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

taneously lowering costs (Allen and Stein 1998; Debakey 1995; Flaherty et al. 1995; GAO 1997; Grigsby 1995; IOM 1996b; ORHP 1997; OTA 1995). Telemedicine can result in significant economies of scale as expertise can be centralized and utilized more frequently (COC 1996; DOC 1997; GAO 1997; IOM 1996b; ORHP 1997). Telemedicine brings its own additional problems in creating and maintaining trust. Telemedicine projects involve multidisciplinary teams, and virtual processes may differ from what health care providers are used to. Telemedicine blurs individual boundaries and thrusts participants into unfamiliar interactions. Technology changes the mode of presentation on which trust formation is based. Technology may change how and what cues are noticed or ignored relative to face-to-face interactions, and it may force health care providers to interact with team members in a manner different from what they are used to. The installation of telemedicine infrastructure may be interpreted as threatening by remote health care providers, who may see it as the first step toward their replacement by nonlocal providers (GAO 1997). Interpersonal trust may not exist for many telemedicine projects. While there remain significant variations in conceptualizations of trust by organizational researchers, a general consensus has been reached that trust is a psychological state based on confident expectations and beliefs that another party will act in a certain manner, and that the trusting party must in some way be vulnerable under conditions of risk and interdependency to actions by the other party (Kramer 1999; Rousseau et al. 1998). Our study focuses on the relationship of interpersonal trust with virtual collaborative relationship performance. Identifying types of interpersonal trust is a contentious and confusing issue. Researchers identify different types of interpersonal trust, use different terminology for similar types of trust, or use similar terms for different types of trust, and subcategorize the same type of trust in different ways. These various subcategories introduce complexity into the study of trust (Bigley and Pearce 1998).

186

MIS Quarterly Vol. 28 No. 2/June 2004

For example, cognition trust differs in its definitions and, for some, comprises subcategories of trust that others consider separate types of trust. Some researchers consider calculative trust a core type of interpersonal trust, while others do not consider it a type of trust at all. In this section, we present our model of interpersonal trust and the hypothesized relationships with VCR performance. We have attempted to bring to light the major types of trust in the literature, how definitions of the same type of trust differ, and why we chose the definitions and operationalizations we did. Figure 1 presents our model of interpersonal trust and the hypothesized relationships with VCR performance. We have identified three types of interpersonal trust—calculative, competence, and relational—and combined them for a fourth type of interpersonal trust—integrated—which includes all three types of trust in our model.

Calculative Trust Calculative trust is based on conceptualizing trust as a form of economic exchange (Child 1998; Lane 1998; Lewicki and Bunker 1996). Also termed rational trust (Gambetta 1988; Lewicki and Bunker 1996; Mayer et al. 1995; Williamson 1993), calculus-based trust (Rousseau et al. 1998), commitment trust (Newell and Swan 2000), and contractual trust (Sako 1991, 1992), calculative trust is an ongoing, market-oriented, economic calculation where each party assesses the benefits and costs to be derived from creating and sustaining a relationship (Child 1998; Lewicki and Bunker 1996). Calculative trust is a form of contractual agreement where parties can be relied on to deliver according to the details of the contract (Newell and Swan 2000; Sako 1991, 1992). The parties choose whether or not to participate in a trusting relationship based on some form of costbenefit analysis. Individuals are assumed to be economically rational beings motivated by their desire to maximize expected gains or minimize expected losses in their transactions (Kramer 1999).

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

Interpersonal Trust

Self-Interest Ability Empathy

Virtual Collaborative Relationship Performance

Integrated

Figure 1. Interpersonal Trust and VCR Performance Model

The concept of calculative or rational trust is not without its detractors. Some argue that trust is not the result of conscious calculation (Kramer 1999) nor is it a form of economic exchange (March and Olsen 1989), concluding that calculative trust does not exist. Trust is needed only when conditions of information uncertainty exist; however, calculative trust is effective only when there is little or no need to trust because there are only limited, identifiable conditions of information uncertainty (Child 1998; Lane 1998). Therefore, calculative trust, and the ability to assess the costs and benefits of engaging in a relationship, comes into play only under conditions in which the need for trust is limited. Despite these objections, we include calculative trust in our model because the condition that the trusting party must be vulnerable to the nonperformance of the other party in our definition of trust holds in this case. Trust includes motivational components (Kramer 1999; Shepard and Tuchinsky 1996), and some of these motivational components may be calculative. This leads to our first hypothesis.

Competence Trust Competence trust is whether the other party is capable of doing what it says it will do (Butler 1991; Butler and Cantrell 1984; Mayer et al. 1995; Mishra 1996; Sako 1991, 1992, 1998). There appears to be a definitional consensus about competence trust in the research community. It is an assessment of the expertise and abilities of the other parties, and is important in a knowledgebased economy because it acts as an indicator of the other party’s ability to perform as anticipated (Rousseau et al. 1998). Competence trust is required in complexity reducing collaborative efforts when the skills needed to perform a task are not found within one person (Newall and Swan 2000). A party is more likely to engage in a collaborative relationship if they perceive the individuals in the other party as being capable. H2: There is a positive association between competence trust and virtual collaborative relationship performance.

Relational Trust H1: There is a positive association between calculative trust and virtual collaborative relationship performance.

The third type of trust in our model, relational or benevolence trust, is the extent one feels a

MIS Quarterly Vol. 28 No. 2/June 2004

187

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

personal attachment to the other party and wants to do good by the other party, regardless of egocentric profit motives (Jarvenpaa et al. 1998; Mayer et al. 1995). Variations of relational trust include normative trust (Child 1998), goodwill trust (Sako 1991, 1992, 1998), affect-based trust (McAllister 1995), identification trust (Lewicki and Bunker 1996), companion trust (Newall and Swan 2000), and fairness (Zaheer et al. 1998). A motivation to do good by the other party is key to these definitions. These definitions all include one party empathizing with the other party, and specifically excludes the notion of calculative trust. However, there are significant differences in these definitions as well. Some definitions may or may not include friendship (Lewicki and Bunker 1996; Newell and Swan 2000), affect (Kramer 1999; McAllister 1995; Zaheer et al. 1998), shared identity (Lewicki and Bunker 1996), goodwill (Newell and Swan 2000), common values (Child 1998; Lane 1998), mutual understanding (Lewicki and Bunker 1996) and dependability (Zaheer et al. 1998). Some researchers also include cognitive and motivational underpinnings of relational trust (Kramer 1999). For our purposes, the key to relational trust is that one party empathizes with the other party and wants to do good by them for altruistic reasons. Relational trust is thought to be especially important to the success of collaborative activities (Jarvenpaa and Leidner 1998; Sako 1998). However, prior research (McAllister 1995) has found only partial support for a relationship between performance and affect-based interpersonal trust. This leads to our third hypothesis. H3: There is a positive association between relational interpersonal trust and virtual collaborative relationship performance.

1995). Trust can take different forms in different relationships, and different forms of trust may mix together and interact in some situations. Trust may have a bandwidth that can vary in both scope and degree (Rousseau et al. 1998). For example, some relationships may rely more on a combination of calculative and competence trust, while other relationships may be based more on a combination of relational and competence trust. Further, one type of trust may evolve into another, deeper type of trust. Rousseau et al. (1998) speculated that in terms of interpersonal trust, calculative trust was more important in the early stages of a relationship, while relational trust was more influential in the later stages. Other authors have proposed similar ideas, with the initial creation of calculative trust paving the way for the equivalent of relational trust development (Bachman 1998; Child 1998; Lewicki and Bunker 1996; Newall and Swan 2000). In addition, there is the issue of whether the various types of trust can compensate for each other (Mayer et al. 1995). Can highly positive relational trust compensate for negative calculative trust, and vice versa? Can negative competence trust be offset by highly positive calculative and/or relational trust? While we do not specify a temporal relationship between the different interpersonal types of trust, we argue that the different types of interpersonal trust affect and are affected by each other, and that a combination of the different types of interpersonal trust together impact collaborative relationship performance. However, a prior study (Zaheer et al. 1998) did not find support for a relationship between integrated interpersonal trust and performance. This leads to our final hypothesis. H4: There is a positive association between integrated trust and virtual collaborative relationship performance.

Integrated Trust The integrated perspective of interpersonal trust (Lewicki and Bunker 1996; Mayer et al. 1995; Zaheer et al. 1998) combines the different types of trust. The different types of trust are related to each other, even though they may be separable and vary independently of each other (Mayer et al.

188

MIS Quarterly Vol. 28 No. 2/June 2004

Types of Trust Excluded from the Model A number of types of trust identified in the literature were excluded from our model because we felt that they either were not a type of inter-

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

personal trust or, given our definitions, were effectively included in the model. Foremost among these was cognition trust (Child 1998), also called cognition-based trust (McAllister 1995), and knowledge-based trust (Lewicki and Bunker 1996). We chose to exclude cognition trust because the definitions of it were often contradictory or overlapped with the definitions in our model. Some authors included dependability (McAllister 1995), reliability (McAllister 1995), and competence and/or benevolence (Jarvenpaa et al. 1998; Mayer et al. 1995; McKnight et al. 1998) within their definitions of cognitive trust. Trust as a psychological state implies that all trust is cognitive in that it is based on beliefs about another party. The types of trust in our model (calculative, competence, relational) are all cognitive in that they are judgments or beliefs. The confusing definitions of cognition trust may explain why prior research (McAllister 1995) had difficulty in testing a relationship between cognition trust and performance. A common aspect of cognition trust is the inclusion of predictability, which is the degree of consistency in intended behavior. It is also perceived as a type of trust in and of itself (Zaheer et al. 1998). We have not included predictability as a type of trust because predictability is not enough to explain trust (Bachman 1998), and to be meaningful, trust must go beyond predictability (Deutsch 1958; Mayer et al. 1995). Instead, we include predictability from both an economic standpoint and a relational standpoint in two of our types of trust. In calculative trust, one party is perceived as predictable if it is in their own selfinterest to do so, while in relational trust, one party is seen as predictable because they empathize with the other party. Some authors include reliability and/or dependability in their definitions of competence trust or cognition trust, where reliability is whether the other party can be relied on to fulfill their obligations (Anderson and Weitz 1989; Zaheer et al. 1998). Reliability is also sometimes perceived as a type of trust in and of itself (Zaheer et al. 1998). Reliability was excluded from our model because professional ethos and fear of malpractice tended

to ensure that one party thought the other party reliable prior to agreeing to engage in the telemedicine relationship. It was thus a necessary precondition for the parties to engage in a VCR. Further, reliability dropped out of Zaheer et al.’s (1998) measures of interpersonal trust. Deterrence-based trust was excluded from our model because we agree with the assessment of other researchers (Hagen and Choe 1998; Rousseau et al. 1998). The ability to impose costly sanctions on the other party is a substitute for trust. It is not a form of trust. We excluded experienced-based trust from our model. Experience-based trust is derived from a conceptualization of interpersonal trust as being based on past interactions with the other party (Deutsch 1958, 1960; Kramer 1999). However, Luhmann (1979, 1988) argued that familiarity was a precondition to and not a form of trust, and Gefen (2000) found that familiarity was separate from trust.

Method Research Setting This research studies the association of interpersonal trust with VCR performance in the health care delivery environment. This environment was selected because it is a knowledge-based industry whose body of knowledge is expanding rapidly. Traditionally, health care delivery has utilized collaboration as a means of complexity reduction (IOM 1996a; Silberman 1992). Advances in information technology make virtual collaboration possible. Telemedicine is the process of two or more geographically separated health care providers collaborating via information technology to provide value-added health care delivery (IOM 1996b), and it is usually believed that telemedicine can increase the access to and quality of health care delivery while simultaneously lowering costs.

Research Design A mixed research design was utilized in this study. Given the sensitivity of trust to the context in which

MIS Quarterly Vol. 28 No. 2/June 2004

189

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

it exists and the emergent nature of VCRs in general, comparative case studies were utilized as the primary research design because they can capture the nuances and the richness of these phenomena and increase the robustness and generalizability of findings through replication (Yin 1994). To enhance methodological rigor, we utilized facet theory (Borg and Groenen 1997; Borg and Shye 1995; Canter 1985a; Guttman 1959, 1968, 1982a; Levy 1994; Shye et al. 1994), a systematic approach to research design and data analysis developed in the field of psychology. The design of the study may appear to be closely related to grounded theory; however, it isn’t. We chose facet theory instead of grounded theory because we had some a priori hypotheses we wanted to test. Facet theory can be complementary to grounded theory in that it provides a means to structure the theory generating and testing process. Appendix A presents an overview of facet theory and Table A1 presents a glossary of facet theory terms for reference. Figure 2 presents the facet theory mapping sentence used for this research. The unit of analysis was the telemedicine relationship—multiple health care providers utilizing telemedicine equipment to collaborate. Two internal content facets were identified. Facet A was the collaboration/environmental interaction facet representing both the telemedicine relationship’s interpersonal trust and impact values. Facet B was the agency facet representing the different parties of the telemedicine relationship— the health sciences center (HSC) and remote site. Facet A was somewhat unusual in that the hypothesized correlates were elements of the same facet, where interpersonal trust and impact were regarded as two aspects of interaction between the collaborative relationship and its environment, with the former acting as an input and the latter acting as an output. In this facet, interpersonal trust was further broken down to self-interest, ability, empathy, and integrated interpersonal trust, and impact was further broken down to access, cost, and quality. Calculative trust was operationalized as self-interest, defined as the extent to which the individuals in one party perceive the benefits from directly participating in the collaborative relationship are greater than the

190

MIS Quarterly Vol. 28 No. 2/June 2004

costs (Kramer 1999; Lewicki and Bunker 1996; Williamson 1993), while competence trust was operationalized as ability, defined as the degree to which the individuals in one party perceive the individuals in the other party as having the necessary expertise and capability (Jarvenpaa et al. 1998; Tyler and Kramer 1996; Zucker et al. 1996). Relational trust was operationalized as empathy, defined as the degree to which the individuals in one party desire to do well by those in the other party (Jarvenpaa et al. 1998; Tyler and Kramer 1996). Telemedicine’s impact on remote site health care delivery was measured as the perceived change relative to the conditions prior to the advent of telemedicine. Access is the timely use of personal health services to achieve the best possible health outcomes (IOM 1993). Quality is the degree to which health care services for individuals and populations increase the likelihood of desired health outcomes and are consistent with current professional knowledge (IOM 1990). The impact on cost is relative to the cost of similar episodes of medical care prior to the start of the telemedicine project and therefore excluded both the operating and capital costs of the telemedicine equipment. All of these elements were present in all of the relationships (as represented by the brackets around Facet A’s elements), and the range facet to the right of the arrow represented all of the possible values for all of Facet A’s elements—very positive, positive, slightly positive, neither positive nor negative, slightly negative, negative, and very negative.

Sample Three telemedicine networks located in the United States and involving at least three operational telemedicine projects were studied. Each of the networks had at its hub a university-affiliated health sciences center (HSC), and the spokes of the networks were located in rural areas where per capita income levels were substantially below the national level (Census Bureau 1990). HSCs were selected because the vast majority of telemedicine projects involve university-affiliated

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

In telemedicine relationship (x), the extent of A: Collaboration-Environmental Interaction 1. Interpersonal Trust {1.1 Self-interest} {1.2 Ability} {1.3 Empathy} {1.4 Overall} 2. Impact {2.1 Access} {2.2 Quality} {2.3 Cost}

as reported by individuals at

B: Agency

Range Facet

(1. HCS) (2. Remote Site) was

(Very Positive) (Positive) (Slightly Positive) (Neither Positive nor Negative) (Slightly Negative) (Negative) (Very Negative)

in the sense of facet A (CollaborationEnvironmental Interaction)

Figure 2. Mapping Sentence of Model

health sciences (or medical) centers (IOM 1996b; ORHP 1997), and they are charged with providing health care to remote areas. The telemedicine projects studied were part of the normal practice of medicine either as revenue generating or cost reducing projects. Ten telemedicine projects involving five teleconsultation, three distance learning, and two teleradiology telemedicine activities were examined. Within each of these telemedicine projects, two relationships were studied: the relationship between the relevant individuals at the HSC and those at the remote site, and the relationship between the individuals at the remote site and those at the HSC. Theoretical (or purposeful) sampling was utilized in an effort to address potential threats to the external validity and construct validity of this research (Strauss and Corbin 1998). HSCs were selected because they and their telemedicine project partners tended to have certain characteristics that naturally accounted for alternative explanations for the impact or the lack thereof of installed telemedicine projects (Paul et al. 1999). For example, all of the HSCs had a mission to improve health care for their respective rural populations, and all of the specialists were paid a salary by the state. Therefore, the lack of specialist reimbursement was less likely to be an inhibiting factor in the short run. All of the telemedicine projects studied were intrastate, eliminating potential interstate physician licensing

barriers. Malpractice liability insurance concerns were less an issue because all of the specialists were engaged in HSC sanctioned telemedicine projects; therefore, their activities were covered by their respective facility’s umbrella liability coverage. High start-up and operating costs were not inhibitors because almost all of the projects in this research received external funding. Site selection was based on four criteria. First, each site had to have at least three active telemedicine projects. Second, each site had to have one of each of the three types of telemedicine activities: teleconsultation, distance learning, and teleradiology. These two criteria enabled both within and between network comparisons of different telemedicine projects. Third, the sites could not involve military or correction facilities because the voluntariness of participation and the dynamics of trust in such situations may be different from those in civilian projects. Fourth, each site had to have been operational for a minimum of six months to allow the inevitable technological and procedural bugs to be addressed and to allow the novelty of telemedicine to pass.2

2

The one exception was a pediatric oncology teleconsultation project that was discontinued by the pediatric oncologists after a period of four months.

MIS Quarterly Vol. 28 No. 2/June 2004

191

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

The World Wide Web was searched to find sites that met these criteria. The second criterion— different types of telemedicine activities within each project—was discarded because sites meeting this criterion could not be found. Although a number of potential sites claimed to have all three types of telemedicine activities operational at the time of this study, only one actually did. Indeed, a number of potential sites that claimed on their Web pages to have active telemedicine projects did not have any active telemedicine projects at the time of this study. This exaggeration of the state of active telemedicine projects was not uncommon. The ORHP (1997) found that approximately 25 percent of the hospitals they surveyed which claimed to have at least one active telemedicine project in fact had no operational telemedicine projects. Each site selected included at least one teleconsultation project, which enabled teleconsultation activities to be compared across the telemedicine networks. Both distance learning and teleradiology projects occurred at two sites, enabling at least one between network comparison for these telemedicine activities. Table 1 presents a summary of the telemedicine relationships studied. Two of the five teleconsultation projects involved multiple specialties where the patient and his or her family were usually present during the sessions. The multiple specialties teleconsultation project involving primary care physicians at a rural hospital was judged to have a positive impact on remote site health care, while the other involving a physician assistant at a rural health clinic was not. An infectious diseases teleconsultation project between HSC specialists and a rural hospital’s primary care physicians with patients usually not present during the sessions was judged to have a very positive impact on remote health care delivery while a pediatric oncology teleconsultation project involving pediatric oncologists at the HSC and nurses with patients and their parents present during the sessions was not. A bone marrow transplant teleconsultation involving HSC transplant specialists, nurses, psychologists, and administrative staff and two rural oncologists at a private clinic with the patient and his or her family almost always present during the sessions was

192

MIS Quarterly Vol. 28 No. 2/June 2004

judged to have a positive impact on remote site health care delivery. This teleconsultation project was utilized for initial consultations to determine if the patient was a viable candidate physically for a transplant and psychologically for the extended stay in isolation the transplant entailed, and whether the patient wanted to undergo a high mortality treatment for a life-threatening disease with these particular specialists. It was also used for follow-up after the transplantation procedure. A distance learning rural primary care residency telemedicine project involving HSC specialists and residents receiving specialized training in rural health care at a regional medical center where the patients were sometimes present during the sessions was deemed to have a positive impact on remote site health care delivery, while a distance learning rural primary care clerkship telemedicine project involving HSC specialists and third-year medical students at a primary care clinic without patients present during the sessions was not. A distance learning rural telemedicine project involving HSC specialists and a podiatry residency rotation for the HSC’s first-year podiatry residents at a group of three federally funded health clinics and the local state hospital where the patients were sometimes present during the sessions was deemed to have a very positive impact on remote site health care delivery. Diabetes was a major health care problem in the region, and the lack of access to preventive care due to economic factors for the population most at risk resulted in the region having a rate of amputations among diabetics that was significantly above the national average. The two teleradiology projects involved digitized radiographic images being sent to the respective HSCs, where the radiologists would read the images and provide a diagnosis by e-mail or telephone, depending on the urgency of the situation. In some cases, the radiologists would telephone the rural primary care physicians for additional information. Both teleradiology projects had undergone at least a four month trial period during which the transmitted image quality was deemed to be an adequate basis for making a diagnosis. One teleradiology project was determined to have a positive impact on remote site health care, while the other was not.

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

Table 1. Telemedicine Relationships Overview Relationship

Type*

A

HSC!R

B

R!HSC

C

HSC!R

D

R!HSC

E

HSC!R

F

R!HSC

G

HSC!R

H

R!HSC

I

HSC!R

J

R!HSC

K

HSC!R

L

R!HSC

M

HSC!R

N

R!HSC

O

HSC!R

P

R!HSC

Activity (Project Duration) Teleradiology (2 years)

Distance Learning—Rural Residency (1 year)

Teleconsultation— Oncology (9 months) Teleconsultation— Pediatric Oncology (4 months)** Distance Learning—Medical Student Clerkship (1 year) Distance Learning—Podiatry Residency (9 months)

Teleconsultation— Infectious Diseases (1.5 years)

Teleradiology (1 year)

Score Structuple & (Impact Values)***

Team Composition & Patient Involvement

Team Stability

Radiologist

Stable

217 (12)

Attending Primary Care Physician (PCP)

Stable

772 (12)

Supervising Physician (SP) & Rotating Specialists, Patient (sometimes)

Semistable

777 (21)

Residents & SPs, Patient (sometimes)

Stable

747 (21)

Oncologists, Nurses, Psychologist, Admin.

Stable

747 (19)

Oncologist, Patient & Family (always)

Stable

777 (19)

Pediatric Oncologists

Stable

146 (11)

Nurse, Patient & Parents (always)

Stable

742 (11)

SP & Rotating Specialists, Patient (rarely)

Semistable

267 (12)

Medical Students

Stable

746 (12)

SP & Residents, Patient (sometimes)

Stable

777 (21)

Stable

775 (21)

Stable

777 (21)

PCPs

Stable

777 (21)

Radiologist

Stable

447 (16)

Attending PCP

Stable

765 (16)

Residents, SP, & Area Podiatrists, Patient (sometimes) Infectious Diseases Specialist & Other Specialists (rarely)

MIS Quarterly Vol. 28 No. 2/June 2004

193

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

Table 1. Telemedicine Relationships Overview (Continued) Relationship

Type*

Q

HSC!R

R

R!HSC

S

HSC!R

T

R!HSC

Activity (Project Duration) Teleconsultation— Multiple Specialties (8 years)

Teleconsultation— Multiple Specialties (5 years)

Team Composition & Patient Involvement

Team Stability

Score Structuple & (Impact Values)***

Different Specialists, Telemedicine Administrator

Semistable

771 (21)

PCP & Patient (sometimes)

Stable

777 (21)

Different Specialists, Telemedicine Administrator

Semistable

441 (12)

Physician Assistant, Patient (sometimes)

Stable

753 (12)

NOTES: * ** ***

HSC!R means the relationship between the relevant individuals at the HSC and those at the remote site. R!HSC means the relationship between the individuals at the remote site and those at the HSC Project was discontinued by the pediatric oncologists. The number in the first line is the relationship’s score structuple value. • The first number represents ability. • The second number represents empathy. • The third number represents self-interest. The number in parenthesis in the second line is the relationship’s impact on remote site health care delivery.

Data Collection Issue-focused, semi-structured interviews of key informants provided thick and richly textured data (Orlikowski 1993; Sackmann 1991) and eliminated the problem of item non-response which plagued earlier telemedicine studies (ORHP 1997). In all, 74 health care professionals were interviewed face-to-face, and the interviews were audiotaped and transcribed. Key informants were members of one of three groups: clinicians (physicians, physician assistants, or nurse practitioners), administrators, and information technology professionals.3 They were selected based on current or

3

In some of the smaller health care facilities, individuals often played dual roles. For example, at one remote site, the residency program administrator was also a physician who taught courses. In these cases, key informants were classified only in one role but were asked questions about both roles played.

194

MIS Quarterly Vol. 28 No. 2/June 2004

past direct involvement in their organization’s telemedicine project. Figure 3 presents a summary of the key informants by position and location. The interviews were approximately equally split between those conducted at the HSCs and the remote sites, and the number of interviews per HSC was proportional to the number of telemedicine projects they had active. Further, the proportion of key informants who were clinicians, administrators, and IT professionals was fairly evenly distributed across the sites. Construct validity and reliability were enhanced by triangulated data collection (Eisenhardt 1989; Yin 1994). This was achieved by interviewing multiple key informants and collecting multiple types of data. Teleconsultations or videotapes of teleconsultations were observed when possible, and documentation and archival data such as grant proposals and follow-up, needs assessments, and strategic plans were collected when available.

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

Health Total by HSC I Admin. IT Prof. Care Prof. Location

Health Total by HSC II Admin. IT Prof. Care Prof. Location

HSC (A,C,E)

4

4

5

13

HSC (G,I,K,M)

4

3

8

B

1

1

2

4

H

1

3

0

4

D

1

2

1

4

J

1

1

3

5

F Total

1 7

0 7

2 10

3 24

L *N

1 0

0 0

2 1

3 1

Total

7

7

14

28

Health Total by HSC III Admin. IT Prof. Care Prof. Location

15

Health Total by TOTAL Admin. IT Prof. Care Prof. Location

HSC (O,Q,S)

3

3

5

11

P

1

1

2

4

HSC I

7

7

10

24

R

1

3

2

6

HSC II

7

7

14

28

T Total

0 5

0 7

1 10

1 22

HSC III Total

5 19

7 21

10 34

22 74

* NOTE: N Key Informants also include H Administrator and IT Professionals

Figure 3. Key Informants by Site and Position

The perceived impact on health care delivery was used as legal issues prevented the researchers from having access to patient records, and most telemedicine sites tended not to maintain such records (ORHP 1997). The impact on remote health care delivery was consistent with the conceptualization of the outcome of collaboration as the value added in terms of complexity reduction and the impact of trust on collaborative relationships as improved performance (Goshal and Moran 1996; Sako 1998). It was also consistent with the output of collaboration being “the enhancement of transaction value” (Zaheer et al. 1998, p. 155).

Data Analysis Qualitative Analysis The transcribed interviews were analyzed and coded. The coding scheme was theoretically based (Martin and Turner 1986), where quotations were categorized according to facet and element values. Internal validity was enhanced through

the use of pattern matching (Strauss and Corbin 1998; Yin 1994), and the explicit specification of the mapping sentence provided structure by which to engage in constant comparative analysis both within and between cases. Key informant interviews established the need for the telemedicine projects and eliminated alternative explanations such as technology problems for the failure of a telemedicine project. They also established trust as the critical factor for the success or failure of the telemedicine relationships, and were used to determine the type and level of trust involved in the different telemedicine relationships. Key informant interviews were also used to assess the impact the telemedicine project had on the cost, quality of, and access to remote site health care. Note that the coding examples presented in the Qualitative Results section involve only one key informant. In most cases (including those presented), multiple confirming comments from different key informants involved in that particular telemedicine relationship, as well as other forms of evidence, were used to determine the coding value.

MIS Quarterly Vol. 28 No. 2/June 2004

195

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

Table 2. Weak Monotonicity Coefficient (Guttman’s mu2) Matrix ABILITY

EMPATHY SELFINT HSC

REMOTE IMPACT

ACCESS COST

QUALITY

ABILITY

1.000

EMPATHY

0.868

1.000

SELFINT

-0.311

-0.016

HSC

-1.000

-0.389

0.259

1.000

REMOTE

1.000

0.389

-0.259

-1.000

1.000

IMPACT

0.937

0.848

0.633

0.000

0.000

1.000

ACCESS

0.913

0.812

0.582

0.000

0.000

1.000

1.000

COST

0.914

0.802

0.602

0.000

0.000

0.997

0.954

1.000

QUALITY

0.821

0.768

0.623

0.000

0.000

0.988

0.987

0.920

1.000

1.000

Table 3. Impact Values by Telemedicine Relationship A&B

C&D

E&F

G&H

I&J

K&L

M&N

O&P

Q&R

S&T

Access

5

7

6

4

4

7

7

5

7

4

Quality

5

7

6

4

4

7

7

7

7

4

Cost

2

7

7

3

4

7

7

4

7

4

OVERALL

12

21

19

11

12

21

21

16

21

12

Quantitative Analysis The coded qualitative data were converted into quantitative data using an ordinal seven-point Likert scale for the trust and perceived impact elements. The reliability of coding of the qualitative data was enhanced by the use of computer aided qualitative data analysis software (Kelle 1995; Morris 1994; Wolfe et al. 1993). In addition, the researchers’ coding of the trust and impact variables for each telemedicine relationship was assessed by an information systems professor whose research interests included trust. The third party assessor concurred with the researchers’ coding 94 percent of the time. Table 2 presents the weak monotonicity coefficient (Guttman’s mu2) matrix for the elements of the two content facets. The original conceptualization of the model anticipated analyzing the association between the different types of trust and how they were perceived as impacting the access, cost, and quality of health care delivery in

196

MIS Quarterly Vol. 28 No. 2/June 2004

the remote areas. However, as exhibited in Table 2, the impact values for each site—access, cost, and quality—were highly correlated (the minimum correlation was 0.92), and testing each impact type was unlikely to provide additional meaningful information or insights. Therefore, the three impact types were summed for each relationship, and only this one figure, the overall impact on health care delivery (IMPACT), was used in the data analysis. The maximum score possible for impact was 21 points and the minimum possible score was 3 points. In this research, the highest score was 21 points, and the lowest score was 11 points. The impact facet was unbalanced in that impact values could be positive and very positive but, from a practical standpoint, could not be negative or very negative because it was difficult for telemedicine relationships to have a negative or very negative impact in terms of either access to, or quality of, remote site health care delivery. Table 3 presents the impact scores for each of the telemedicine projects.

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

Model Dimensionality and Construct Validity Smallest space analysis (SSA) was used to evaluate the dimensionality and construct validity of the model, and a two-dimensional model was selected as the best model. The coefficient of alienation for the two-dimensional model was 0.036, which indicated a good fit for the monotonic relationship between input coefficients and output distances (Guttman 1968). The proportion of variance explained (RSQ) value was 0.997, and neither it nor the coefficient of alienation could be significantly improved upon by adding additional dimensions to the model. The shape of the Shepherd diagram for the two dimensional model (exhibited in Figure 4)—negatively sloped monotone series of points—indicated satisfactory goodness-of-fit for the model. The partitioning of the SSA space (exhibited in Figure 5) indicated a strong correspondence between the mapping sentence and the spatial contiguity of the data (Borg and Shye 1995; Shye et al. 1994). All of this suggested that the model presented in Figure 1 had good construct validity.

Hypotheses Testing Partial order scalogram analysis with base coordinates (POSAC) was used to test the hypotheses. Table 4 presents the POSAC coordinates for the calculative trust and agency elements’ values for each profile and Figure 6 exhibits the POSAC diagram of the different telemedicine relationships suggested in Hypothesis 1. In Figure 6, agency is represented by HSC and R, which represent health sciences centers and remotes sites, respectively. The number in brackets is the calculative trust (internal content facet element) value for those relationships (Borg and Shye 1995; Brown 1985). The final number(s) is(are) the external item—perceived impact value for those relationships with the corresponding calculative trust value.4 The model’s CORREP value,

4

It is standard practice in POSAC diagrams to label the points by either their profile ID or their structuple values. A separate diagram, called external item diagram, which superimposes the external item values for each profile in

exhibited at the bottom of Table 5, was 0.9800 (out of a possible 1.000), which indicated an excellent fit of the profiles in the POSAC space in terms of their relationship to each other. The perceived impact on remote site health care delivery element values are usually exhibited by themselves in the external item diagram. The patterns formed by these values represented the regional hypotheses and were used to test the hypotheses by assessing whether there was a relationship between the content facets’ elements and the external item (Levy and Guttman 1985; Shye and Amar 1985). The hypothesis is the higher the order of a telemedicine relationship as represented by its interpersonal trust values, or some combination thereof, then the greater the perceived impact of that telemedicine relationship on the remote site’s health care delivery. In Figure 6, those relationships whose perceived impact value is consistent with the hypothesized relationship with calculative trust are in plain text, while those that are not are bolded and italicized. The weak monotonicity coefficient values of calculative trust, HSC, and the perceived impact on remote site health care delivery elements in relation to the X, Y, joint, and lateral axis are exhibited in Table 5. The coefficient values were used to determine the roles the different facets and their elements played and whether the hypothesis was empirically supported. The X axis represented self-interest, the calculative trust facet element whose value increased as the X value increased. The Y axis represented the agency facet, with the remote sites located on the top half of the POSAC diagram and the HSCs located on the bottom half of the diagram. The hypothesis was supported if the perceived impact’s weak monotonicity coefficient value with the joint axis approached one and its value with the lateral axis approached zero. This support also could be viewed graphically in the POSAC diagram if the telemedicine relation-

the POSAC diagram, is usually presented by itself and is the means by which hypotheses are tested (Levy and Guttman 1985; Shye and Amar 1985). Due to space considerations, the POSAC diagram and the external item diagram were combined in this paper.

MIS Quarterly Vol. 28 No. 2/June 2004

197

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

3

Distances

2

1

0 -1.5

-1.0

-0.5

0.0 Data

0.5

1.0

1.5

Figure 4. Shepard Diagram of Two Dimensional SSA of VCR Performance Model

2

EMPATHY

Dimension-2

1

IMPACT 0

ABILITY

HSC

REMOTE

SELFINT

-1

-2

-2

1

0 Dimension-1

1

Figure 5. Two Dimensional SSA of VCR Performance Model

198

MIS Quarterly Vol. 28 No. 2/June 2004

2

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

Table 4. POSAC Coordinates and Coefficients for Calculative Trust (H1) Telemedicine Relationship

Agency

Selfinterest

X

Y

Joint

A, C, E, I, K, M, O

HSC

7

0.943

0.577

0.760

0.683

12, 12, 16, 19, 21, 21, 21

D, F, N, R

Remote

7

0.745

0.943

0.844

0.401

19, 21, 21, 21

G

HSC

6

0.882

0.333

0.608

0.774

11

J

Remote

6

0.667

0.882

0.774

0.392

12

L, P

Remote

5

0.471

0.816

0.644

0.327

16, 21

T

Remote

3

0.577

0.745

0.661

0.416

12

B, H

Remote

2

0.333

0.667

0.500

0.333

11, 12

Q, S

HSC

1

0.816

0.471

0.644

0.673

12, 21

Lateral

Impact

Proportion of profile pairs correctly represented (CORREP): 0.980

1.0 R [7] 19, 21, 21, 21

R [6] 12 R [5] 16, 21

0.8

R [3] 12 R [2] 11, 12

HSC [7] 12, 12, 16, 19, 21, 21, 21

Agency

0.6

HSC [1] 12, 21

0.4 HSC [6] 11

0.2

0.0

0.0

0.2

0.4

0.6 Self-Interest

0.8

1.0

Figure 6. POSAC Diagram of Calculative Trust Model (H1)

MIS Quarterly Vol. 28 No. 2/June 2004

199

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

Table 5. H1: Calculative Trust Weak Monotonicity Coefficient Matrix X

Y

JOINT

LATERAL

IMPACT

0.4310

0.5034

0.8008

-0.0152

SELFINT

0.7797

0.4466

0.9640

0.3552

HSC

1.0000

-1.0000

0.0949

1.0000

ships with the highest perceived impact were located in the upper right-hand corner of the POSAC diagram, partitioned by a line parallel to the lateral axis.

Results Qualitative Results Interviews with key informants established the need for the telemedicine projects. For example, prior to the infectious diseases teleconsultation project, the infectious diseases patients—the majority of whom were uninsured—often incurred additional health problems that required specialists the local state hospital lacked. However, these patients were not in the condition to travel, and even if they made it to the medical center 300 miles away, the medical center would not accept these patients because their contract with the state did not allow for the reimbursement for expenses incurred as a result of providing care to uninsured patients not in stable condition. The local state hospital physicians perceived the teleconsultation project as a means of providing the specialty care the infectious diseases patients needed to survive. State Hospital Physician: The worst case scenarios are our infectious diseases cases again. They [the medical center 300 miles away] don’t want to touch them with a 10 foot pole. We’ve even gotten feedback that they’re not salvageable. Well, we’ve lost two patients out of 40—both of them died of heart disease, not of infectious diseases.

200

MIS Quarterly Vol. 28 No. 2/June 2004

Our cure rates are looking like they are going to be very good. But in the meantime, we’ve got to get them through their massive bleeds and stuff like that and if no one wants to help us take care of them because they don’t think they’re viable, we’re stuck literally putting your finger in the dike and it’s a real uncomfortable feeling as a clinician. We’re trying to fix that real quickly, but in [state]—you always sent the patient where they needed to go. Would you like to get in the ambulance or the lifeflight helicopter with somebody with a very contagious disease? Key informant interviews eliminated alternative explanations such as technology problems for the failure of a telemedicine project and established trust as a critical factor. Questions about the usability, reliability, and sufficiency of the technology were specifically asked, but there was no indication that the technology was a cause of project failure, nor were there any indications that there was a lack of interpersonal trust of the technologists. For example, a pediatric oncologist involved in a teleconsultation project felt the technology worked but the teleconsultation project failed because of the nurses at the remote state hospital. HSC Pediatric Oncologist: I think overall, we thought we could do a really good job with [the telemedicine equipment]. The only—the only—problem there was was with the operator on the other end. Key informant interviews also established the importance of trust in telemedicine, particularly projects involving multidisciplinary teams. There

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

was a tendency for HSC specialists to denigrate the ability of the remote physicians, and remote physicians would often refuse to work with specialists who, consciously or not, treated them as second class citizens. A rural hospital physician involved with the multiple specialties teleconsultation project discussed what it was like working with the HSC specialists: Interviewer: How has it been dealing with the doctors up there (the HSC)? Remote Physician: Well, pretty good in general, and that’s an important key to the success of the program, because, you know, if the doctors make you look good, and let you learn something, then you’ll want to come back. But if they make you look bad, then you don’t want to do it again. And most of the doctors are really good about...you know, even if they feel like you’ve done exactly the wrong thing, they’ll say, “Yes, that was an excellent thought. And I would ask a couple of other things.” But there are some guys that are jerks over it. There was one infectious disease guy that, wooh! I mean, he would sit back and he would say, “What kind of bull****, what kind of [jerk], nah, nah, nah, nah....” And you could hear him off the screen. You know, I could hear him talking about this off [camera], you know, and...we’d just never consult him again. You’d say, “Never, ever get this guy again.” The interviews were also utilized to determine the type and level of trust involved in the different telemedicine relationships. For example, the pediatric oncologists had a very negative assessment of the ability of the local state hospital’s nurses involved in the teleconsultation project. They felt the nurses did not have the training to perform the necessary neurological and abdominal exams. HSC Pediatric Oncologist: In order for it [telemedicine] to work, a condition that is absolutely necessary is to have a qualified person that will link the patient and

the physician. And that person has to be a registered nurse practitioner or a physician’s assistant. Because you’re practicing medicine and seeing the patient and then treating the patient and [you are] responsible for what happens to the patient, then that person [on the remote end] needs to do a good physical examination with an abdominal exam, heart sounds, and everything. They need to relate to you what is happening to the patient….They [physicians’ assistants and nurse practitioners] are fully capable. They have done thousands of physical examinations; they know what’s normal, so whatever is abnormal they report. It just takes some training to— the hardest thing to do is a neurological exam and an abdominal [exam] probably. Interviewer: Because? HSC Pediatric Oncologist: Neurological exam takes some training, special training in trying to do it right and interpreting it right, and then the abdominal exam—you really have to put your hands in there and try to feel the liver and the spleen—and that takes awhile to develop the expertise. And if you miss it, it’s serious. So you have to have somebody well trained because if you are not able to find out that person [the patient] has that finding, it may be a serious problem of missing very important information. That’s why I think the link between the patient and [specialist] physician is absolutely necessary. You cannot practice telemedicine without that link, I think— practice it safely and without decreasing the standards of care, because if you have somebody who’s not capable and you see a patient with fever and you don’t do an exam and you don’t know that the patient has a large spleen, you see, you misdiagnose. You have half the information that you really have to have. That link is very important.

MIS Quarterly Vol. 28 No. 2/June 2004

201

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

In the case of the bone marrow transplant teleconsultation project, the rural oncologist originally was suspicious of motives of the the HSC’s subspecialists, but that concern dissipated and he now perceived himself as being a partner with instead of a lackey of the HSC’s subspecialists. Rural Oncologist: There’s been—like I said at the beginning—so many people asked me about that you feel like you’re kind of being like a lackey or something—you know just like a student rather than a doctor, not playing a major role and I think that, now I feel like a total partner even though it’s a transplant, I feel like I’m doing as much as they are even though I’m not getting into the mechanics of it and I think that that’s how it’s evolved to that level and so at the beginning, I started having my own doubts about well I’m just letting these people come into our office and dictating the care of my patients. But it’s evolved so that, I didn’t feel that way at the beginning but so many people brought that up that I was beginning to wonder and now I feel just the opposite, that I have a lot more control….I think that because of the communication that’s developed and it’s more than just having the telemedicine, it’s talking to these doctors and gaining confidence both ways that we can communicate well, that we respect each other’s ideas, we have a feeling of how things are done and they’re more comfortable about sending the patients back. Key informant interviews were also used to assess the impact the telemedicine project had on the quality of and access to remote site health care. A physician at a rural hospital involved in a multiple specialties teleconsultation project talked about how teleconsultations improved both the quality and access to care. Prior to telemedicine, the rural physicians were on their own when it came to difficult or rare cases because the local population was too poor to afford to travel to the HSC or its closer affiliated center.

202

MIS Quarterly Vol. 28 No. 2/June 2004

Interviewer: How were some of these cases handled prior to telemedicine? Remote Physician: We just took a wild guess. We did, because a lot of these folks can’t get up there. You know, you can try calling, but it’s not the same thing as looking at it, and so a lot of it was just....You just used your best judgment and went on... Interviewer: They couldn’t get there in terms of… Remote Physician: Economics…normally, it’s economics. Physical condition occasionally, but mostly, economics. Most of the people in [rural area] do not have insurance. That’s an economic fact of life. Interviewer: Medicaid?

Are they covered by

Remote Physician: Doesn’t make any difference if you have Traveler’s Insurance. If you don’t have enough money to put [gas] in your gas tank, you can’t get to the doctor. That’s the common problem. Key informant interviews were also used to assess the impact the telemedicine project had on the cost of remote site health care. For example, the bone marrow transplant teleconsultation project allowed patients to return home earlier. Prior to the teleconsultation project, the rural oncologist’s patients who underwent the bone marrow transplant procedure were severely immuno-compromised and had to spend 6 to 10 weeks living in short-term apartments near the HSC in order to undergo blood transfusions or other types of therapy on an outpatient basis. With the advent of the teleconsultation project, these patients (and members of their family who often stayed with them at the HSC apartments) could now return home after three weeks in the hospital and receive much of their post-transplant support at the oncologists’ clinic. As a result, the patients and

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

their insurance companies were spared the cost of the short-stay apartment rental, and the teleconsultation project enabled the patient’s friends or relatives to continue working on at least a parttime basis while the patient underwent posttransplant activities at the local oncologist’s office. Rural Oncologist: They [the patient’s family] get home. They don’t have to pay for a hotel room. It’s no little thing that a lot of them can still work here and keep their health insurance or whatever. So it’s kind of hard to explain to people but I think most people that have a family—if I think what would happen if I had to go spend a week up in [HSC city] to my family and be away from my job. It would hit home and then these are people not even with illness. There are a lot of social and economic implications.

Quantitative Results POSAC analysis was utilized to assess the relationship between individual profiles (the telemedicine relationships)—represented by their interpersonal trust and agency facet values acting as input variables—and the perceived impact on remote site health care delivery facet element values acting as an output variable. An advantage of nonparametric MDS is that its analysis with as few as seven data points can be very robust (Borg and Lingoes 1987; Shye 1985a, 1985b), and this data set consisted of 20 telemedicine relationships where 12 of the 20 (60 percent) were perceived as positively impacting remote site health care delivery. Two tables and one figure are presented for each hypothesis. The first table presents the POSAC coordinates for the relevant trust and agency elements’ values for each profile, and the CORREP value for the model, and the second table presents the weak monotonicity coefficient values for the different element values and the X and Y axes. The figure presents the different telemedicine relationships item diagram based on the relevant trust and agency content structuple

values, with the impact on remote site health care delivery content structuple value superimposed on the different telemedicine relationships profiles. Those relationships whose perceived impact value is consistent with the hypothesized relationship with type of trust being tested are in plain text, while those whose are not are bolded and italicized.

H1: Calculative Trust and VCR Performance The POSAC diagram correctly represented 98 percent of the profile pairs. The HSC weak monotonicity coefficient values in Table 5 indicate that the X axis represents the self-interest element value, while the Y axis represents the agency facet element values. The perceived impact’s weak monotonicity coefficient values with the joint axis (0.801) and lateral axis (-0.015) approached one and zero, respectively, and provided support for Hypothesis 1. The item diagram in Figure 7 also supported the hypothesis in that it was partitioned in a manner that it discriminated between those telemedicine relationships positively impacting remote site health care delivery and those who do not at a rate better than chance. In all, 15 of the 20 (75 percent) of the telemedicine relationships were correctly discriminated by the optimal partitioning of the external item diagram. All 12 (100 percent) of the telemedicine relationships having a positive impact were correctly discriminated by the partitioning of the diagram, and 3 of the 8 (37.5 percent) telemedicine relationships not having a positive impact were correctly discriminated by the partitioning. H1 was therefore supported, and the impact on remote site health care delivery of virtual collaborative relationships is monotonically associated with calculative interpersonal trust.

H2: Competence Trust and VCR Performance The POSAC diagram correctly represented 100 percent of the profile pairs. As indicated in Table 7 the weak monotonicity coefficient values of the HSC indicates that the X axis represents the agency facet element values, while the Y axis re-

MIS Quarterly Vol. 28 No. 2/June 2004

203

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

1.0 R [7] 19, 21, 21, 21

R [6] 12 R [5] 16, 21

0.8

R [3] 12 HSC [7] 12, 12, 16, 19, 21, 21, 21

R [2] 11, 12

Agency

0.6

HSC [1] 12, 21 0.4 HSC [6] 11

0.2

0.0 0.0

0.2

0.4

0.6 Self-Interest

0.8

1.0

Figure 7. External Item Diagram of VCR Performance for Calculative Interpersonal Trust (H1)

Table 6. POSAC Coordinates and Coefficients for Competence Trust (H2) Telemedicine Relationship H, B, J, T, P, F, D, L, N, R

Agency Ability Remote

7

X

Y

0.408

0.913

Joint 0.661

Lateral 0.248

Impact 11, 12, 12, 12, 16, 19, 21, 21, 21, 21

E, C, K, M, Q

HSC

7

0.913

0.816

0.865

0.548

19, 21, 21, 21, 21

O, S

HSC

4

0.816

0.707

0.762

0.555

12, 16

A, I

HSC

2

0.707

0.577

0.642

0.565

12, 12

G

HSC

1

0.577

0.408

0.493

0.585

11

Proportion of profile pairs correctly represented (CORREP): 1.000

Table 7. H2: Competence Trust Weak Monotonicity Coefficient Matrix X

Y

IMPACT

0.3789

ABILITY HSC

204

JOINT

LATERAL

0.6728

0.8424

-0.0551

-0.4278

1.0000

0.7383

-1.0000

1.0000

-1.0000

0.7115

1.0000

MIS Quarterly Vol. 28 No. 2/June 2004

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

1.0

R, [7] 11, 12, 12, 12, 16, 19, 21, 21, 21, 21

0.8

HSC [7] 19, 21, 21, 21, 21 HSC [4] 12, 16

0.6 Ability

HSC [2] 12, 12

0.4

HSC [1] 11

0.2

0.0 111 0.0

0.2

0.4

0.6 Agency

0.8

1.0

Figure 8. External Item Diagram of VCR Performance for Competence Trust (H2)

presents the ability element. The perceived impact’s weak monotonicity coefficient values with the joint axis (0.842) and lateral axis (-0.055) approached one and zero, respectively, and provided support for Hypothesis 2. The partitioning of the item diagram space also provided support for the hypothesis. As indicated in Figure 8, the item diagram was partitioned in a manner that it discriminated between those telemedicine relationships having a positive impact on the remote site health care delivery and those who did not at a rate better than chance. In all, 15 of the 20 (75 percent) of the telemedicine relationships were correctly discriminated by the optimal partitioning of the external item diagram. All 12 (100 percent) of the telemedicine relationships having a positive impact were correctly discriminated by the partitioning of the diagram, and 3 of the 8 (37.5 percent) telemedicine relationships not having a positive impact were correctly discriminated by the partitioning. H2 was therefore supported.

H3: Relational Trust and VCR Performance The POSAC diagram correctly represented 80.1 percent of the profile pairs. As exhibited in Table 9, the weak monotonicity coefficient values indicate that the X axis represents the empathy facet element values, while the Y axis represents the agency element. The perceived impact’s weak monotonicity coefficient value with the joint axis of 0.643 was positive but only moderately approached one, and its value with the lateral axis of 0.230 only moderately approached zero. This provided only modest support for the hypothesis. However, the partitioning of the item diagram space provided additional support for the hypothesis. As indicated in Figure 9, the item diagram was partitioned in a manner that it discriminated between those telemedicine relationships having a positive impact on the remote site health care delivery and those that did not at a rate better than chance. In all, 16 of the 20 (80 percent) of the

MIS Quarterly Vol. 28 No. 2/June 2004

205

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

Table 8. POSAC Coordinates and Coefficients for Relational Trust (H3) Telemedicine Relationship

Agency

Empathy

X

B, F, L, N, R

Remote

7

0.745

0.882

0.814

0.432

12, 19, 21, 21, 21

C, K, M, Q

HSC

7

0.882

0.577

0.730

0.652

21, 21, 21, 21

I

HSC

6

0.943

0.471

0.707

0.736

12

P

Remote

6

0.667

0.943

0.805

0.362

16

T

Remote

5

0.471

0.816

0.644

0.327

12

D, H, J

Remote

4

0.333

0.745

0.539

0.294

11, 12, 21

Y

Joint

Lateral

Impact

E, G, O, S

HSC

4

0.816

0.667

0.742

0.575

11, 12, 16, 19

A

HSC

1

0.577

0.333

0.455

0.622

12

Proportion of profile pairs correctly represented (CORREP): 0.801

Table 9. H3 Relational Interpersonal Trust Weak Monotonicity Coefficient Matrix X

Y

JOINT

IMPACT

0.5245

0.2312

0.6432

0.2302

EMPATHY

0.7477

0.5591

0.9458

0.1578

HSC

0.9294

-1.0000

-0.0666

1.0000

telemedicine relationships were correctly discriminated by the optimal partitioning of the external item diagram. A total of 11 of the 12 (91.7 percent) telemedicine relationships having a positive impact were correctly discriminated by the partitioning of the diagram, and 5 of the 8 (62.5 percent) telemedicine relationships not having a positive impact were correctly discriminated by the partitioning. H3 was therefore supported.

H4: Integrated Trust and VCR Performance The POSAC diagram correctly represented 99.1 percent of the profile pairs. As indicated in Table 11, the weak monotonicity coefficient values of the HSC indicates that the X axis represents the Agency facet element values, while the Y axis represents the ability element. Self-interest plays an accentuating role, as illustrated by the inverted L shaped dotted line starting at approximately

206

MIS Quarterly Vol. 28 No. 2/June 2004

LATERAL

0.85 on the Y axis. Empathy plays an attenuating role as illustrated by the L shaped dashed lines starting at approximately 0.21 and 0.45 on the X axis which partition the POSAC solution space. The perceived impact’s weak monotonicity coefficient value with the joint axis of 0.977 approached one, and its value with the lateral axis of -0.109 approached zero. This provided strong support for the hypothesis. The partitioning of the item diagram space based also provided strong support for the hypothesis. As indicated in Figure 10, the item diagram was partitioned in a manner that it discriminated between those telemedicine relationships having a positive impact on the remote site health care delivery and those that did not at a rate better than chance. In all, 19 of the 20 (95 percent) of the telemedicine relationships were correctly discriminated by the optimal partitioning of the external item diagram. A total of 11 of the 12 (91.7 percent) telemedicine relationships having a positive impact were correctly dis-

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

1.0 R [6] 16 R [7] 12, 19, 21, 21, 21

R [5] 12 0.8

HSC [4] 11, 12, 16, 19

R [4] 11, 12, 21 0.6 Agency

HSC [7] 21, 21, 21, 21

HSC [6] 12

0.4 HSC [1] 12

0.2

111

0.0 0.0

0.2

0.4

0.6 Empathy

0.8

1.0

Figure 9. External Item Diagram of VCR Performance for Relational Trust (H3)

criminated by the partitioning of the diagram, and all 8 (100 percent) telemedicine relationships not having a positive impact were correctly discriminated by the partitioning. H4 was therefore strongly supported.

Discussion This article examined the association of interpersonal trust with virtual collaborative relationship performance. Trust enables collaboration as a means of complexity reduction, and virtual collaboration extends face-to-face collaborative relationships by substituting technology for collocation. When virtual collaborative relationships fail, technology is often blamed while in many cases it is trust, or the lack thereof, that is at fault. Four types of interpersonal trust—calculative, competence, relational, and integrated—were

identified and each was operationalized as a single construct. All four hypotheses were supported. The impact on remote site health care delivery of VCRs was monotonically associated with each of the individual types of trust and integrated trust. However, it was the fourth hypothesis—an association between integrated trust and VCR performance—that provided the most interesting findings in terms of the roles the different types of trust played and how they were interrelated. The ability element can be interpreted that if one party’s assessment of the other party’s competency is negative, then the VCR will have no impact on remote site health care delivery. In each of the three cases where ability was negative, the VCRs had no impact on remote site health care delivery. Therefore, a nonnegative assessment of the other party’s competence is a necessary but not sufficient condition in order for a VCR to have a positive impact on remote site health care delivery.

MIS Quarterly Vol. 28 No. 2/June 2004

207

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

Table 10. POSAC Coordinates and Coefficients for Integrated Trust (H4) Telemedicine Relationship

Agency

Ability, Empathy, Self-interest

X

Y

Joint

Lateral

F, N, R

Remote

777

0.686

0.970

0.828

0.358

19, 21, 21

C, K, M

HSC

777

0.970

0.686

0.828

0.642

21, 21, 21

L

Remote

775

0.642

0.874

0.758

0.384

21

E

HSC

747

0.874

0.594

0.734

0.640

19

D

Remote

747

0.420

0.907

0.664

0.256

21

P

Remote

765

0.594

0.804

0.699

0.395

16

Impact

J

Remote

746

0.343

0.939

0.641

0.202

12

B

Remote

772

0.485

0.840

0.663

0.322

12

O

HSC

447

0.907

0.542

0.725

0.683

16

Q

HSC

771

0.767

0.642

0.704

0.563

21

I

HSC

267

0.939

0.420

0.680

0.760

12

T

Remote

753

0.542

0.767

0.655

0.388

12

H

Remote

742

0.243

0.728

0.485

0.257

11

G

HSC

146

0.804

0.343

0.574

0.731

11

A

HSC

217

0.840

0.243

0.541

0.799

12

S

HSC

441

0.728

0.485

0.606

0.621

12

Proportion of profile pairs correctly represented (CORREP): 0.991

Table 11. H4: Integrated Trust Weak Monotonicity Coefficient Matrix X IMPACT

0.5116

ABILITY

Y

JOINT

LATERAL

0.6697

0.9770

-0.1088

-0.7589

0.9986

0.8133

-0.9933

EMPATHY

0.2836

0.7539

0.9545

-0.3344

SELFINT

0.6900

0.0488

0.7562

0.4170

HSC

1.0000

-1.0000

-0.0009

1.0000

208

MIS Quarterly Vol. 28 No. 2/June 2004

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

1.0

R [777] 19, 21, 21

R [746] 12 R [747] 21

R [775] 21

R [772] 12

0.8

R [742] 11

R [765] 16 HSC [777]

R [753] 12

Positive Empathy 21, 21, 21

Positive Ability

HSC [771] 21 HSC [747] 19 HSC [447] 16 HSC [441] 12

Neutral Ability

HSC [267] 12

Self--interest Negative Self

0.4

Neutral Empathy 0.2

Negative Ability

Negative Empathy 0.0 0.0

0.2

0.4

Agency

0.6

Self--interest Positive Self

Ability

0.6

HSC [146] 11 HSC [217] 12

0.8

1.0

Figure 10. External Item Diagram of VCR Performance for Integrated Trust (H4)

Empathy played an attenuating role. Without a positive ability assessment, positive empathy is not likely. It appears that if ability and self-interest values are split—that is, one type of trust is positive and the other is negative—then, consistent with its attenuating role, empathy doesn’t matter. There are not enough data points to draw a definitive conclusion, but we speculate that if empathy is neutral, it acts as a drag on the impact of those collaborative relationships. Self-interest played an accentuating role. It appears that self-interest cannot offset negative assessments of the other party’s ability or empathy, but it can be the deciding factor when ability and empathy are neutral or positive. The integrated trust model provided interesting insights into the interrelationships of the different

types of trust on VCR performance. There was a strong relationship between all three types of trust being positive and VCR performance being strongly positive as well. In all eight cases where one party’s assessment of the three types of trust was all positive, the virtual collaborative relationship had a strongly positive impact on remote site health care delivery. This relationship generally held when all the types of trust were either neutral or positive. In 11 out of the 12 cases where all the types of trust were either neutral or positive, then impact was positive as well. There was also a strong relationship between trust and performance when at least one type of trust was negative. In seven out of the eight cases where at least one type of trust was negative, there was no impact on remote site health care delivery.

MIS Quarterly Vol. 28 No. 2/June 2004

209

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

In the other cases where the three types of trust were neutral or positive, it appeared that calculative trust generally dominated, although in some cases, relational trust could overcome neutral calculative trust values. In five of the six cases where one party’s assessment of the other party’s ability was either neutral or positive, and the assessment of the other party’s empathy was neutral, self-interest dominated. In two of these six cases where self-interest was neutral, there was no impact on remote site health care delivery. In three of these six cases where ability was either neutral or positive, empathy was neutral, and selfinterest was positive, then impact was positive as well. In these six cases, there was only one case where the assessment of ability was positive, empathy was neutral, self-interest was positive, and there was no impact on remote site health care delivery. It seems that in this case, while it was in one party’s self-interest to participate in this telemedicine relationship, it was not strong enough to warrant participation. In two of the three cases where ability and empathy were positive but self-interest was neutral, the virtual collaborative relationship had no impact on remote site health care delivery. This meant that in the one case where impact was positive, a positive empathy value dominated a neutral self-interest value. Therefore, one party chose to participate in a telemedicine relationship despite it not being in their own interest because they identified with the other party.

Contributions to Research and Practice The results of this article extended prior studies by finding that interpersonal trust is a primary determinant as to whether VCRs can address complex situations. The results presented here confirm, extend, and apparently contradict prior studies of interpersonal trust. Our finding of an association between performance and relational trust was consistent with and confirmed McAllister’s (1995) finding of only partial

210

MIS Quarterly Vol. 28 No. 2/June 2004

support for a relationship between affect-based interpersonal trust and performance. We also found support for an association between calculative trust and performance. Our research extended prior research by examining the roles the different types of trust play, and provided strong support for the hypothesis that there is a positive association between integrated interpersonal trust and VCR performance. A neutral or positive assessment of the other party’s competence is a necessary but not sufficient condition if a VCR’s performance is to be positive. Calculative trust plays an accentuating role, implying that it tends to sharpen the differentiation delineated by competence trust. In contrast, relational trust plays an attenuating role where it tends to temper the differentiation delineated by the other types of trust. Our research also extended prior research by examining how the different types of trust interact and are interrelated. Our findings indicate that positive assessments of all three types of trust are necessary if VCRs are to have a strongly positive performance. However, positive performance is still possible if all three types of trust are nonnegative; that is, if at least one type of trust is positive and the others are neutral, then VCR performance is likely to be positive as well, but not as positive as if each of the three types of trust are all positive. Our findings also establish that if any one type of trust is negative, then it is very likely that VCR performance will not be positive. We also found that the different types of trust are interrelated. Self-interest dominates when ability is either neutral or positive, and empathy is neutral—as long as one party perceives that it is strongly enough in their own interest to participate. Empathy may dominate self-interest in cases where ability is positive but self-interest is neutral. Our findings indicate that it is not one particular type of interpersonal trust that is critical for successful VCRs; rather, it is the integrated types of interpersonal trust that matter. The different types of interpersonal trust interact and are interdependent, and these interrelationships and interdependencies must be taken into account by

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

researchers as they attempt to understand the impact of trust on VCR performance. It also means that managers of virtual collaborative processes in both the project design phase and the management of the project must not depend on a single type of interpersonal trust to enhance their adoption; instead, they should pay attention to all of the types of interpersonal trust examined in this research and realize that they are related and interdependent. Our findings also apparently contradict prior research on integrated interpersonal trust and performance. Unlike Zaheer et al. (1998), we found a positive association between integrated interpersonal trust and performance. We believe there is a methodological explanation for our contradictory findings. From a data analysis perspective, Zaheer et al. used structured equation modeling in their study, while we used facet theory. The data assumptions in facet theory are less restrictive than those of structured equation modeling in that the ordinal data are sufficient and the relationships between variables need not be linear (Borg and Shye 1995; Shenkar et al. 1995; Shye et al. 1994), and are consistent with many aspects of human behavior (Guttman 1944; Levy and Guttman 1985). Consistent with prior applications of facet theory, the result here has been to find relationships between variables that other, more restrictive methods have been unable to find (Canter 1985b; Shye et al. 1994). Thus, our contradictory findings may well be the result of the different analysis techniques utilized. This study also makes a methodological contribution by introducing facet theory to information systems research. The study demonstrated how facet theory can be used in the study of complex social systems (both in the research design and data analysis phases), and how it has the potential to address many of the concerns and challenges information systems researchers face in performing field research. This article also makes significant contributions to practice. For those organizations considering deploying VCRs, this research demonstrates interpersonal trust factors must be addressed if

the benefits of these newer work relationships are to be reaped. Further, it is the integrated interpersonal trust factors that must be taken into account. This research also benefits those health care delivery organizations utilizing or considering utilizing VCRs as a means of more effectively and efficiently deploying health care resources by identifying the social context in which such telemedicine projects exist as a major contributing factor in determining project impact. In doing so, it suggests new considerations in the ongoing concerns about the failure of more extensive utilization and adoption of telemedicine in general. This article has implications for information technology professionals as well. Information technology professionals need to be less obsessed with the technology and the optimal configuration to support virtual collaboration activities, and instead focus more on the social context in which the technologies exist. In doing so, information technology professionals must also understand that it is integrated interpersonal trust, and not any one specific type of interpersonal trust, that matters.

Limitations and Suggestions for Future Research This article makes important contributions to the understanding of interpersonal trust and its association with VCR performance. However, trust research is sensitive to the context in which it occurs (Rousseau et al. 1998), and this research is subject to the same concerns about generalizability as is all trust research. The use of facet theory design techniques facilitates replicating these findings in other contexts. It may seem as though different types of telemedicine activities require different degrees of trust, and therefore the association between interpersonal trust and virtual collaborative relationship performance may be moderated by the type of activities in which the VCRs engaged. The data indicate otherwise—that the type of telemedicine activities in which the different relationships were engaged did not impact the results. Additionally,

MIS Quarterly Vol. 28 No. 2/June 2004

211

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

of the telemedicine relationships, 60 percent were perceived as having a positive impact, while 40 percent were not; of the teleconsultations, 60 percent were perceived as having a positive impact while 40 percent were not; and of the distance learning projects, 67 percent were perceived as having a positive impact, while 33 percent were not. Finally, of the teleradiology relationships, 50 percent were found to have a positive impact, while 50 percent were not. The findings that 50 percent of the teleradiology relationships were not perceived as having a positive impact was particularly supportive of the proposition that the type of telemedicine activities in which the relationship was engaged does not affect whether the relationship positively impacted remote site health care delivery. Teleradiology is not very sophisticated from a technology perspective because it involves asynchronous file transfer. Further, digital radiography is well accepted in the radiology profession, and teleradiology is one of the few telemedicine activities for which the specialists were reimbursed for participating. Finally, teleradiology involves the fewest process changes from the perspective of the radiologist, the primary care provider, and the patient. These facts suggest that if there were a telemedicine activity for which interpersonal trust would matter the least, it would be teleradiology. Yet one-half of the teleradiology projects did not have a positive impact because of interpersonal trust values. Trust is a messy concept and there are strong theoretical and data analytical reasons why integrated trust should be included in the model. Numerous other researchers have theorized about the concept of integrated trust. From a data analysis perspective, a strength of facet theory is that it can assess the interaction effect of different variables by determining whether a facet or element plays an attenuating or accentuating role. However, the integrated perspective of interpersonal trust is called integrated trust, but it may well be a perspective of how these different types of trust combine. Future research needs to determine whether integrative trust is a new kind of trust or just a mixture of the other types of trust.

212

MIS Quarterly Vol. 28 No. 2/June 2004

Telemedicine is not a new type of medicine; rather, it is a different way of providing existing medical care. Future research might investigate the impact telemedicine and trust relationships have on conventional health care delivery. Future research needs to address how the different types of interpersonal trust interact and the temporal relationships between these types of trust. Our findings are consistent with others’ proposition that calculative trust is needed if relational trust is to develop (Child 1998; Lewicki and Bunker 1996; Rousseau et al. 1998) and a collaborative relationship is to have a positive impact, but the timeline of when one type of trust may or may not be more important was not studied. Future research is needed to better understand these interactions and the temporal factors involved.

References Allen, A., and Stein, S. “Cost Effectiveness of Telemedicine,” Telemedicine Today (6:5), 1998, pp. 10-15. Anderson, E., and Weitz, B. “Determinants of Continuity in Converntional Industrial Channel Dyads,” Marketing Science (8:4), 1989, pp. 310-324. Bachmann, R. “Conclusion: Trust—Conceptual Aspects of a Complex Phenomenon,” in Trust Within and Between Organizations: Conceptual Issues and Empirical Applications, C. Lane and R. Bachmann (Eds.), Oxford University Press, Oxford , 1998, pp. 298-322. Bigley, G. A., and Pearce, J. L. “Straining for Shared Meaning in Organizational Science: Problems of Trust and Distrust,” Academy of Management Review (23:3), 1998, pp. 405421. Borg, I., and Groenen, P. Modern Multidimensional Scaling: Theory and Applications, Springer-Verlag, New York, 1997. Borg, I., and Lingoes, J. C. Multidimensional Similarity Structure Analysis, Springer, New York, 1987. Borg, I., and Shye, S. Facet Theory: Form and Content, Sage, Thousand Oaks, CA, 1995. Brown, J. “An Introduction to the Uses of Facet Theory,” in Facet Theory: Approaches to

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

Social Research, D. Canter (Ed.) Springer-Verlag, New York, 1985, pp. 17-58. Brown, J., and Barnett, J. “Facet Theory: An Approach to Research,” in Research Methods in Psychology, G. M. Breakwell, S. Hammond, and C. Fife-Schaw (Eds.), Sage, Thousand Oaks, CA:, 2000, pp. 105-118. Butler, J. K. “Toward Understanding and Measuring Conditions of Trust: Evolution of a Conditions of Trust Inventory,” Journal of Management (17:3), 1991, pp. 643-663. Butler, J. K., and Cantrell, R. S. “A Behavioral Decision Theory Approach to Modeling Dyadic Trust in Superiors and Subordinates,” Psychological-Reports (55:1), 1984, pp. 19-28. Canter, D. Facet Theory: Approaches to Social Research, Springer-Verlag, New York, 1985a. Canter, D. “How to Be a Facet Researcher,” in Facet Theory: Approaches to Social Research, D. Canter (Ed.) Springer-Verlag, New York, 1985b, pp. 265-276. Cattell, R. B. “The Scree Test for the Number of Factors,” Multivariate Behavioral Research (1:2), 1966, pp. 140-161. Census Bureau. “1990 Census of Population and Housing, Summary Tape File 3a,” United States Census Bureau, 1990 (available onine at http://venus.census.gov/cdrom/lookup; accessed March 3, 1997). Child, J. “Trust and International Strategic Alliances: The Case of Sino-Foreign Joint Ventures,” in Trust Within and Between Organizations: Conceptual Issues and Empirical Applications, C. Lane and R. Bachmann (Eds.), Oxford University Press, Oxford, 1998, pp. 241272. COC “Highway To Health: Transforming US Health Care in the Information Age,” Council on Competitiveness, Washington, DC, 1996. Collins, T., and Bicknell, D. CRASH: Learning from the World’s Worst Computer Disasters, Simon and Schuster, London, 1998. Debakey, M. E. “Telemedicine Has Now Come of Age,” Telemedicine Journal (1:1), 1995, pp. 34. Deutsch, M. “The Effect of Motivational Orientation upon Trust and Suspicion,” Human Relations (13), 1960, pp. 123-140. Deutsch, M. “Trust and Suspicion,” Conflict Resolution (2:4), 1958, pp. 265-279.

DOC. “Telemedicine Report to Congress,” United States Department of Commerce, 1997 (available online at http://www.ntia.doc.gov/ reports/telemed/, accessed February 18, 1997). Edmundson, E. W., Koch, W. R., and Silverman, S. “A Facet Analysis Approach to Content and Construct Validity,” Educational and Psychological Measurement (53:2), 1993, pp. 351-369. Eisenhardt, K. “Building Theories from Case Study Research,” Academy of Management Review (14), 1989, pp. 532-550. Flaherty, R. D., Perednia, D. A., Singer, T., and Orbuch, P. The Western Governors’ Association Telemedicine Action Report, Western Governors’ Association, Denver, CO, 1995. Gambetta, D. “Can We Trust Trust?,” in Trust: Making and Breaking Cooperative Relations, D. Gambetta (Ed.) Basil Blackwell, New York, 1988, pp. 213-238. GAO. Telemedicine: Federal Strategy Is Needed to Guide Investments, United States General Accounting Office, United States Senate, Washington, DC, 1997. Gefen, D. “E-Commerce: The Role of Familiarity and Trust,” Omega (28), 2000, pp. 725-737. George, J. M., and Jones, G. R. “The Experience and Evolution of Trust: Implications for Cooperation and Teamwork,” Academy of Management Review (23:3), 1998, pp. 531-546. Ghoshal, S., and Moran, P. “Bad for Practice: A Critique of the Transaction Cost Theory,” Academy of Management Review (21:1), 1996, pp. 35-48. Grigsby, J. “Current Status of Domestic Telemedicine,” Journal of Medical Systems (19:1), 1995, pp. 19-27. Guttman, L. “A Basis for Scaling Qualitative Data,” American Sociological Review (9), 1944, pp. 139-150. Guttman, L. “Facet Theory, Smallest Space Analysis, and Factor Analysis,” Perceptual and Motor Skills (54), 1982a, pp. 491-493. Guttman, L. “A General Nonmetric Technique for Finding the Smallest Coordinate Space for a Configuration of Points,” Psychometrika (33:4), 1968, pp. 469-506. Guttman, L. “A Structural Theory for Intergroup Beliefs and Actions,” American Sociological Review (24), 1959, pp. 318-328.

MIS Quarterly Vol. 28 No. 2/June 2004

213

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

Guttman, L. “‘What Is Not What’ in Theory Construction,” in Social Structure and Behavior, R. M. Hauser, D. Mechanic, and A. Haller (Eds.), Academic Press, New York, 1982b, pp. 331-348. Guttman, R., and Greenbaum, C. W. “Facet Theory: Its Development and Current Status,” European Psychologist (3:1), 1998, pp. 13-36. Hagen, J. M., and Choe, S. “Trust in Japanese Interfirm Relations: Institutional Sanctions Matter,” Academy of Management Review (23:3), 1998, pp. 601-620. Handy, C. “Trust and the Virtual Organization,” Harvard Business Review (73), 1995, pp. 4050. IOM. Access To Health Care In America, Institute of Medicine, National Academy Press, Washington, DC, 1993. IOM. Clinical Practice Guidelines: Directions for a New Program, Institute of Medicine, National Academy Press, Washington, DC, 1990. IOM. Primary Care: America’s Health in an New Era, Institute of Medicine, National Academy Press, Washington, DC, 1996a. IOM. Telemedicine: A Guide to Assessing Telecommunications in Health Care, Institute of Medicine, National Academy Press, Washington, DC, 1996b. Jarvenpaa, S. L., Knoll, K., and Leidner, D. E. “Is Anybody Out There? Antecedents of Trust in Global Virtual Teams,” Journal of Management Information Systems (14:4), 1998, pp. 29-64. Jarvenpaa, S. L., and Leidner, D. E. “Communication and Trust in Global Virtual Teams,” Organizational Science (10:6), 1999, pp. 791815. Kelle, U. Computer-Aided Qualitative Data Analysis: Theory Methods and Practice, Sage Publications, London, 1995. Kramer, R. M. “Trust and Distrust in Organizations: Emerging Perspectives, Enduring Questions,” Annual Review of Psychology (50), 1999, pp. 569-598. Lane, C. “Introduction: Theories and Issues in the Study of Trust,” in Trust Within and Between Organizations: Conceptual Issues and Empirical Applications, C. Lane and R. Bachmann (Eds.), Oxford University Press, Oxford, 1998, pp. 1-30.

214

MIS Quarterly Vol. 28 No. 2/June 2004

Latane, B., Liu, J. H., Nowak, A., Bonevento, M., and Zheng, L. “Distance Matters: Physical Space and Social Impact,” Personality and Social Psychology Bulletin (21:8), 1995, pp. 795-895. Levy, S. Louis Guttman on Theory and Methodology: Selected Writings, Dartmouth Publishing, Brookfield, VT, 1994. Levy, S., and Guttman, L. “The Partial-Order of Severity of Thyroid Cancer with the Prognosis of Survival,” in Analysis in Real Life Environment: Ins and Outs of Solving Problems, J. F. Marcotorchino, J. M. Proth, and J. Janssen (Eds.), Elsevier Science, Amsterdam, 1985, pp. 111-119. Lewicki, R. J., and Bunker, B. B. “Developing and Maintaining Trust in Work Relationships,” in Trust in Organizations: Frontiers of Theory and Research, R. Kramer and T. Tyler (Eds.), Sage Publications, Thousand Oaks, CA, 1996, pp. 114-139. Lewicki, R. J., McAllister, D. J., and Bies, R. J. “Trust and Distrust: New Relationships and Realities,” Academy of Management Review (23:3), 1998, pp. 438-458. Lewis, J. D., and Weigert, A. “Trust as a Social Reality,” Social Forces (63:4), 1985, pp. 967985. Luhmann, N. Trust and Power, Wiley, Chichester, UK, 1979. Luhmann, N. “Familiarity, Confidence, Trust: Problems, and Alternatives,” in Trust: Making and Breaking Cooperative Relations, D. Gambetta (Ed.) Basil Blackwell, New York, 1988, pp. 94-108. March, J., and Olsen, J. Rediscovering Institutions: The Organizational Basis for Politics, Free Press, New York, 1989. Martin, P. Y., and Turner, B. A. “Grounded Theory and Organizational Research,” Journal of Applied Behavioral Science (22:2), 1986, pp. 141-157. Mason, R. O., and Mitroff, I. I. “A Program for Research on Management Information Systems,” Management Science (19:5), 1973, pp. 475-487. Mayer, R. C., Davis, S., and Schoorman, F. D. “An Integrative Model of Organizational Trust,” Academy of Management Review (20), 1995, pp. 709-734.

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

McAllister, D. J. “Affect - and Cognition-Based Trust as Foundations for Interpersonl Cooperation in Organizations,” Academy of Management Journal (38:1), 1995, pp. 24-59. McKnight, D. H., Cummings, L. L., and Chervany, N. L. “Initial Trust Formation in New Organizational Relationships,” Academy of Management Review (23:3), 1998, pp. 473-490. Meyerson, D., Weick, K. E., and Kramer, R. M. “Swift Trust and Temporary Groups,” in Trust in Organizations: Frontiers of Theory and Research, R. Kramer and T. Tyler (Eds.), Sage Publications, Thousand Oaks, CA, 1996, pp. 166-195. Mishra, A. K. “Organizational Responses to Crisis: The Centrality of Trust,” in Trust in Organizations: Frontiers of Theory and Research, R. Kramer and T. Tyler (Eds.), Sage Publications, Thousand Oaks, CA, 1996, pp. 261-287. Morris, R. “Computerized Content Analysis in Management Research: A Demonstration of Advantages and Limitations,” Journal of Management (20), 1994, pp. 903-932. Newell, S., and Swan, J. “Trust and InterOrganizational Networking,” Human Relations (53:10), 2000, pp. 1287-1328. Nohria, N., and Eccles, R. G. “Face-to-Face: Making Network Organizations Work,” in Networks and Organizations: Structure, Form, and Action, N. Nohria and R. G. Eccles (Eds.), Harvard Business School Press, Boston, 1992, pp. 288-308. OHRA. Exploratory Evaluation of Rural Applications of Telemedicine, Office of Rural Health Policy, United States Department of Health and Human Services, Rockville, MD, 1997. OTA. Bringing Health Care Online: The Role of Information Technologies, Office of Technology Assessment, United States Government Printing Office, Washington, DC, 1995. Orlikowski, W. J. “CASE Tools as Organizational Change: Investigating Incremental and Radical Changes in Systems Development,” MIS Quarterly (17:3), 1993, pp. 309-340. Paul, D. L., Pearlson, K. E., and McDaniel, R. R. “Technological Barriers to Telemedicine: Technology Management Implications,” IEEE Transactions on Engineering Management (6:2), 1999, pp. 279-288.

Ring, P. S. “Fragile and Resilient Trust and Their Roles in Economic Exchange,” Business & Society (35:2), 1996, Ring, P. S., and Van de Ven, A. H. “Developmental Processes of Cooperative Interorganizational Relationships,” Academy of Management Review (19:1), 1994, pp. 90-118. Rousseau, D. M., Sitkin, S. B., Burt, R. S., and Camerer, C. “Not So Different After All: A Cross-Discipline View of Trust,” Academy of Management Review:3), 1998, pp. 393-404. Sackmann, S. A. Cultural Knowledge in Organizations: Exploring the Collective Mind, Sage Publications, Newbury Park, CA, 1991. Sako, M. “Does Trust Improve Business Performance?,” in Trust Within and Between Organizations: Conceptual Issues and Empirical Applications, C. Lane and R. Bachmann (Eds.), Oxford University Press, Oxford, 1998, pp. 88-117. Sako, M. Prices, Quality, and Trust: Inter-firm Relations in Britain and Japan, Cambridge University Press, Cambridge, UK, 1992. Sako, M. “The Role of ‘Trust’ in Japanese BuyerSupplier Relationships,” Ricerche Economiche (45), 1991, pp. 449-474. Shapira, Z. “A Facet Analysis of Leadership Styles,” Journal of Applied Psychology (61), 1976, pp. 136-139. Shenkar, O., Aranya, N., and Almor, T. “Construct Dimensions in the Contingency Model: An Analysis Comparing Metric and Non-Metric Multivariate Instruments,” Human Relations (48), 1995, pp. 559-580. Shepard, B., and Tuchinsky, M. “Micro-OB and the Network Organization,” in Trust in Organizations: Frontiers of Theory and Research, R. Kramer and T. Tyler (Eds.), Sage Publications, Thousand Oaks, CA, 1996, pp. 140-165. Shepard, R. N. “The Analysis of Proximities: Multidimensional Scaling with an Unknown Distance Function. Part I,” Psychometrika (27:2), 1962a, pp. 125-140. Shepard, R. N. “The Analysis of Proximities: Multidimensional Scaling with an Unknown Distance Function. Part II,” Psychometrika (27:3), 1962b, pp. 219-246. Shye, S. “Modern Facet Theory: Content Design and Measurement in Behavioral Research,”

MIS Quarterly Vol. 28 No. 2/June 2004

215

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

European Journal of Psychological Assessment (14:2), 1998, pp. 160-171. Shye, S. Multiple Scaling: The Theory and Application of Partial Order Scalogram Analysis, North-Holland, Amsterdam, 1985a. Shye, S. “Nonmetric Multivariate Models for Behavioral Action Systems,” in Facet Theory: Approaches to Social Research, D. Canter (Ed.) Springer-Verlag, New York, 1985b, pp. 97-148. Shye, S., and Amar, R. “Partial-Order Scalogram Analysis by Base Coordinates and Lattice Mapping of the Items by Their Scalogram Scales,” in Facet Theory: Approaches to Social Research, D. Canter (Ed.) Springer Verlag, New York, 1985, pp. 277-298. Shye, S., Elizur, D., and Hoffman, M. Introduction to Facet Theory: Content Design and Intrinsic Data Analysis in Behavioral Research, Sage Publications, Thousand Oaks, CA, 1994. Silberman, C. E. “What Is it like to Be a Patient in the 1990s?,” in Technology and Health Care in an Era of Limits, Institute of Medicine, National Academy Press, Washington, DC, 1992, pp. 165-197. Staw, B. M., Sandelands, L. E., and Dutton, J. E. “Threat-Rigidity Effects in Organizational Behavior: A Multilevel Analysis,” Administrative Science Quarterly (26:4), 1981, pp. 501-524. Strauss, A., and Corbin, J. Basics of Qualitative Research: Grounded Theory Procedures and Techniques (2nd ed.), Sage Publications, Newbury Park, CA, 1998. Tyler, T., and Kramer, R. M. “Whither Trust?,” in Trust in Organizations: Frontiers of Theory and Research, R. Kramer and T. Tyler (Eds.), Sage Publications, Thousand Oaks, CA, 1996, pp. 115. Whitener, E. M., Brodt, S. E., Korsgaard, M. A., and Werner, J. M. “Managers as Initiators of Trust: An Exchange Relationship Framework for Understanding Managerial Trustworthy Behavior,” Academy of Management Review (23:3), 1998, pp. 513-531. Williamson, O. E. “Calculativeness, Trust, and Economic Organization,” Journal of Law and Economics (30), 1993, pp. 131-145. Williamson, O. E. The Economic Institutions of Capitalism, The Free Press, New York, 1985.

216

MIS Quarterly Vol. 28 No. 2/June 2004

Williamson, O. E. “The Economics of Organization: The Transaction Cost Approach,” American Journal of Sociology (87), 1981, pp. 548577. Wolfe, R., Gephart, R., and Johnson, T. “Computer Facilitated Qualitative Data Analysis: Potential Contributions to Management Research,” Journal of Management (19), 1993, pp. 637-660. Yin, R. K. Case Study Research: Design and Methods (2nd ed.), Sage Publications, Thousand Oaks, CA, 1994. Zaheer, A., McEvily, B., and Perrone, V. “Does Trust Matter? Exploring the Effects of Interorganizational and Interpersonal Trust on Performance,” Organization Science (9:2), 1998, pp. 141-159. Zand, D. E. “Trust and Managerial Problem Solving,” Administrative Science Quarterly (17), 1972, pp. 229-239. Zucker, L. G., Darby, M. R., Brewer, M. B., and Peng, Y. “Collaboration Structure and Information Dilemmas in Biotechnology: Organizational Boundaries as Trust Production,” in Trust in Organizations: Frontiers of Theory and Research, R. Kramer and T. Tyler (Eds.), Sage Publications, Thousand Oaks, CA, 1996, pp. 90-113.

About the Authors David L. Paul is an assistant professor in the Department of Information Technology and Electronic Commerce at the Daniels College of Business of the University of Denver. He received a B.S. from the Wharton School of the University of Pennsylvania, an M.B.A. from the Anderson School at the University of California, Los Angeles, and a Ph.D. from the Graduate School of Business at the University of Texas at Austin. Dr Paul’s research interests include telemedicine, virtual collaboration, trust, and complex adaptive systems. Dr. Paul has published articles in journals such as IEEE Transactions on Engineering Management, International Journal of Healthcare Technology and Management, and Computational and Mathematical Organizational Theory, and he

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

has presented papers at the Hawaii International Conference on System Sciences and numerous INFORMS Conferences. He was the recipient of the 2000 International Conference on Information Systems Best Dissertation Award. Reuben R. McDaniel, Jr. holds the Charles and Elizabeth Prothro Regents Chair in Health Care Management in the McCombs School of Business at The University of Texas at Austin and is a professor of Management Science and Information Systems. He received his Ed.D. from Indiana University, his M.S. from Akron University, and his

B.S. from Drexel University. His present research interest is in management of complex adaptive systems, with a particular interest in organizational designs and information systems for more effective sensemaking and decision-making in health care organizations. A partial list of journals he has published in includes The Academy of Management Journal, Decision Sciences, Health Care Management Review, Health Services Research, The Journal of Applied Behavioral Sciences, The Journal of the National Medical Association, Management Science, and Organizational Behavior and Human Decision Processes.

Appendix A Overview of Facet Theory This appendix presents an overview of facet theory in order to demonstrate how facet theory can address the concerns and challenges information systems researchers face in performing field research and why it was an appropriate methodology for this study. Facet theory is a systematic approach to facilitating and integrating research construction, design, and data analysis of complex social systems. It utilizes multidimensional data analysis guided by a theoretical framework (Borg and Shye 1995; Guttman and Greenbaum 1998; Shye 1998). Facet theory was initially developed by Louis Guttman, who defined theory (1982b, p. 335) as an “hypothesis of a correspondence between a definitional system for a universe of observations and an aspect of the empirical structure of those observations, together with a rationale for such a hypothesis.” Facet theory provides a system of concepts, definitions, and theorems on research design and data analysis (Borg and Shye 1995). The advantages of its research design concepts are that they provide techniques for defining the observations and constructing hypotheses that link features of the design with empirical aspects of the data.

Research Design Using Facet Theory Table A1 provides a glossary of facet theory terms. The mapping sentence is the primary facet theory tool in the design phase. It is a verbal statement used to construct the formal definitional framework for research design and theory testing. In the mapping sentence, the theoretical constructs of the research and the type of observations needed to test it are simultaneously identified and explicated, and it is from the mapping sentence that hypotheses are generated (Borg and Shye 1995; Brown 1985; Canter 1985b; Guttman and Greenbaum 1998).

MIS Quarterly Vol. 28 No. 2/June 2004

217

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

Table A1. Facet Theory Terminology Term

Definition

Accentuating

A facet or element in an accentuating role partitions the POSAC item diagram with an inverted “L” shaped line and the facet or element’s weak monotonicity coefficient values should approach zero for the X axis and one for the Y axis. A facet or element in an accentuating role is structurally dependent on the base-coordinate (X or Y) items and induces a finer division of the X and Y base coordinates by emphasizing or sharpening whatever trend is indicated by the base coordinates. The structural dependence is such that values of an accentuating item tend to increase with an increase in the compound values (either or both) of the base coordinates, and its role is to sharpen the differentiation delineated by the base coordinates (Borg and Shye 1995; Shye 1985a; Shye et al. 1994).

Attenuating

An facet or element in an attenuating role partitions the POSAC item diagram with an “L” shaped line and the facet or element’s weak monotonicity coefficient values should approach one for the X axis and zero for the Y axis. A facet or element in an attenuating role is structurally dependent on the base-coordinate (X or Y) items and induces a finer division of the X and Y base coordinates by moderating whatever trend is indicated by the base coordinates. The structural dependence is such that values of an attenuating item tend to increase with an increase in the concurrent simultaneous increase in values of both of the base coordinates, and its role is to moderate the differentiation delineated by the base coordinates (Borg and Shye 1995; Shye 1985a; Shye et al. 1994).

Coefficient of Alienation

The coefficient of alienation is a measure of the model’s goodness-of-fit (loss function) determined by the extent to which the distance between pairs of points and a two dimensional space adhere to the rule regarding the monotonic relationship between input coefficients and output distances. The coefficient of alienation value can be between 0 and 1, inclusive, where perfect fit is represented by a 0 value and the worst possible fit is given by the value 1. As a general rule, a coefficient of alienation value of less than 0.15 indicates a good fit (Brown 1995; Guttman 1968; Shapira 1976).

Comparable Profiles

Profiles (score structuples) are comparable if all the content structuple values of one profile are equal to or greater than the corresponding content structuple values of the other profile, and one profile is of a higher order than another if and only if it is higher on at least one item and not lower on any other items (Levy and Guttman 1985).

Content Facets

The “Q” in PQ!R, content facets represent the domain of interest of the research in that they are the attributes or conditions by which the population of interest is compared or classified (Borg and Shye 1995).

Content Structuple

A subset of a score structuple, a content structuple is a definitional structuple consisting of element values for some but not all of the different content facets (Shye et al. 1994).

CORREP

The coefficient of correct representation (CORREP) is a measure of the goodnessof-fit of the POSAC diagram. It specifies the proportion of score structuple pairs, weighted by their observed frequencies, whose comparability and incomparability relations, respectively, are correctly represented in a POSAC diagram. It can have values between 0 and 1, inclusive, with 1 representing a perfect representation (Shye and Amar 1985).

218

MIS Quarterly Vol. 28 No. 2/June 2004

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

Table A1. Facet Theory Terminology (Continued) Term

Definition

Elements

Elements define the universe of a facet by specifying the exact values subjects may be assigned within a particular facet. Elements within a facet are mutually exclusive and jointly exhaustive in that the elements in a particular facet ideally make up all of the possible values for that particular universe (Brown 1985; Edmundson et al. 1993; Shye et al. 1994).

Extension

The addition of elements to existing facets, usually as justified by empirical observation (Shye et al. 1994).

External Item

An external item is a classification of the content facets and their elements specified in terms of the content facets’ properties or behavior with respect to criterion external to those content facets (Borg and Shye 1995). The external item enables the comparison of an association between one or more of the internal content facets and an external concept.

External Item Diagram

A variation of an item diagram, where the external item values for each profile are superimposed on the profiles in the POSAC diagram, which is then used to assess whether there is a relationship between the content facets’ elements and an external item (Levy and Guttman 1985; Shye and Amar 1985).

Facet

A facet is a set of elements playing the role of a component set of a Cartesian set that together represent underlying conceptual and semantic components within a content universe (Guttman and Greenbaum 1998).

Facet Roles

Facets may play an axial, modular, polar, or joint role (Levy and Guttman 1985). An axial role is played by a facet whose elements are ordered but the order of the elements is uncorrelated with the order of the elements of the other facets. Axial facets partition the SSA space into sections using horizontal or vertical lines. A modular facet also consists of ordered elements, but it may be related to other facets. A modulating facet forms concentric circles radiating from a particular origin. A polar facet consists of unordered elements where the elements form a circular distribution of points termed a circumplex geometric structure. Polar facets represent a qualitative facet where no obvious beginning or end from which order among the elements exists (Levy and Guttman 1985). They partition space into wedge-shaped regions, with each region representing an element of the facet (Brown 1985). A fourth role is that of the joint role where two ordered facets act together to represent one dimension of a model. Facets play a joint role when two or more of them have a common notion of order, and combinations of elements divide the SSA into conceptually consistent regions (Borg and Shye 1995, Shye 1998). Joint facets tend to divide the SSA space diagonally. Note that these partitions presented above are ideal types; in practice, the partitioning of the SSA space will not be perfect (Shye 1985a).

MIS Quarterly Vol. 28 No. 2/June 2004

219

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

Table A1. Facet Theory Terminology (Continued) Term

Definition

Geometric Data Structures

Facets, in their different roles, can be combined to form different geometric structures of the data. Three of the most common structures are the duplex, radex, and cylindrex. A duplex structure is formed by combining two axial facets, where one of the facets partitions the two-dimensional SSA space horizontally while the other facet divides the space vertically, resulting in the partitioning of the space into squares or rectangles. A radex structure is formed in the two-dimensional SSA space by combining polar and modular facets, and has a common origin for both the polar and modular facets. The cylindrex divides a three-dimensional SSA space into ordered or layered radexes (Borg and Shye 1995; Shye 1998). Note that these geometric data structures are ideal types; in practice, these structures will not be perfectly formed (Shye 1985a).

Guttman’s mu2

Also called the weak monotonicity coefficient, it is a nonparametric measure of association which makes the weakest possible assumptions about empirical observations in that data are assumed to have ordinal properties only so that its calculation does not require the distribution of the variables to be known. Guttman’s mu2 assesses the extent to which an increase in one variable is accompanied by an increase (or no decrease) in the other. Like other measures of association, the value of Guttman’s mu2 can vary between -1 to +1 (Canter 1985; Shye 1985a).

Incomparable Profiles

Two profiles (score structuples) are not comparable if and only if one profile is higher on at least one content structuple while the other profile is also higher on at least one other content structuple (Levy and Guttman 1985).

Intension

The addition of new facets to an existing mapping sentence, usually as justified by empirical observation (Shye et al. 1994).

Item

An item is an observational question together with its range of admissible answers (Borg and Shye 1995). It is used to represent possible facet element values for each content structuple.

Item Diagram

An item diagram is a reproduction of the POSAC diagram except that instead of the profile ID, the profile’s score in any item is presented. Item diagrams show how an item partitions the POSAC space by the item values (Levy and Guttman 1985).

Joint Axis (in POSAC)

The joint axis, obtained by rotating the X axis by 45/, represents quantitative (summative) aspects of the observed phenomena. It is used to compare the order of different profiles in the sample by ranking profiles according to the underlying common range (Shye et al. 1994).

Lateral Axis (in POSAC)

The lateral axis obtained by rotating of the Y axis 45/, represents qualitative (differential) aspects of the observed phenomena and is utilized to order profiles that have the same joint coordinate value and to determine the role the different content facets play (Shye et al. 1994).

Mapping Sentence

A verbal statement of both the domain (population and content facets) and the range (range facets) which connects the facets in ordinary language. The mapping sentence serves to define a priori exactly what is being studied—the population, the content variables, and the range of possible responses serving as the definitional and conceptual base of the problem to be studied (Borg and Shye 1995).

220

MIS Quarterly Vol. 28 No. 2/June 2004

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

Table A1. Facet Theory Terminology (Continued) Term

Definition

Monotonic Relationship

A monotonic relationship exists when there is a consistent upward or downward trend between the two variables of interest (Shye 1985a).

Ordered Facet

A facet is ordered to the degree there is a clear hierarchy in the relationships of its elements (Levy and Guttman 1985).

Partial Order

A partial order is a relation on a set of elements such that determining which element relative to another has a higher or lower order is not possible (Levy and Guttman 1985).

Partial Order Scalogram Analysis by base Coordinates (POSAC)

Partial order scalogram analysis by base coordinates (POSAC) is a procedure for fitting observed profiles into a two dimensional coordinate space representation subject to the constraint that ordered relations, including incomparability, are preserved. POSAC is used to investigate the structural characteristics of subject, enabling the comparison (or partial comparison) of the subjects (profiles) in the sample based on differences in their content facet element values. It also can be used to assess whether there is an association between the different subjects— represented by different combinations of their content facets—and an external item (Shye and Amar 1985).

POSAC Diagram

A two dimensional partial order scalogram of an iteratively calculated configuration of points for a set of profiles based on some or all of the content facet values where the partial order is preserved as best as possible (Levy and Guttman 1985).

PQ! !R

The mathematical representation of a mapping sentence, where P is the population of interest, Q the domain of interest, and R the response given the Cartesian product of sets P and Q (Borg and Shye 1995).

Profile

See Score Structuple

Range Facet

The range facet is an ordered set of possible responses to the content facets (Brown and Barnett 2000). A special case of a range facet is the common meaning range, where the range of responses (elements) are ordered from high to low by a common meaning in such a way that, in all content facets, high numerical values indicate presence (or absence) of the attribute (Shye et al. 1994).

Regional Hypotheses

Regional hypothesis assess how the data points partition a geometric region into different spaces based on the different elements rather than the clustering of points within that space, and are used in SSA to evaluate the structure of the model and in POSAC to compare the different sample profiles (Levy and Guttman 1985; Shye and Amar 1985).

Scalogram

A scalogram is a collection of observed profiles in which each profile is represented as a point in a two dimensional space and the comparability between two profiles is represented by a line segment (Shye 1985a).

Score Structuple

Also called a profile, a score structuple is obtained from the classification of elements for any item, where the profile has a specific element value in each facet. Each subject is represented by one score structuple (Guttman and Greenbaum 1998).

MIS Quarterly Vol. 28 No. 2/June 2004

221

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

Table A1. Facet Theory Terminology (Continued) Term

Definition

Shepard Diagram

A scatterplot of distances between points in the SSA plot against the observed dissimilarities or similarities whose shape is used to determine the adequacy of the model. A model’s goodness-of-fit is indicated by a relatively smooth negatively sloped monotone series of points for similarities or a positively sloped monotone series of points for dissimilarities (Shepard 1962a, 1962b).

Smallest Space Analysis (SSA)

A method of multidimensional scaling in which a set of variables and their intercorrelations are geometrically portrayed in space while preserving the rank order of relations, where smallest space means the “space of lowest dimensionality” (Shye et al. 1994). SSA aims at establishing the structural properties of variables and is often used to test the hypotheses stated in the mapping sentence by relating the conceptual structure of observations on the universe of features of the empirical structure of observations on that universe (Guttman and Greenbaum 1998).

Struct

The actual value for a particular element in a structuple of a specific subject (Borg and Shye 1995).

Unordered Facet

A facet is unordered to the degree there is no clear hierarchy in the relationships of its elements (Levy and Guttman 1985).

Weak Monotonicity Coefficient

See Guttman’s mu2

The mapping sentence consists of a minimum of three components that can be represented as PQ!R. The population of interest (P) specifies the universe of possible subjects and a single profile (p) defines the unit of analysis for a particular research problem. Q and R are two different types of facets, where a facet is a component set of a Cartesian set within a content universe (Guttman and Greenbaum 1998). The content facets (Q) are the attributes or conditions by which the population of interest (P) is compared or classified (Borg and Shye 1995). In the mapping sentence, the items to the left of the arrow are the conditions of the observation—the subject (profile) and its attributes of interest (content facets). The item to the right of the arrow is the range facet (R), which represents the range of possible observations used to classify individual subjects within the population of interest (P). Each facet consists of elements which specify the exact values subjects may be assigned within a particular facet. Content facet elements may be either ordered or unordered, while range facet element values must be ordered. Ordered facets are those facets whose elements represent quantitative distinctions, while unordered facets are those facets whose elements represent only qualitative distinctions. Elements within a facet are mutually exclusive and jointly exhaustive; therefore, the elements in a particular facet ideally represent all possible values for that particular universe (Brown 1985; Edmundson et al. 1993; Shye et al. 1994). The mapping sentence thus is a verbal statement mapping the logical relationships between the content facets and the range facets for a specific population, where the Cartesian product of facet sets P × Q represents the whole possible universe of the population asked all possible questions, and R represents the whole possible universe of answers to those questions (Borg and Shye 1995). The mapping sentence provides the theoretical basis by which to assess the structure of the data and compare different profiles. At the same time, facet theory techniques of intension—the addition of new facets to an existing mapping

222

MIS Quarterly Vol. 28 No. 2/June 2004

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

sentence, usually as justified by empirical observation—and extension—the addition of elements to existing facets, usually as justified by empirical observation—provides the researcher with the flexibility of pursuing unexpected opportunities (Shye et al. 1994), particularly in case studies of emergent phenomenon. The mapping sentence may also include a fourth component—an external item, which is a “classification of the content facets and their elements specified in terms of the content facets’ properties or ‘behavior’ with respect to criterion external to those content facets” (Borg and Shye 1995, p. 161). The external item enables the comparison of an association between one or more of the internal content facets and an external concept.

Data Analysis in Facet Theory Facet theory data analysis methods involve two related multidimensional scaling (MDS) techniques. The goal is to determine whether a correspondence between the definitional system—the mapping sentence—and the observed data exists. Smallest space analysis (SSA) is used to evaluate the structure of the model, while partial order scalogram analysis by base coordinates (POSAC) is used to compare—or at least partially compare if direct comparison is not possible—differences in the sample profiles, and in cases where an external item exists, how the different profiles or content facets are associated with the external item. Both techniques use a Euclidean space in which facet elements in SSA and profiles in POSAC are presented as points to examine if they partition the space into contiguous regions. Both techniques are a form of discriminant analysis where regional hypotheses—how the data points partition a geometric region into different spaces based on the different elements or profiles rather than the clustering of points within that space—are used to evaluate the structure of the model and to compare the different sample profiles (Levy and Guttman 1985). Data assumptions in facet theory require that the data need only be ordinal (Shenkar et al. 1995; Shye et al. 1994), and relationships between variables need not be linear (Borg and Shye 1995). Such assumptions are appropriate in behavioral and social research because they are consistent with many aspects of human behavior (Guttman 1944; Levy and Guttman 1985). Thus, while parametric measures of association may be used, nonparametric measures of association are often used. An advantage of using nonparametric measures of association in MDS is that with as few as seven data points the data analysis results can be quite robust (Borg and Lingoes 1987; Shye 1985a). A nonparametric measure of association often used in facet theory is the weak monotonicity coefficient (also called Guttman’s mu2), which makes the weakest possible assumptions about empirical observations in that data are assumed to have ordinal properties only (Canter 1985b). The weak monotonicity coefficient assesses the extent to which an increase in one variable is accompanied by an increase (or no decrease) in the other. Like other measures of association, the value of the weak monotonicity coefficient can vary between -1 to +1. However, its calculation does not require the distribution of the variables to be known; as such, the order of interpoint distances is sufficient for determining a unique configuration of points in a geometric space (Shye 1985a).

Evaluation of the Model Structure Smallest space analysis is used to evaluate the structure of the model in terms of the dimensionality and construct validity of the model. SSA comprises a class of MDS models that represent similarity (or dissimilarity) coefficients among a set of objects by distances in a multidimensional space (Borg and Lingoes 1987). SSA of the content facets for different dimensional models is performed to determine the optimal dimensionality of the model.

MIS Quarterly Vol. 28 No. 2/June 2004

223

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

As in factor analysis, specific rules for determining the structure of a model and its appropriate dimensionality do not exist; instead, guidelines or criteria are utilized (Borg and Shye 1995; Shye 1998). Decision-making criteria include the coefficient of alienation values and the proportion of variance explained by the SSA solution, the change in those values when adding a dimension, model parsimony, the shape of the different models’ Shepard diagrams, and the construct validity of the different models. The coefficient of alienation is a measure of the model’s goodness-of-fit determined by the extent to which the distance between pairs of points and a two dimensional space adhere to the rule regarding the monotonic relationship between input coefficients and output distances. A monotonic relationship exists when there is a consistent upward or downward trend between the two variables of interest (Shye 1985a). The coefficient of alienation value can be between 0 and 1, inclusive, where the best fit is represented by a 0 value and the worst fit is given by the value 1 (Guttman 1968). As a general rule, a coefficient of alienation value of less than 0.15 indicates a good fit (Brown 1985; Guttman 1968; Shapira 1976). The shape of the Shepherd diagram of the SSA solution is also used to determine the adequacy of the model. The Shepard diagram (Shepard 1962a, 1962b) is a scatterplot of distances between points in the MDS plot against the observed dissimilarities or similarities. A model’s goodness-of-fit is indicated by a relatively smooth negatively sloped monotone series of points for similarities or a positively sloped monotone series of points for dissimilarities. An approach similar to the scree test (Cattell 1966) in factor analysis is used to determine the appropriate dimensionality of a model in facet theory. Changes in the coefficient of alienation and in the proportion of variance explained when the model is tested in different dimensions are examined, and the model’s appropriate dimensionality is where adding another dimension would account for only small changes in the values of the coefficient of alienation and the proportion of variance explained. The construct validity of the model is also used to assess the structure of the model and its appropriate dimensionality. Construct validity is assessed through tests of regional hypotheses. SSA is a technique for plotting the different facets and elements in an n-dimensional space. Unlike other MDS techniques, SSA does not attempt to identify the dimensions of the model (Borg and Shye 1995). Instead, the researcher specifies them based on theoretical considerations and an examination of the SSA solution. Using regional hypotheses, construct validity of a model is determined by assessing the extent to which the hypothesized model—represented by the mapping sentence—and the partitioning of the SSA space—a geometric representation of the empirical association of the facet elements—correspond with each other (Borg and Shye 1995). SSA is a form of discriminant analysis, and the ability of facet theory data points to discriminate or divide the space into the different elements rather than the clustering of points within that space is the basis by which construct validity is judged (Borg and Lingoes 1987; Borg and Shye 1995; Shye 1998). Partitioning of the SSA space is not haphazard. The partitions must be consistent with the type of facet and role they are hypothesized to fill in the mapping sentence. The type of facet—ordered or unordered— and the role of that facet determines the shapes by which regional hypothesis partition the SSA space. Different types of facets partition the SSA space into different shapes, depending on the role such facets played in the conceptualized mapping sentence. Facets may play an axial, modular, polar, or joint role (Levy and Guttman 1985). An axial facet is one whose elements are ordered but the order of the elements is uncorrelated with the order of the elements of the other facets. Axial facets partition the SSA space into sections using horizontal or vertical lines. A modular facet also consists of ordered elements, but it may be related to other facets. A modulating facet

224

MIS Quarterly Vol. 28 No. 2/June 2004

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

forms concentric circles radiating from a particular origin. A polar facet consists of unordered elements where the elements form a circular distribution of points termed a circumplex geometric structure. Polar facets represent a qualitative facet where there is no obvious beginning or end from which order among the elements may be determined (Levy and Guttman 1985). They partition space into wedge-shaped regions, with each region representing an element of the facet (Brown 1985). A fourth role occurs when two ordered facets act together to represent one dimension of a model. Facets play a joint role when two or more of them have a common notion of order, and combinations of elements divide the SSA into conceptually consistent regions (Borg and Shye 1995, Shye 1998). Joint facets tend to divide the SSA space diagonally. The facets, in their different roles, can be combined to form different structures of the data. Three of the most common structures are the duplex, radex, and cylindrex. A duplex structure is formed by combining two axial facets, where one of the facets partitions the two-dimensional SSA space horizontally while the other facet divides the space vertically, resulting in the partitioning of the space into squares or rectangles. A radex structure is formed in the two-dimensional SSA space by combining polar and modular facets, and has a common origin for both the polar and modular facets. The cylindrex divides a three-dimensional SSA space into ordered or layered radexes (Borg and Shye 1995; Shye 1998). Note that the partitions presented above are ideal types; in practice, the partitioning of the SSA space will not be perfect (Shye 1985a). The manner in which the SSA space is partitioned is guided by the need to replicate the findings across different samples. Regional hypotheses are more likely to be replicated if the original SSA space is partitioned by regular, simple lines. Therefore, while it is possible to partition particular samples with irregular lines or shapes, this is not desirable because it is less likely such irregular partitions will be replicated. The more irregular lines or shapes utilized to partition a particular space, the more likely the regional hypothesis will not be empirically supported over time (Borg and Shye 1995, Shye 1998).

Comparisons of Subjects and Their Attributes Partial order scalogram analysis by base coordinates (POSAC) is used to compare—or partially compare—the profiles of subjects in the sample based on differences in their content facet element values. It also can be used to assess whether there is an association between the different subjects—represented by different combinations of their content facets—and the external item. The subjects (p) are represented by their score structuples, which consist of element values for all the different content facets. The actual value for a particular element in a score structuple of a specific subject is a struct (Borg and Shye 1995; Brown 1985). Subjects also can be represented by their content structuples, which consist of a subset of their score structuples. The different subjects can be ordered or partially ordered by comparing score or content structuple values. A partial order is a relation on a set of elements such that determining which element relative to another has a higher or lower order is not possible (Borg and Lingoes 1987). Structuples are comparable if all the structuple values of one structuple are equal to or greater than the corresponding structuple values of the other structuple, and is of a higher order than another if and only if it is higher on at least one item and not lower on any other items. However, two structuples are not comparable if and only if one structuple is higher on at least one structuple while the other structuple is also higher on at least one other structuple (Levy and Guttman 1985). A scalogram facilitates assessing the partial order of a set of different profiles. A scalogram is a collection of observed profiles in which each profile is represented as a point in a two-dimensional space and the comparability between two profiles is represented by a line segment. POSAC is a procedure for fitting observed profiles into a two-dimensional coordinate space subject to the constraint that ordered relations,

MIS Quarterly Vol. 28 No. 2/June 2004

225

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

including incomparability, are preserved (Shye 1985a). A configuration of points for a set of profiles based on some or all of the content facet values is iteratively calculated, and these points, labeled by their profile ID, are used to graph a two-dimensional partial order scalogram such that the partial order is preserved as best as possible (Shye and Amar 1985). A two dimensional space representation (termed a POSAC diagram) is utilized because it is the smallest dimensionality that preserves the partial order (Levy and Guttman 1985). POSAC calculates two sets of coordinates for each profile. The first set of coordinates, Dim(1) and Dim(2), represent the location of the profile on the X and Y axis, respectively, where the X and Y axis are a mathematically optimal case of base axis for the empirical partial order (Levy and Guttman 1985). The second set of coordinates is for the joint (J) and lateral (L) axis, which are the X and Y axis rotated clockwise 45 degrees. The joint axis has a positive slope of one relative to the X and Y axis, and is used to compare the order of different profiles in the sample. The lateral axis has a negative slope of one relative to the X and Y axis, and is utilized to order profiles that have the same joint coordinate value and to determine the role the different content facets play (Borg and Shye 1995; Levy and Guttman 1985; Shye and Amar 1985). A goodness-of-fit measure for POSAC models is the coefficient of correct representation (CORREP), which assesses the percentage of profiles whose order relations are correctly represented by POSAC. CORREP can have values between 0 and 1, with 1 meaning the model order is perfectly represented (Shye and Amar 1985). The order represented in the POSAC diagram is based on the content facets only, where the order of a profile relative to other profiles is a function of their position relative to the joint axis. Generally speaking, those profiles located in the upper right-hand corner of the POSAC diagram are of a higher order than those in the lower left-hand corner. While the joint axis coordinates facilitate ordering the profiles, they are not sufficient to order the profiles when only partial order exists. Not all profiles are comparable. Two methods to assess which profiles are comparable or not comparable, and the order of comparable profiles, are available. The first method is based on the slope of the line connecting two profile points. Two comparable score structuples are represented by two points on a common line having a positive slope, with the points closer to the upper right corner having a higher order. Noncomparable score structuples have their points aligned along a negative slope; that is, they are oriented toward a direction orthogonal to the joint direction, namely the lateral direction (Levy and Guttman 1985; Shye and Amar 1985). The second method of assessing profile order is by drawing two lines—one parallel to the X axis and the other parallel to the Y axis—through the profile of interest, dividing the POSAC diagram into four quadrants. Those profiles in the upper right-hand quadrant have profiles of a higher order than the profile intersected by the two lines, while those profiles in the lower left-hand corner have profiles of a lower order than the profile intersected by the two lines. The profiles in the remaining two quadrants—the upper left-hand and lower right-hand quadrants—are not comparable to the profiles intersected by the two lines (Levy and Guttman 1985; Shye and Amar 1985). One of the most useful aspects of the POSAC analysis is the item diagram, which is the reproduction of the POSAC diagram with an item (content facet element) score replacing the profile name. This enables the comparison of the different profiles based on part or all of their score structuple values. Regional hypotheses of the item diagram(s) are used to assess the role of that particular item. The POSAC space for the different item diagrams are partitioned based on the role the facets elements are hypothesized to play. Facets or elements hypothesized to play either an axial or polar role should partition the POSAC item diagram with lines parallel to either the X or Y axis, and the weak monotonicity coefficient for the facet or element should approach one and negative one for the X and Y axis (or vice versa). A facet or element hypothesized to play an attenuating role partitions the POSAC item diagram with an “L” shaped line and the facet or element’s weak monotonicity coefficient values should approach one for the X axis and zero

226

MIS Quarterly Vol. 28 No. 2/June 2004

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

for the Y axis. Such a facet or element is structurally dependent on the base-coordinate (X or Y) items such that values of an attenuating item tend to increase with an increase in the concurrent simultaneous increase in values of both of the base coordinates, and its role is to moderate the differentiation delineated by the base coordinates (Borg and Shye 1995; Shye 1985a; Shye et al. 1994). A facet or element hypothesized to play an accentuating role partitions the POSAC item diagram with an inverted “L” shaped line and the facet or element’s weak monotonicity coefficient values should approach zero for the X axis and one for the Y axis. Such a facet or element is structurally dependent on the base-coordinate (X or Y) items such that values of an accentuating item tend to increase with an increase in the compound values (either or both) of the base coordinates, and its role is to sharpen the differentiation delineated by the base coordinates (Borg and Shye 1995; Shye 1985a; Shye et al. 1994). Note that the partitions presented above are ideal types; in practice, the partitioning of the POSAC space will not be perfect (Shye 1985). Ideally, the profiles in the POSAC diagram would be perfectly represented (CORREP = 1.0) and therefore the weak monotonicity coefficient values of the various elements and axes would have no error as well. This rarely is going to be the case, and the results are going to require some interpretation. Regional hypotheses of the external item diagram, which superimposes the external item values for each profile in the POSAC diagram, are used to assess whether there is a relationship between the content facets’ elements and an external item (Levy and Guttman 1985; Shye and Amar 1985). This relationship is tested empirically by superimposing the external facet content structuple values of the profiles on their respective points in the POSAC diagram and calculating the external item’s weak monotonicity coefficient value with the joint axis and lateral axis. A content facet has a positive association with an external item if the external item’s weak monotonicity coefficient value with the joint axis approaches one and its value with the lateral axis approaches zero (Shye 1985a). A positive relationship between the content facets and an external item is represented by the partitioning of the external item diagram of the POSAC space by a line parallel to the lateral axis (Levy and Guttman 1985). The hypothesis is empirically supported if the profiles with the strongest association with the external item are located in the upper right-hand corner of the POSAC diagram which has been partitioned by the line parallel to the lateral axis. POSAC is a form of discriminant analysis; as such, the ability of facet theory data points to discriminate or divide the space based on the different elements and not the clustering of points within that space is the basis by which the acceptability of the partitioning is judged (Levy and Guttman 1985). Exact rules for determining a POSAC model’s ability to discriminate between the different facet elements, or in the case of the external item diagram between the external items based on content facet values, are not available. Acceptable discriminatory power is a function of the research purpose and design, quality of the mapping sentence and data collection methods, and the choice of facets and sample. Replication across different samples ultimately is the key criteria for determining the acceptability of the partitioning of the POSAC space as a means of understanding the role of the different content facets or for establishing a relationship between the content facets and the external item (Borg and Shye 1995; Shye and Amar 1985). The POSAC diagram can be an effective tool for predicting the outcome/value of an external item given the values of content facet elements once the partitioning of the POSAC space has been determined acceptable. Facet theory is an appropriate methodology for addressing the challenges information systems researchers face because it provides a systematic approach to developing a research design and completing data analysis in the study of complex social systems. The development of a mapping sentence is one key to the success of this methodology. The mapping sentence enables the modeling of the appropriate facets and their relationships. Multidimensional scaling provides useful strategies for evaluating the structures of the resulting models and for evaluating the hypotheses derived from these models.

MIS Quarterly Vol. 28 No. 2/June 2004

227