A Field Study of the Effect of Interpersonal Trust on ...€¦ · MIS Quarterly Vol. 28 No. 2, pp. 183-227/June 2004 183 RESEARCH ARTICLE A FIELD STUDY OF THE EFFECT OF INTERPERSONAL

Paul & McDaniel/Effect of Interpersonal Trust on VCR Performance

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

RESEARCH ARTICLE

A FIELD STUDY OF THE EFFECT OFINTERPERSONAL TRUST ON VIRTUALCOLLABORATIVE RELATIONSHIPPERFORMANCE1

By: David L. PaulDepartment of Information Technology

and Electronic CommerceDaniels College of BusinessUniversity of Denver2101 S. University BoulevardDenver, CO [email protected]

Reuben R. McDaniel, Jr.Department of Management Science and

Information SystemsMcCombs School of BusinessUniversity of Texas at AustinAustin, TX [email protected]

Abstract

This article examines the relationship betweeninterpersonal trust and virtual collaborative rela-tionship (VCR) performance. Findings from a

1Michael D. Myers was the accepting senior editor forthis paper.

study of 10 operational telemedicine projects inhealth care delivery systems are presented. Theresults presented here confirm, extend, and ap-parently contradict prior studies of interpersonaltrust. Four types of interpersonal trust�calcula-tive, competence, relational, and integrated�areidentified and operationalized as a single con-struct. We found support for an association be-tween calculative, competence, and relationalinterpersonal trust and performance. Our findingof a positive association between integrated inter-personal trust and performance not only yields thestrongest support for a relationship between trustand VCR performance but also contradicts priorresearch. Our findings indicate that the differenttypes of trust are interrelated in that positive as-sessments of all three types of trust are necessaryif VCRs are to have strongly positive performance.The study also established that if any one type oftrust is negative, then it is very likely that VCR per-formance will not be positive. Our findings indi-cate that integrated types of interpersonal trust areinterdependent, and the various patterns of inter-action among them are such that they are mutu-ally reinforcing. These interrelationships andinterdependencies of the different types of inter-personal trust must be taken into account byresearchers as they attempt to understand theimpact of trust on virtual collaborative relationshipperformance.


184 MIS Quarterly Vol. 28 No. 2/June 2004

Keywords: Interpersonal trust, collaboration,virtual teams, telemedicine

Introduction

One should expect trust to be in-creasingly in demand as a means ofenduring the complexity of the futurewhich technology will generate (Luh-mann 1979, p. 16).

In a complex world, trust is a necessity. Trusteffectively and efficiently reduces complexity byenabling parties with different knowledge basesand experiences to collaborate (Gefen 2000;Luhmann 1979; Lewis and Weigert 1985). Suchcollaborative relationships can be either virtual orface-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 betweeninterpersonal trust and virtual collaborative rela-tionship (VCR) performance. Findings resultingfrom data collected in a field study of VCRs inhealth care delivery, specifically 10 operationaltelemedicine projects, are presented. VCRs inhealth care delivery have characteristics thatmake them exemplary subjects for studying trustin a context where trust is very important, yet theyoften operate in a context hostile to developingand maintaining trust.

This study makes several contributions to re-search on interpersonal trust in VCRs. First, theformation of collaborative relationships, be theyvirtual or face-to-face, does not guarantee theywill be effectively utilized. Collaborative relation-ships are inherently social constructions, and theirsuccess or failure may be due to the social con-text in which they exist, and not the quality of therelationships themselves. This study extendsresearch on the relationship between inter-personal trust and performance. Partial supportfor a relationship between performance and affect-based interpersonal trust has been found(McAllister 1995), but no relationship between

interpersonal trust and performance has also beenfound (Zaheer et al. 1998). Prior research foundthat antecedents to trust were a good predictor oftrust development in virtual teams (Jarvenpaa etal. 1998), while another study found a relationshipbetween communication processes and swift trustdevelopment (Jarvenpaa and Leidner 1999).

Second, this study helps us to understand whytechnically sufficient information systems may notbe adopted. Deploying information technologysystems is often predicated on the assumptionthat the value of an information system�s informa-tion quality and increased information processingcapabilities is sufficient to justify system use.However, this assumption does not always hold,and decision makers often rely on sources theytrust the most�regardless of the timeliness,quantity, and quality of information from thesesources (Mishra 1996; Staw et al. 1981). Thereare many reasons why systems are not adopted(Collins and Bicknell 1998). This study attemptsto differentiate between complexity reduction ef-forts that fail due to insufficient information pro-cessing capabilities, such as inadequate hardwareor software capabilities, and others that fail due toa lack of trust within the relationship.

Third, this study introduces a methodology, facettheory, previously not utilized in information sys-tems research. Facet theory originated in psycho-logy and is a systematic approach to facilitatingand integrating research construction, design, anddata analysis of complex social systems. Itutilizes multidimensional data analysis guided bya theoretical framework (Borg and Shye 1995;Guttman and Greenbaum 1998; Shye 1998).Facet theory has the potential to address many ofthe concerns and challenges information systemsresearchers face in performing field research.

Fourth, this research makes a significant contri-bution to practice by demonstrating that organi-zations must address interpersonal trust factors ifthey want to reap the benefits of newer workrelationships. It also suggests new considerationsin the ongoing concerns about the more extensiveadoption of virtual relationships, specifically tele-medicine, in the effective and efficient deploymentof health care resources.


MIS Quarterly Vol. 28 No. 2/June 2004 185

Interpersonal Trust

Trust in an organizational setting is an effectiveenabler of complexity reduction�especially whenimportant decisions and new technology areconcerned (Gefen 2000; Lewis and Weigert 1985;Ring and Van de Ven 1994). Trust plays a keyrole as a foundation for effective collaboration(Kramer 1999; Mayer et al. 1995; Rousseau et al.1998; Whitener 1998) and is the salient factor indetermining the effectiveness of many rela-tionships (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 oncollaborative relationship performance is facili-tating the learning and innovation (Goshal andMoran 1996; Newall and Swan 2000; Sako 1998)needed to address the ambiguity and unstructurednature of wicked decision problems (Mason andMitroff 1973). A direct link between trust andcollaborative relationship performance exists;once the need for collaboration is established,trust becomes the salient factor in determiningperformance.

Trust is particularly important in newer organiza-tional forms such as virtual collaborative relation-ships (McKnight et al. 1998; Meyerson et al. 1996;Newall and Swan 2000; Ring 1996). In the virtualworld, trust is a way to �manage people whom youdo not see� (Handy 1995, p. 41). Yet trust invirtual teams is difficult to build, and it has beenargued that face-to-face contact is irreplaceablefor building trust (Nohria and Eccles 1992). Thereplacement of technology for collocation under-mines the emotional relationship aspects of trust.Collocation reinforces social similarity, sharedvalues, and expectations, and increases the im-mediacy of threats from failing to meet commit-ments (Jarvenpaa and Leidner 1999; Latane et al.1995; Sako 1998).

Trust is especially important in health caredelivery because health care providers rely on col-laboration as a primary means of complexityreduction. Health care delivery is a collaborativeactivity whose quality, efficiency, and respon-siveness is enhanced by the use of inter-

disciplinary teams (IOM 1996a). Traditionally,health care providers have reduced complexity bycollaborating with other health care knowledgeworkers (IOM 1996a; Silberman 1992). Suchcomplexity results from the co-morbidity of patientconditions that requires providers to simul-taneously deal with multiple problems that interactwith each other and often preclude assignment ofcausal relationships with any certainty (IOM1996b). Patient symptoms and test results aresubject to multiple plausible, but conflicting, inter-pretations (IOM 1996b). Multiple treatmentcourses with multiple, contradictory, and oftenuncertain outcomes are available (Silberman1992), and patient and family members vary intheir treatment and quality of life preferences.

While trust is critically important in health caredelivery relationships, the health care deliveryenvironment does not facilitate the creation andmaintenance of trust (IOM 1996b; OTA 1995;).Collaborative consultations can expose one tomalpractice liabilities due to the actions of others.A health care provider engaging in collaborativeconsultations risks losing patients to the otherparty, and must trust that the other party will notattempt to steal patients. Health care deliveryinvolves multidisciplinary teams, where teammembers have to rely on others whose trainingand perspective are different from their own, andwhere there are significantly different power andstatus relationships. Trust in such differentiatedteams can be especially difficult to create andmaintain (Luhmann 1979). Indeed, it can beargued that lack of trust is the natural state of thehealth care delivery environment.

The importance of and difficulty in creating andmaintaining trust is especially in force in tele-medicine. Telemedicine is the process of two ormore geographically separated health careproviders collaborating via information technologyto provide value-added health care delivery (IOM1996b). The standard measures of health caredelivery�and measures of the effectiveness ofcomplexity reduction efforts�are access, cost,and quality (IOM 1990, 1993, 1996a), and it isclaimed that telemedicine can increase the accessto and quality of health care delivery while simul-



taneously lowering costs (Allen and Stein 1998;Debakey 1995; Flaherty et al. 1995; GAO 1997;Grigsby 1995; IOM 1996b; ORHP 1997; OTA1995). Telemedicine can result in significant eco-nomies of scale as expertise can be centralizedand utilized more frequently (COC 1996; DOC1997; GAO 1997; IOM 1996b; ORHP 1997).

Telemedicine brings its own additional problemsin creating and maintaining trust. Telemedicineprojects involve multidisciplinary teams, andvirtual processes may differ from what health careproviders are used to. Telemedicine blurs indivi-dual boundaries and thrusts participants intounfamiliar interactions. Technology changes themode of presentation on which trust formation isbased. Technology may change how and whatcues are noticed or ignored relative to face-to-faceinteractions, and it may force health care pro-viders to interact with team members in a mannerdifferent from what they are used to. The installa-tion of telemedicine infrastructure may beinterpreted as threatening by remote health careproviders, who may see it as the first step towardtheir replacement by nonlocal providers (GAO1997). Interpersonal trust may not exist for manytelemedicine projects.

While there remain significant variations inconceptualizations of trust by organizationalresearchers, a general consensus has beenreached that trust is a psychological state basedon confident expectations and beliefs that anotherparty will act in a certain manner, and that thetrusting party must in some way be vulnerableunder conditions of risk and interdependency toactions by the other party (Kramer 1999;Rousseau et al. 1998). Our study focuses on therelationship of interpersonal trust with virtualcollaborative relationship performance.

Identifying types of interpersonal trust is a con-tentious and confusing issue. Researchers iden-tify different types of interpersonal trust, use dif-ferent terminology for similar types of trust, or usesimilar terms for different types of trust, and sub-categorize the same type of trust in different ways.These various subcategories introduce complexityinto the study of trust (Bigley and Pearce 1998).

For example, cognition trust differs in its defini-tions and, for some, comprises subcategories oftrust that others consider separate types of trust.Some researchers consider calculative trust acore type of interpersonal trust, while others donot consider it a type of trust at all. In this section,we present our model of interpersonal trust andthe hypothesized relationships with VCR perfor-mance. We have attempted to bring to light themajor types of trust in the literature, how defini-tions of the same type of trust differ, and why wechose the definitions and operationalizations wedid.

Figure 1 presents our model of interpersonal trustand the hypothesized relationships with VCRperformance. We have identified three types ofinterpersonal trust�calculative, competence, andrelational�and combined them for a fourth type ofinterpersonal trust�integrated�which includes allthree types of trust in our model.

Calculative Trust

Calculative trust is based on conceptualizing trustas a form of economic exchange (Child 1998;Lane 1998; Lewicki and Bunker 1996). Alsotermed rational trust (Gambetta 1988; Lewicki andBunker 1996; Mayer et al. 1995; Williamson1993), calculus-based trust (Rousseau et al.1998), commitment trust (Newell and Swan 2000),and contractual trust (Sako 1991, 1992), calcula-tive trust is an ongoing, market-oriented, econo-mic calculation where each party assesses thebenefits and costs to be derived from creating andsustaining a relationship (Child 1998; Lewicki andBunker 1996). Calculative trust is a form of con-tractual agreement where parties can be relied onto deliver according to the details of the contract(Newell and Swan 2000; Sako 1991, 1992). Theparties choose whether or not to participate in atrusting relationship based on some form of cost-benefit analysis. Individuals are assumed to beeconomically rational beings motivated by theirdesire to maximize expected gains or minimizeexpected losses in their transactions (Kramer1999).



Interpersonal Trust

Self-Interest

Ability

Empathy

Integrated

VirtualCollaborativeRelationshipPerformance

Interpersonal Trust

Self-Interest

Ability

Empathy

Integrated

VirtualCollaborativeRelationshipPerformance

Figure 1. Interpersonal Trust and VCR Performance Model

The concept of calculative or rational trust is notwithout its detractors. Some argue that trust is notthe result of conscious calculation (Kramer 1999)nor is it a form of economic exchange (March andOlsen 1989), concluding that calculative trust doesnot exist. Trust is needed only when conditions ofinformation uncertainty exist; however, calculativetrust is effective only when there is little or noneed to trust because there are only limited, iden-tifiable conditions of information uncertainty (Child1998; Lane 1998). Therefore, calculative trust,and the ability to assess the costs and benefits ofengaging in a relationship, comes into play onlyunder conditions in which the need for trust islimited.

Despite these objections, we include calculativetrust in our model because the condition that thetrusting party must be vulnerable to the non-performance of the other party in our definition oftrust holds in this case. Trust includes motiva-tional components (Kramer 1999; Shepard andTuchinsky 1996), and some of these motivationalcomponents may be calculative. This leads to ourfirst hypothesis.

H1: There is a positive association be-tween calculative trust and virtual colla-borative relationship performance.

Competence Trust

Competence trust is whether the other party iscapable of doing what it says it will do (Butler1991; Butler and Cantrell 1984; Mayer et al. 1995;Mishra 1996; Sako 1991, 1992, 1998). There ap-pears to be a definitional consensus about com-petence trust in the research community. It is anassessment of the expertise and abilities of theother parties, and is important in a knowledge-based economy because it acts as an indicator ofthe other party�s ability to perform as anticipated(Rousseau et al. 1998). Competence trust is re-quired in complexity reducing collaborative effortswhen the skills needed to perform a task are notfound within one person (Newall and Swan 2000).A party is more likely to engage in a collaborativerelationship if they perceive the individuals in theother party as being capable.

H2: There is a positive association be-tween competence trust and virtual colla-borative relationship performance.

Relational Trust

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



personal attachment to the other party and wantsto do good by the other party, regardless of ego-centric profit motives (Jarvenpaa et al. 1998;Mayer et al. 1995). Variations of relational trustinclude normative trust (Child 1998), goodwill trust(Sako 1991, 1992, 1998), affect-based trust(McAllister 1995), identification trust (Lewicki andBunker 1996), companion trust (Newall and Swan2000), and fairness (Zaheer et al. 1998). A moti-vation to do good by the other party is key tothese definitions. These definitions all include oneparty empathizing with the other party, and spe-cifically excludes the notion of calculative trust.

However, there are significant differences in thesedefinitions as well. Some definitions may or maynot include friendship (Lewicki and Bunker 1996;Newell and Swan 2000), affect (Kramer 1999;McAllister 1995; Zaheer et al. 1998), shared iden-tity (Lewicki and Bunker 1996), goodwill (Newelland Swan 2000), common values (Child 1998;Lane 1998), mutual understanding (Lewicki andBunker 1996) and dependability (Zaheer et al.1998). Some researchers also include cognitiveand motivational underpinnings of relational trust(Kramer 1999). For our purposes, the key to rela-tional trust is that one party empathizes with theother party and wants to do good by them foraltruistic reasons. Relational trust is thought to beespecially important to the success of colla-borative activities (Jarvenpaa and Leidner 1998;Sako 1998). However, prior research (McAllister1995) has found only partial support for a relation-ship between performance and affect-based inter-personal trust. This leads to our third hypothesis.

H3: There is a positive associationbetween relational interpersonal trustand virtual collaborative relationshipperformance.

Integrated Trust

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

1995). Trust can take different forms in differentrelationships, and different forms of trust may mixtogether and interact in some situations. Trustmay have a bandwidth that can vary in both scopeand degree (Rousseau et al. 1998). For example,some relationships may rely more on a combina-tion of calculative and competence trust, whileother relationships may be based more on acombination of relational and competence trust.Further, one type of trust may evolve into another,deeper type of trust. Rousseau et al. (1998) spe-culated that in terms of interpersonal trust, calcu-lative trust was more important in the early stagesof a relationship, while relational trust was moreinfluential in the later stages. Other authors haveproposed similar ideas, with the initial creation ofcalculative trust paving the way for the equivalentof relational trust development (Bachman 1998;Child 1998; Lewicki and Bunker 1996; Newall andSwan 2000).

In addition, there is the issue of whether thevarious types of trust can compensate for eachother (Mayer et al. 1995). Can highly positiverelational trust compensate for negative calcula-tive trust, and vice versa? Can negative compe-tence trust be offset by highly positive calculativeand/or relational trust? While we do not specify atemporal relationship between the different inter-personal types of trust, we argue that the differenttypes of interpersonal trust affect and are affectedby each other, and that a combination of the dif-ferent types of interpersonal trust together impactcollaborative relationship performance. However,a prior study (Zaheer et al. 1998) did not find sup-port for a relationship between integrated inter-personal trust and performance. This leads to ourfinal hypothesis.

H4: There is a positive associationbetween integrated trust and virtualcollaborative relationship performance.

Types of Trust Excludedfrom the Model

A number of types of trust identified in the litera-ture were excluded from our model because wefelt that they either were not a type of inter-



personal trust or, given our definitions, were effec-tively included in the model. Foremost amongthese was cognition trust (Child 1998), also calledcognition-based trust (McAllister 1995), andknowledge-based trust (Lewicki and Bunker1996). We chose to exclude cognition trust be-cause the definitions of it were often contradictoryor overlapped with the definitions in our model.Some authors included dependability (McAllister1995), reliability (McAllister 1995), and compe-tence and/or benevolence (Jarvenpaa et al. 1998;Mayer et al. 1995; McKnight et al. 1998) withintheir definitions of cognitive trust.

Trust as a psychological state implies that all trustis cognitive in that it is based on beliefs aboutanother party. The types of trust in our model(calculative, competence, relational) are all cogni-tive in that they are judgments or beliefs. Theconfusing definitions of cognition trust may explainwhy prior research (McAllister 1995) had difficultyin testing a relationship between cognition trustand performance.

A common aspect of cognition trust is theinclusion of predictability, which is the degree ofconsistency in intended behavior. It is also per-ceived as a type of trust in and of itself (Zaheer etal. 1998). We have not included predictability asa type of trust because predictability is not enoughto explain trust (Bachman 1998), and to bemeaningful, trust must go beyond predictability(Deutsch 1958; Mayer et al. 1995). Instead, weinclude predictability from both an economicstandpoint and a relational standpoint in two of ourtypes of trust. In calculative trust, one party isperceived as predictable if it is in their own self-interest to do so, while in relational trust, one partyis seen as predictable because they empathizewith the other party.

Some authors include reliability and/or depend-ability in their definitions of competence trust orcognition trust, where reliability is whether theother party can be relied on to fulfill their obliga-tions (Anderson and Weitz 1989; Zaheer et al.1998). Reliability is also sometimes perceived asa type of trust in and of itself (Zaheer et al. 1998).Reliability was excluded from our model becauseprofessional ethos and fear of malpractice tended

to ensure that one party thought the other partyreliable prior to agreeing to engage in the tele-medicine relationship. It was thus a necessaryprecondition 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 ourmodel because we agree with the assessment ofother researchers (Hagen and Choe 1998;Rousseau et al. 1998). The ability to imposecostly sanctions on the other party is a substitutefor trust. It is not a form of trust. We excludedexperienced-based trust from our model. Experi-ence-based trust is derived from a conceptuali-zation of interpersonal trust as being based onpast interactions with the other party (Deutsch1958, 1960; Kramer 1999). However, Luhmann(1979, 1988) argued that familiarity was a precon-dition 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 inter-personal trust with VCR performance in the healthcare delivery environment. This environment wasselected because it is a knowledge-based industrywhose body of knowledge is expanding rapidly.Traditionally, health care delivery has utilizedcollaboration as a means of complexity reduction(IOM 1996a; Silberman 1992). Advances in infor-mation technology make virtual collaborationpossible. Telemedicine is the process of two ormore geographically separated health care pro-viders collaborating via information technology toprovide value-added health care delivery (IOM1996b), and it is usually believed that telemedicinecan increase the access to and quality of healthcare 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



it exists and the emergent nature of VCRs ingeneral, comparative case studies were utilized asthe primary research design because they cancapture the nuances and the richness of thesephenomena and increase the robustness andgeneralizability of findings through replication (Yin1994). To enhance methodological rigor, weutilized facet theory (Borg and Groenen 1997;Borg and Shye 1995; Canter 1985a; Guttman1959, 1968, 1982a; Levy 1994; Shye et al. 1994),a systematic approach to research design anddata analysis developed in the field of psychology.The design of the study may appear to be closelyrelated to grounded theory; however, it isn�t. Wechose facet theory instead of grounded theorybecause we had some a priori hypotheses wewanted to test. Facet theory can be complemen-tary to grounded theory in that it provides a meansto structure the theory generating and testingprocess.

Appendix A presents an overview of facet theoryand Table A1 presents a glossary of facet theoryterms for reference. Figure 2 presents the facettheory mapping sentence used for this research.The unit of analysis was the telemedicinerelationship�multiple health care providers utili-zing telemedicine equipment to collaborate. Twointernal content facets were identified. Facet Awas the collaboration/environmental interactionfacet representing both the telemedicine rela-tionship�s interpersonal trust and impact values.Facet B was the agency facet representing thedifferent parties of the telemedicine relationship�the health sciences center (HSC) and remote site.Facet A was somewhat unusual in that the hypo-thesized correlates were elements of the samefacet, where interpersonal trust and impact wereregarded as two aspects of interaction betweenthe collaborative relationship and its environment,with the former acting as an input and the latteracting as an output. In this facet, interpersonaltrust was further broken down to self-interest,ability, empathy, and integrated interpersonaltrust, and impact was further broken down toaccess, cost, and quality. Calculative trust wasoperationalized as self-interest, defined as theextent to which the individuals in one party per-ceive the benefits from directly participating in thecollaborative relationship are greater than the

costs (Kramer 1999; Lewicki and Bunker 1996;Williamson 1993), while competence trust wasoperationalized as ability, defined as the degree towhich the individuals in one party perceive theindividuals in the other party as having the neces-sary 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 inone party desire to do well by those in the otherparty (Jarvenpaa et al. 1998; Tyler and Kramer1996).

Telemedicine�s impact on remote site health caredelivery was measured as the perceived changerelative to the conditions prior to the advent oftelemedicine. Access is the timely use of per-sonal health services to achieve the best possiblehealth outcomes (IOM 1993). Quality is the de-gree to which health care services for individualsand populations increase the likelihood of desiredhealth outcomes and are consistent with currentprofessional knowledge (IOM 1990). The impacton cost is relative to the cost of similar episodes ofmedical care prior to the start of the telemedicineproject and therefore excluded both the operatingand capital costs of the telemedicine equipment.All of these elements were present in all of therelationships (as represented by the bracketsaround Facet A�s elements), and the range facetto the right of the arrow represented all of thepossible values for all of Facet A�s elements�verypositive, positive, slightly positive, neither positivenor negative, slightly negative, negative, and verynegative.

Sample

Three telemedicine networks located in the UnitedStates and involving at least three operationaltelemedicine projects were studied. Each of thenetworks had at its hub a university-affiliatedhealth sciences center (HSC), and the spokes ofthe networks were located in rural areas whereper capita income levels were substantially belowthe national level (Census Bureau 1990). HSCswere selected because the vast majority of tele-medicine projects involve university-affiliated



In telemedicine relationship (x), the extent of

A: Collaboration-EnvironmentalInteraction

B: Agency Range Facet

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}

asreported

byindividuals

at

(1. HCS)(2. Remote Site)

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

wasin the sense of facet A

(Collaboration-Environmental

Interaction)

In telemedicine relationship (x), the extent of

A: Collaboration-EnvironmentalInteraction

B: Agency Range Facet

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}

asreported

byindividuals

at

(1. HCS)(2. Remote Site)

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

wasin the sense of facet A

(Collaboration-Environmental

Interaction)

Figure 2. Mapping Sentence of Model

health sciences (or medical) centers (IOM 1996b;ORHP 1997), and they are charged with providinghealth care to remote areas. The telemedicineprojects studied were part of the normal practiceof medicine either as revenue generating or costreducing projects. Ten telemedicine projects in-volving five teleconsultation, three distancelearning, and two teleradiology telemedicineactivities were examined. Within each of thesetelemedicine projects, two relationships werestudied: the relationship between the relevantindividuals at the HSC and those at the remotesite, and the relationship between the individualsat the remote site and those at the HSC.

Theoretical (or purposeful) sampling was utilizedin an effort to address potential threats to theexternal validity and construct validity of thisresearch (Strauss and Corbin 1998). HSCs wereselected because they and their telemedicineproject partners tended to have certain charac-teristics that naturally accounted for alternativeexplanations for the impact or the lack thereof ofinstalled telemedicine projects (Paul et al. 1999).For example, all of the HSCs had a mission toimprove health care for their respective ruralpopulations, and all of the specialists were paid asalary by the state. Therefore, the lack of specia-list reimbursement was less likely to be an in-hibiting factor in the short run. All of the tele-medicine projects studied were intrastate, eli-minating potential interstate physician licensing

barriers. Malpractice liability insurance concernswere less an issue because all of the specialistswere engaged in HSC sanctioned telemedicineprojects; therefore, their activities were covered bytheir respective facility�s umbrella liability cover-age. High start-up and operating costs were notinhibitors because almost all of the projects in thisresearch received external funding.

Site selection was based on four criteria. First,each site had to have at least three activetelemedicine projects. Second, each site had tohave one of each of the three types of telemedi-cine activities: teleconsultation, distance learning,and teleradiology. These two criteria enabledboth within and between network comparisons ofdifferent telemedicine projects. Third, the sitescould not involve military or correction facilitiesbecause the voluntariness of participation and thedynamics of trust in such situations may bedifferent from those in civilian projects. Fourth,each site had to have been operational for aminimum of six months to allow the inevitabletechnological and procedural bugs to be ad-dressed and to allow the novelty of telemedicineto pass.2

2The one exception was a pediatric oncology tele-consultation project that was discontinued by thepediatric oncologists after a period of four months.



The World Wide Web was searched to find sitesthat met these criteria. The second criterion�different types of telemedicine activities withineach project�was discarded because sitesmeeting this criterion could not be found.Although a number of potential sites claimed tohave all three types of telemedicine activitiesoperational at the time of this study, only oneactually did. Indeed, a number of potential sitesthat claimed on their Web pages to have activetelemedicine projects did not have any activetelemedicine projects at the time of this study.This exaggeration of the state of active tele-medicine projects was not uncommon. TheORHP (1997) found that approximately 25 percentof the hospitals they surveyed which claimed tohave at least one active telemedicine project infact had no operational telemedicine projects.Each site selected included at least one tele-consultation project, which enabled teleconsul-tation activities to be compared across thetelemedicine networks. Both distance learningand teleradiology projects occurred at two sites,enabling at least one between network com-parison for these telemedicine activities.

Table 1 presents a summary of the telemedicinerelationships studied. Two of the five telecon-sultation projects involved multiple specialtieswhere the patient and his or her family wereusually present during the sessions. The multiplespecialties teleconsultation project involvingprimary care physicians at a rural hospital wasjudged to have a positive impact on remote sitehealth care, while the other involving a physicianassistant at a rural health clinic was not. Aninfectious diseases teleconsultation projectbetween HSC specialists and a rural hospital�sprimary care physicians with patients usually notpresent during the sessions was judged to have avery positive impact on remote health caredelivery while a pediatric oncology teleconsul-tation project involving pediatric oncologists at theHSC and nurses with patients and their parentspresent during the sessions was not. A bonemarrow transplant teleconsultation involving HSCtransplant specialists, nurses, psychologists, andadministrative staff and two rural oncologists at aprivate clinic with the patient and his or her familyalmost always present during the sessions was

judged to have a positive impact on remote sitehealth care delivery. This teleconsultation projectwas utilized for initial consultations to determine ifthe patient was a viable candidate physically for atransplant and psychologically for the extendedstay in isolation the transplant entailed, andwhether the patient wanted to undergo a highmortality treatment for a life-threatening diseasewith these particular specialists. It was also usedfor follow-up after the transplantation procedure.

A distance learning rural primary care residencytelemedicine project involving HSC specialists andresidents receiving specialized training in ruralhealth care at a regional medical center where thepatients were sometimes present during the ses-sions was deemed to have a positive impact onremote site health care delivery, while a distancelearning rural primary care clerkship telemedicineproject involving HSC specialists and third-yearmedical students at a primary care clinic withoutpatients present during the sessions was not. Adistance learning rural telemedicine project in-volving HSC specialists and a podiatry residencyrotation for the HSC�s first-year podiatry residentsat a group of three federally funded health clinicsand the local state hospital where the patientswere sometimes present during the sessions wasdeemed to have a very positive impact on remotesite health care delivery. Diabetes was a majorhealth care problem in the region, and the lack ofaccess to preventive care due to economic factorsfor the population most at risk resulted in the re-gion having a rate of amputations among diabeticsthat was significantly above the national average.

The two teleradiology projects involved digitizedradiographic images being sent to the respectiveHSCs, where the radiologists would read theimages and provide a diagnosis by e-mail ortelephone, depending on the urgency of thesituation. In some cases, the radiologists wouldtelephone the rural primary care physicians foradditional information. Both teleradiology projectshad undergone at least a four month trial periodduring which the transmitted image quality wasdeemed to be an adequate basis for making adiagnosis. One teleradiology project was deter-mined to have a positive impact on remote sitehealth care, while the other was not.



Table 1. Telemedicine Relationships Overview

Relation-ship Type*

Activity(Project Duration)

Team Composition &Patient Involvement

TeamStability

Score Struc-tuple & (Impact

Values)***

A HSC!RTeleradiology

(2 years)

Radiologist Stable217(12)

B R!HSCAttending Primary Care

Physician (PCP)Stable

772(12)

C HSC!R DistanceLearning�Rural

Residency(1 year)

Supervising Physician(SP) & Rotating

Specialists,Patient (sometimes)

Semi-stable

777(21)

D R!HSCResidents & SPs,

Patient (sometimes)Stable

747(21)

E HSC!R Teleconsultation�Oncology

(9 months)

Oncologists, Nurses,Psychologist, Admin.

Stable747(19)

F R!HSCOncologist, Patient &

Family (always)Stable

777(19)

G HSC!R Teleconsultation�Pediatric Oncology

(4 months)**

Pediatric Oncologists Stable146(11)

H R!HSCNurse, Patient &Parents (always)

Stable742(11)

I HSC!R DistanceLearning�MedicalStudent Clerkship

(1 year)

SP & RotatingSpecialists, Patient

(rarely)

Semi-stable

267(12)

J R!HSC Medical Students Stable746(12)

K HSC!R DistanceLearning�Podiatry

Residency(9 months)

SP & Residents,Patient (sometimes)

Stable777(21)

L R!HSCResidents, SP, & Area

Podiatrists, Patient(sometimes)

Stable775(21)

M HSC!R Teleconsultation�Infectious Diseases

(1.5 years)

Infectious DiseasesSpecialist & OtherSpecialists (rarely)

Stable777(21)

N R!HSC PCPs Stable777(21)

O HSC!RTeleradiology

(1 year)

Radiologist Stable447(16)

P R!HSC Attending PCP Stable765(16)



Table 1. Telemedicine Relationships Overview (Continued)

Relation-ship Type*

Activity(Project Duration)

Team Composition &Patient Involvement

TeamStability

Score Struc-tuple & (Impact

Values)***

Q HSC!R Teleconsultation�Multiple Specialties

(8 years)

Different Specialists,TelemedicineAdministrator

Semi-stable

771(21)

R R!HSCPCP & Patient(sometimes)

Stable777(21)

S HSC!R Teleconsultation�Multiple Specialties

(5 years)

Different Specialists,TelemedicineAdministrator

Semi-stable

441(12)

T R!HSCPhysician Assistant,Patient (sometimes)

Stable753(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 keyinformants provided thick and richly textured data(Orlikowski 1993; Sackmann 1991) and eliminatedthe problem of item non-response which plaguedearlier telemedicine studies (ORHP 1997). In all,74 health care professionals were interviewedface-to-face, and the interviews were audiotapedand transcribed. Key informants were membersof one of three groups: clinicians (physicians,physician assistants, or nurse practitioners),administrators, and information technology profes-sionals.3 They were selected based on current or

past direct involvement in their organization�stelemedicine project. Figure 3 presents a sum-mary of the key informants by position and loca-tion. The interviews were approximately equallysplit between those conducted at the HSCs andthe remote sites, and the number of interviews perHSC was proportional to the number of tele-medicine projects they had active. Further, theproportion of key informants who were clinicians,administrators, and IT professionals was fairlyevenly distributed across the sites.

Construct validity and reliability were enhanced bytriangulated data collection (Eisenhardt 1989; Yin1994). This was achieved by interviewing multiplekey informants and collecting multiple types ofdata. Teleconsultations or videotapes of telecon-sultations were observed when possible, anddocumentation and archival data such as grantproposals and follow-up, needs assessments, andstrategic plans were collected when available.

3In some of the smaller health care facilities, individualsoften played dual roles. For example, at one remotesite, the residency program administrator was also aphysician who taught courses. In these cases, key infor-mants were classified only in one role but were askedquestions about both roles played.



HSC I Admin. IT Prof.Health

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

Health Care Prof.

Total by Location

HSC (A,C,E) 4 4 5 13 HSC (G,I,K,M) 4 3 8 15B 1 1 2 4 H 1 3 0 4D 1 2 1 4 J 1 1 3 5F 1 0 2 3 L 1 0 2 3

Total 7 7 10 24 *N 0 0 1 1Total 7 7 14 28

HSC III Admin. IT Prof.Health

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

Health Care Prof.

Total by Location

HSC (O,Q,S) 3 3 5 11P 1 1 2 4 HSC I 7 7 10 24R 1 3 2 6 HSC II 7 7 14 28T 0 0 1 1 HSC III 5 7 10 22

Total 5 7 10 22 Total 19 21 34 74

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

HSC I Admin. IT Prof.Health

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

Health Care Prof.

Total by Location

HSC (A,C,E) 4 4 5 13 HSC (G,I,K,M) 4 3 8 15B 1 1 2 4 H 1 3 0 4D 1 2 1 4 J 1 1 3 5F 1 0 2 3 L 1 0 2 3

Total 7 7 10 24 *N 0 0 1 1Total 7 7 14 28

HSC III Admin. IT Prof.Health

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

Health Care Prof.

Total by Location

HSC (O,Q,S) 3 3 5 11P 1 1 2 4 HSC I 7 7 10 24R 1 3 2 6 HSC II 7 7 14 28T 0 0 1 1 HSC III 5 7 10 22

Total 5 7 10 22 Total 19 21 34 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 wasused as legal issues prevented the researchersfrom having access to patient records, and mosttelemedicine sites tended not to maintain suchrecords (ORHP 1997). The impact on remotehealth care delivery was consistent with the con-ceptualization of the outcome of collaboration asthe value added in terms of complexity reductionand the impact of trust on collaborative rela-tionships as improved performance (Goshal andMoran 1996; Sako 1998). It was also consistentwith the output of collaboration being �theenhancement of transaction value� (Zaheer et al.1998, p. 155).

Data Analysis

Qualitative Analysis

The transcribed interviews were analyzed andcoded. The coding scheme was theoreticallybased (Martin and Turner 1986), where quotationswere categorized according to facet and elementvalues. Internal validity was enhanced through

the use of pattern matching (Strauss and Corbin1998; Yin 1994), and the explicit specification ofthe mapping sentence provided structure by whichto engage in constant comparative analysis bothwithin and between cases.

Key informant interviews established the need forthe telemedicine projects and eliminated alter-native explanations such as technology problemsfor the failure of a telemedicine project. They alsoestablished trust as the critical factor for thesuccess or failure of the telemedicine relation-ships, and were used to determine the type andlevel of trust involved in the different telemedicinerelationships. Key informant interviews were alsoused to assess the impact the telemedicine pro-ject had on the cost, quality of, and access toremote site health care. Note that the codingexamples presented in the Qualitative Resultssection involve only one key informant. In mostcases (including those presented), multiple con-firming comments from different key informantsinvolved in that particular telemedicine relation-ship, as well as other forms of evidence, wereused to determine the coding value.



Table 2. Weak Monotonicity Coefficient (Guttman�s mu2) MatrixABILITY EMPATHY SELFINT HSC REMOTE IMPACT ACCESS COST QUALITY

ABILITY 1.000

EMPATHY 0.868 1.000

SELFINT -0.311 -0.016 1.000

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 1.000ACCESS 0.913 0.812 0.582 0.000 0.000 1.000 1.000COST 0.914 0.802 0.602 0.000 0.000 0.997 0.954 1.000QUALITY 0.821 0.768 0.623 0.000 0.000 0.988 0.987 0.920 1.000

Table 3. Impact Values by Telemedicine RelationshipA & 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 4Quality 5 7 6 4 4 7 7 7 7 4Cost 2 7 7 3 4 7 7 4 7 4OVERALL 12 21 19 11 12 21 21 16 21 12

Quantitative Analysis

The coded qualitative data were converted intoquantitative data using an ordinal seven-pointLikert scale for the trust and perceived impactelements. The reliability of coding of the quali-tative data was enhanced by the use of computeraided qualitative data analysis software (Kelle1995; Morris 1994; Wolfe et al. 1993). In addition,the researchers� coding of the trust and impactvariables for each telemedicine relationship wasassessed by an information systems professorwhose research interests included trust. The thirdparty assessor concurred with the researchers�coding 94 percent of the time.

Table 2 presents the weak monotonicity coeffi-cient (Guttman�s mu2) matrix for the elements ofthe two content facets. The original conceptuali-zation of the model anticipated analyzing theassociation between the different types of trustand how they were perceived as impacting theaccess, cost, and quality of health care delivery in

the remote areas. However, as exhibited inTable 2, the impact values for each site�access,cost, and quality�were highly correlated (theminimum correlation was 0.92), and testing eachimpact type was unlikely to provide additionalmeaningful information or insights. Therefore, thethree impact types were summed for each rela-tionship, and only this one figure, the overall im-pact on health care delivery (IMPACT), was usedin the data analysis. The maximum score pos-sible for impact was 21 points and the minimumpossible score was 3 points. In this research, thehighest score was 21 points, and the lowest scorewas 11 points. The impact facet was unbalancedin that impact values could be positive and verypositive but, from a practical standpoint, could notbe negative or very negative because it wasdifficult for telemedicine relationships to have anegative or very negative impact in terms of eitheraccess to, or quality of, remote site health caredelivery. Table 3 presents the impact scores foreach of the telemedicine projects.



Model Dimensionality and Construct Validity

Smallest space analysis (SSA) was used to evalu-ate the dimensionality and construct validity of themodel, and a two-dimensional model was selectedas the best model. The coefficient of alienation forthe two-dimensional model was 0.036, which indi-cated a good fit for the monotonic relationshipbetween input coefficients and output distances(Guttman 1968). The proportion of variance ex-plained (RSQ) value was 0.997, and neither it northe coefficient of alienation could be significantlyimproved upon by adding additional dimensions tothe model. The shape of the Shepherd diagramfor the two dimensional model (exhibited inFigure 4)�negatively sloped monotone series ofpoints�indicated satisfactory goodness-of-fit forthe model. The partitioning of the SSA space (ex-hibited in Figure 5) indicated a strong correspon-dence between the mapping sentence and thespatial contiguity of the data (Borg and Shye1995; Shye et al. 1994). All of this suggested thatthe model presented in Figure 1 had good con-struct validity.

Hypotheses Testing

Partial order scalogram analysis with base coordi-nates (POSAC) was used to test the hypotheses.Table 4 presents the POSAC coordinates for thecalculative trust and agency elements� values foreach profile and Figure 6 exhibits the POSACdiagram of the different telemedicine relationshipssuggested in Hypothesis 1. In Figure 6, agency isrepresented by HSC and R, which representhealth sciences centers and remotes sites, res-pectively. The number in brackets is the calcu-lative trust (internal content facet element) valuefor those relationships (Borg and Shye 1995;Brown 1985). The final number(s) is(are) theexternal item�perceived impact value for thoserelationships with the corresponding calculativetrust value.4 The model�s CORREP value,

exhibited at the bottom of Table 5, was 0.9800(out of a possible 1.000), which indicated anexcellent fit of the profiles in the POSAC space interms of their relationship to each other.

The perceived impact on remote site health caredelivery element values are usually exhibited bythemselves in the external item diagram. Thepatterns formed by these values represented theregional hypotheses and were used to test thehypotheses by assessing whether there was arelationship between the content facets� elementsand the external item (Levy and Guttman 1985;Shye and Amar 1985). The hypothesis is thehigher the order of a telemedicine relationship asrepresented by its interpersonal trust values, orsome combination thereof, then the greater theperceived impact of that telemedicine relationshipon the remote site�s health care delivery. InFigure 6, those relationships whose perceivedimpact value is consistent with the hypothesizedrelationship with calculative trust are in plain text,while those that are not are bolded and italicized.

The weak monotonicity coefficient values of cal-culative trust, HSC, and the perceived impact onremote site health care delivery elements in rela-tion to the X, Y, joint, and lateral axis are exhibitedin Table 5. The coefficient values were used todetermine the roles the different facets and theirelements played and whether the hypothesis wasempirically supported. The X axis representedself-interest, the calculative trust facet elementwhose value increased as the X value increased.The Y axis represented the agency facet, with theremote sites located on the top half of the POSACdiagram and the HSCs located on the bottom halfof the diagram. The hypothesis was supported ifthe perceived impact�s weak monotonicity coeffi-cient value with the joint axis approached one andits value with the lateral axis approached zero.This support also could be viewed graphically inthe POSAC diagram if the telemedicine relation-

4It is standard practice in POSAC diagrams to label thepoints by either their profile ID or their structuple values.A separate diagram, called external item diagram, whichsuperimposes the external item values for each profile in

the POSAC diagram, is usually presented by itself andis the means by which hypotheses are tested (Levy andGuttman 1985; Shye and Amar 1985). Due to spaceconsiderations, the POSAC diagram and the externalitem diagram were combined in this paper.



-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5Data

0

1

2

3

Dis

tanc

es

-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5Data

0

1

2

3

Dis

tanc

es

-2 -1

0 1 2Dimension-1

-2

-1

0

1

2

Dim

ensi

on-2

REMOTEABILITY

EMPATHY

HSC

SELFINT

IMPACT

-2 -1

0 1 2Dimension-1

-2

-1

0

1

2

Dim

ensi

on-2

REMOTEABILITY

EMPATHY

HSC

SELFINT

IMPACT

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

Figure 5. Two Dimensional SSA of VCR Performance Model



0.0 0.2 0.4 0.6 0.8 1.0Self-Interest

0.0

0.2

0.4

0.6

0.8

1.0

Age

ncy

R [2] 11, 12

R [5] 16, 21

R [3] 12

R [6] 12

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

HSC [6] 11

HSC [1]12, 21

R [7] 19,21, 21, 21

0.0 0.2 0.4 0.6 0.8 1.0Self-Interest

0.0

0.2

0.4

0.6

0.8

1.0

Age

ncy

R [2] 11, 12

R [5] 16, 21

R [3] 12

R [6] 12

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

HSC [6] 11

HSC [1]12, 21

R [7] 19,21, 21, 21

Table 4. POSAC Coordinates and Coefficients for Calculative Trust (H1)TelemedicineRelationship Agency

Self-interest X Y Joint Lateral Impact

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

Proportion of profile pairs correctly represented (CORREP): 0.980

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



Table 5. H1: Calculative Trust Weak Monotonicity Coefficient MatrixX 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 werelocated in the upper right-hand corner of thePOSAC diagram, partitioned by a line parallel tothe lateral axis.

Results

Qualitative Results

Interviews with key informants established theneed for the telemedicine projects. For example,prior to the infectious diseases teleconsultationproject, the infectious diseases patients�themajority of whom were uninsured�often incurredadditional health problems that required specia-lists 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 300miles away, the medical center would not acceptthese patients because their contract with thestate did not allow for the reimbursement forexpenses incurred as a result of providing care touninsured patients not in stable condition. Thelocal state hospital physicians perceived theteleconsultation project as a means of providingthe specialty care the infectious diseases patientsneeded to survive.

State Hospital Physician: The worstcase scenarios are our infectious dis-eases cases again. They [the medicalcenter 300 miles away] don�t want totouch them with a 10 foot pole. We�veeven gotten feedback that they�re notsalvageable. Well, we�ve lost two pa-tients out of 40�both of them died ofheart disease, not of infectious diseases.

Our cure rates are looking like they aregoing to be very good. But in the mean-time, we�ve got to get them through theirmassive bleeds and stuff like that and ifno one wants to help us take care ofthem because they don�t think they�reviable, we�re stuck literally putting yourfinger in the dike and it�s a real uncom-fortable feeling as a clinician. We�retrying to fix that real quickly, but in[state]�you always sent the patientwhere they needed to go. Would youlike to get in the ambulance or the life-flight helicopter with somebody with avery contagious disease?

Key informant interviews eliminated alternativeexplanations such as technology problems for thefailure of a telemedicine project and establishedtrust as a critical factor. Questions about the us-ability, reliability, and sufficiency of the technologywere specifically asked, but there was no indica-tion that the technology was a cause of projectfailure, nor were there any indications that therewas a lack of interpersonal trust of the technolo-gists. For example, a pediatric oncologist involvedin a teleconsultation project felt the technologyworked but the teleconsultation project failed be-cause of the nurses at the remote state hospital.

HSC Pediatric Oncologist: I think over-all, we thought we could do a really goodjob with [the telemedicine equipment].The only�the only�problem there waswas with the operator on the other end.

Key informant interviews also established theimportance of trust in telemedicine, particularlyprojects involving multidisciplinary teams. There



was a tendency for HSC specialists to denigratethe ability of the remote physicians, and remotephysicians would often refuse to work with specia-lists who, consciously or not, treated them assecond class citizens. A rural hospital physicianinvolved with the multiple specialties teleconsul-tation project discussed what it was like workingwith the HSC specialists:

Interviewer: How has it been dealingwith the doctors up there (the HSC)?

Remote Physician: Well, pretty good ingeneral, and that�s an important key tothe success of the program, because,you know, if the doctors make you lookgood, and let you learn something, thenyou�ll want to come back. But if theymake you look bad, then you don�t wantto do it again. And most of the doctorsare really good about...you know, even ifthey feel like you�ve done exactly thewrong thing, they�ll say, �Yes, that wasan excellent thought. And I would ask acouple of other things.� But there aresome guys that are jerks over it. Therewas one infectious disease guy that,wooh! I mean, he would sit back and hewould say, �What kind of bull****, whatkind of [jerk], nah, nah, nah, nah....� Andyou could hear him off the screen. Youknow, I could hear him talking about thisoff [camera], you know, and...we�d justnever consult him again. You�d say,�Never, ever get this guy again.�

The interviews were also utilized to determine thetype and level of trust involved in the differenttelemedicine relationships. For example, the pedi-atric oncologists had a very negative assessmentof the ability of the local state hospital�s nursesinvolved in the teleconsultation project. They feltthe nurses did not have the training to perform thenecessary neurological and abdominal exams.

HSC Pediatric Oncologist: In order for it[telemedicine] to work, a condition that isabsolutely necessary is to have a quali-fied person that will link the patient and

the physician. And that person has to bea registered nurse practitioner or a physi-cian�s assistant. Because you�re prac-ticing medicine and seeing the patientand then treating the patient and [youare] responsible for what happens to thepatient, then that person [on the remoteend] needs to do a good physical exami-nation with an abdominal exam, heartsounds, and everything. They need torelate to you what is happening to thepatient�.They [physicians� assistantsand nurse practitioners] are fully capa-ble. They have done thousands of phy-sical examinations; they know what�snormal, so whatever is abnormal theyreport. It just takes some training to�the hardest thing to do is a neurologicalexam and an abdominal [exam]probably.

Interviewer: Because?

HSC Pediatric Oncologist: Neurologicalexam takes some training, specialtraining in trying to do it right and inter-preting it right, and then the abdominalexam�you really have to put your handsin there and try to feel the liver and thespleen�and that takes awhile to developthe expertise. And if you miss it, it�sserious. So you have to have somebodywell trained because if you are not ableto find out that person [the patient] hasthat finding, it may be a serious problemof missing very important information.That�s why I think the link between thepatient and [specialist] physician is abso-lutely necessary. You cannot practicetelemedicine without that link, I think�practice it safely and without decreasingthe standards of care, because if youhave somebody who�s not capable andyou see a patient with fever and youdon�t do an exam and you don�t knowthat the patient has a large spleen, yousee, you misdiagnose. You have halfthe information that you really have tohave. That link is very important.



In the case of the bone marrow transplant tele-consultation project, the rural oncologist originallywas suspicious of motives of the the HSC�s sub-specialists, but that concern dissipated and henow perceived himself as being a partner withinstead of a lackey of the HSC�s subspecialists.

Rural Oncologist: There�s been�like Isaid at the beginning�so many peopleasked me about that you feel like you�rekind of being like a lackey or some-thing�you know just like a studentrather than a doctor, not playing a majorrole and I think that, now I feel like a totalpartner even though it�s a transplant, Ifeel like I�m doing as much as they areeven though I�m not getting into themechanics of it and I think that that�show it�s evolved to that level and so atthe beginning, I started having my owndoubts about well I�m just letting thesepeople come into our office and dictatingthe care of my patients. But it�s evolvedso that, I didn�t feel that way at the be-ginning but so many people brought thatup that I was beginning to wonder andnow I feel just the opposite, that I have alot more control�.I think that because ofthe communication that�s developed andit�s more than just having the tele-medicine, it�s talking to these doctorsand gaining confidence both ways thatwe can communicate well, that werespect each other�s ideas, we have afeeling of how things are done andthey�re more comfortable about sendingthe patients back.

Key informant interviews were also used to assessthe impact the telemedicine project had on thequality of and access to remote site health care.A physician at a rural hospital involved in amultiple specialties teleconsultation project talkedabout how teleconsultations improved both thequality and access to care. Prior to telemedicine,the rural physicians were on their own when itcame to difficult or rare cases because the localpopulation was too poor to afford to travel to theHSC or its closer affiliated center.

Interviewer: How were some of thesecases handled prior to telemedicine?

Remote Physician: We just took a wildguess. We did, because a lot of thesefolks can�t get up there. You know, youcan try calling, but it�s not the same thingas looking at it, and so a lot of it wasjust....You just used your best judgmentand went on...

Interviewer: They couldn�t get there interms of�

Remote Physician: Economics�nor-mally, it�s economics. Physical conditionoccasionally, but mostly, economics.Most of the people in [rural area] do nothave insurance. That�s an economic factof life.

Interviewer: Are they covered byMedicaid?

Remote Physician: Doesn�t make anydifference if you have Traveler�s Insur-ance. If you don�t have enough moneyto put [gas] in your gas tank, you can�tget to the doctor. That�s the commonproblem.

Key informant interviews were also used to assessthe impact the telemedicine project had on thecost of remote site health care. For example, thebone marrow transplant teleconsultation projectallowed patients to return home earlier. Prior tothe teleconsultation project, the rural oncologist�spatients who underwent the bone marrow trans-plant procedure were severely immuno-compro-mised and had to spend 6 to 10 weeks living inshort-term apartments near the HSC in order toundergo blood transfusions or other types oftherapy on an outpatient basis. With the advent ofthe teleconsultation project, these patients (andmembers of their family who often stayed withthem at the HSC apartments) could now returnhome after three weeks in the hospital and receivemuch of their post-transplant support at theoncologists� clinic. As a result, the patients and



their insurance companies were spared the costof the short-stay apartment rental, and the tele-consultation project enabled the patient�s friendsor relatives to continue working on at least a part-time basis while the patient underwent post-transplant activities at the local oncologist�s office.

Rural Oncologist: They [the patient�sfamily] get home. They don�t have topay for a hotel room. It�s no little thingthat a lot of them can still work here andkeep their health insurance or whatever.So it�s kind of hard to explain to peoplebut I think most people that have afamily�if I think what would happen if Ihad 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 arepeople not even with illness. There area lot of social and economic implications.

Quantitative Results

POSAC analysis was utilized to assess therelationship between individual profiles (thetelemedicine relationships)�represented by theirinterpersonal trust and agency facet values actingas input variables�and the perceived impact onremote site health care delivery facet elementvalues acting as an output variable. An advan-tage of nonparametric MDS is that its analysis withas few as seven data points can be very robust(Borg and Lingoes 1987; Shye 1985a, 1985b),and this data set consisted of 20 telemedicinerelationships where 12 of the 20 (60 percent) wereperceived as positively impacting remote sitehealth care delivery.

Two tables and one figure are presented for eachhypothesis. The first table presents the POSACcoordinates for the relevant trust and agency ele-ments� values for each profile, and the CORREPvalue for the model, and the second tablepresents the weak monotonicity coefficient valuesfor the different element values and the X and Yaxes. The figure presents the different tele-medicine relationships item diagram based on therelevant trust and agency content structuple

values, with the impact on remote site health caredelivery content structuple value superimposed onthe different telemedicine relationships profiles.Those relationships whose perceived impact valueis consistent with the hypothesized relationshipwith type of trust being tested are in plain text,while those whose are not are bolded anditalicized.

H1: Calculative Trust and VCR Performance

The POSAC diagram correctly represented98 percent of the profile pairs. The HSC weakmonotonicity coefficient values in Table 5 indicatethat the X axis represents the self-interest elementvalue, while the Y axis represents the agencyfacet element values. The perceived impact�sweak monotonicity coefficient values with the jointaxis (0.801) and lateral axis (-0.015) approachedone and zero, respectively, and provided supportfor Hypothesis 1. The item diagram in Figure 7also supported the hypothesis in that it was parti-tioned in a manner that it discriminated betweenthose telemedicine relationships positively im-pacting remote site health care delivery and thosewho do not at a rate better than chance. In all, 15of the 20 (75 percent) of the telemedicine relation-ships were correctly discriminated by the optimalpartitioning of the external item diagram. All 12(100 percent) of the telemedicine relationshipshaving a positive impact were correctly discri-minated by the partitioning of the diagram, and 3of the 8 (37.5 percent) telemedicine relationshipsnot having a positive impact were correctly dis-criminated by the partitioning. H1 was thereforesupported, and the impact on remote site healthcare delivery of virtual collaborative relationshipsis monotonically associated with calculative inter-personal trust.

H2: Competence Trust and VCR Performance

The POSAC diagram correctly represented 100percent of the profile pairs. As indicated inTable 7 the weak monotonicity coefficient valuesof the HSC indicates that the X axis represents theagency facet element values, while the Y axis re-



0.0 0.2 0.4 0.6 0.8 1.0Self-Interest

0.0

0.2

0.4

0.6

0.8

1.0

Age

ncy

R [2] 11, 12

R [5] 16, 21

R [3] 12

R [6] 12

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

HSC [6] 11

HSC [1]12, 21

R [7] 19,21, 21, 21

0.0 0.2 0.4 0.6 0.8 1.0Self-Interest

0.0

0.2

0.4

0.6

0.8

1.0

Age

ncy

R [2] 11, 12

R [5] 16, 21

R [3] 12

R [6] 12

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

HSC [6] 11

HSC [1]12, 21

R [7] 19,21, 21, 21

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

Table 6. POSAC Coordinates and Coefficients for Competence Trust (H2)TelemedicineRelationship Agency Ability X Y Joint Lateral Impact

H, B, J, T, P, F, D,L, N, R

Remote 7 0.408 0.913 0.661 0.248 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, 21O, S HSC 4 0.816 0.707 0.762 0.555 12, 16A, I HSC 2 0.707 0.577 0.642 0.565 12, 12G HSC 1 0.577 0.408 0.493 0.585 11


Table 7. H2: Competence Trust Weak Monotonicity Coefficient MatrixX Y JOINT LATERAL

IMPACT 0.3789 0.6728 0.8424 -0.0551ABILITY -0.4278 1.0000 0.7383 -1.0000HSC 1.0000 -1.0000 0.7115 1.0000



0.0 0.2 0.4 0.6 0.8 1.0Agency

0.0

0.2

0.4

0.6

0.8

1.0

Abi

lity

111

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

HSC [1] 11

HSC [7]19, 21, 21, 21, 21

HSC [4]12, 16

HSC [2]12, 12

0.0 0.2 0.4 0.6 0.8 1.0Agency

0.0

0.2

0.4

0.6

0.8

1.0

Abi

lity

111

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

HSC [1] 11

HSC [7]19, 21, 21, 21, 21

HSC [4]12, 16

HSC [2]12, 12

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

presents the ability element. The perceived im-pact�s weak monotonicity coefficient values withthe joint axis (0.842) and lateral axis (-0.055)approached one and zero, respectively, and pro-vided support for Hypothesis 2. The partitioning ofthe item diagram space also provided support forthe hypothesis. As indicated in Figure 8, the itemdiagram was partitioned in a manner that it discri-minated between those telemedicine relationshipshaving a positive impact on the remote site healthcare delivery and those who did not at a ratebetter than chance. In all, 15 of the 20 (75 per-cent) of the telemedicine relationships were cor-rectly discriminated by the optimal partitioning ofthe external item diagram. All 12 (100 percent) ofthe telemedicine relationships having a positiveimpact were correctly discriminated by the parti-tioning of the diagram, and 3 of the 8 (37.5 per-cent) telemedicine relationships not having a posi-tive impact were correctly discriminated by thepartitioning. H2 was therefore supported.

H3: Relational Trust and VCR Performance

The POSAC diagram correctly represented80.1 percent of the profile pairs. As exhibited inTable 9, the weak monotonicity coefficient valuesindicate that the X axis represents the empathyfacet element values, while the Y axis representsthe agency element. The perceived impact�sweak monotonicity coefficient value with the jointaxis of 0.643 was positive but only moderatelyapproached one, and its value with the lateral axisof 0.230 only moderately approached zero. Thisprovided only modest support for the hypothesis.However, the partitioning of the item diagramspace provided additional support for the hypo-thesis. As indicated in Figure 9, the item diagramwas partitioned in a manner that it discriminatedbetween those telemedicine relationships havinga positive impact on the remote site health caredelivery and those that did not at a rate better thanchance. In all, 16 of the 20 (80 percent) of the



Table 8. POSAC Coordinates and Coefficients for Relational Trust (H3)TelemedicineRelationship Agency Empathy X Y Joint Lateral Impact

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

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


Table 9. H3 Relational Interpersonal Trust Weak Monotonicity Coefficient MatrixX Y JOINT LATERAL

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 discri-minated by the optimal partitioning of the externalitem diagram. A total of 11 of the 12 (91.7 per-cent) telemedicine relationships having a positiveimpact were correctly discriminated by the parti-tioning of the diagram, and 5 of the 8 (62.5percent) telemedicine relationships not having apositive impact were correctly discriminated by thepartitioning. H3 was therefore supported.

H4: Integrated Trust and VCR Performance

The POSAC diagram correctly represented 99.1percent of the profile pairs. As indicated inTable 11, the weak monotonicity coefficient valuesof the HSC indicates that the X axis represents theAgency facet element values, while the Y axisrepresents the ability element. Self-interest playsan accentuating role, as illustrated by the invertedL shaped dotted line starting at approximately

0.85 on the Y axis. Empathy plays an attenuatingrole as illustrated by the L shaped dashed linesstarting at approximately 0.21 and 0.45 on the Xaxis which partition the POSAC solution space.The perceived impact�s weak monotonicity coef-ficient value with the joint axis of 0.977 ap-proached one, and its value with the lateral axis of-0.109 approached zero. This provided strongsupport for the hypothesis. The partitioning of theitem diagram space based also provided strongsupport for the hypothesis. As indicated inFigure 10, the item diagram was partitioned in amanner that it discriminated between those tele-medicine relationships having a positive impact onthe remote site health care delivery and those thatdid not at a rate better than chance. In all, 19 ofthe 20 (95 percent) of the telemedicine relation-ships were correctly discriminated by the optimalpartitioning of the external item diagram. A total of11 of the 12 (91.7 percent) telemedicine relation-ships having a positive impact were correctly dis-



0.0 0.2 0.4 0.6 0.8 1.0Empathy

0.0

0.2

0.4

0.6

0.8

1.0

Age

ncy

111

R [4] 11, 12, 21

HSC [1] 12

R [6] 16

HSC [6] 12

HSC [7] 21, 21, 21, 21

HSC [4]11, 12, 16, 19

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

R [5] 12

0.0 0.2 0.4 0.6 0.8 1.0Empathy

0.0

0.2

0.4

0.6

0.8

1.0

Age

ncy

111

R [4] 11, 12, 21

HSC [1] 12

R [6] 16

HSC [6] 12

HSC [7] 21, 21, 21, 21

HSC [4]11, 12, 16, 19

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

R [5] 12

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

criminated by the partitioning of the diagram, andall 8 (100 percent) telemedicine relationships nothaving a positive impact were correctly discri-minated by the partitioning. H4 was thereforestrongly supported.

Discussion

This article examined the association of inter-personal trust with virtual collaborative relationshipperformance. Trust enables collaboration as ameans of complexity reduction, and virtual colla-boration extends face-to-face collaborative rela-tionships by substituting technology for colloca-tion. When virtual collaborative relationships fail,technology is often blamed while in many cases itis 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 asingle construct. All four hypotheses were sup-ported. The impact on remote site health caredelivery of VCRs was monotonically associatedwith each of the individual types of trust andintegrated trust. However, it was the fourth hypo-thesis�an association between integrated trustand VCR performance�that provided the mostinteresting findings in terms of the roles the dif-ferent types of trust played and how they wereinterrelated.

The ability element can be interpreted that if oneparty�s assessment of the other party�s compe-tency is negative, then the VCR will have noimpact on remote site health care delivery. Ineach of the three cases where ability was nega-tive, the VCRs had no impact on remote sitehealth care delivery. Therefore, a nonnegativeassessment of the other party�s competence is anecessary but not sufficient condition in order fora VCR to have a positive impact on remote sitehealth care delivery.



Table 10. POSAC Coordinates and Coefficients for Integrated Trust (H4)TelemedicineRelationship Agency

Ability, Empathy,Self-interest X Y Joint Lateral Impact

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

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


Table 11. H4: Integrated Trust Weak Monotonicity Coefficient MatrixX Y JOINT LATERAL

IMPACT 0.5116 0.6697 0.9770 -0.1088

ABILITY -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



0.0 0.2 0.4 0.6 0.8 1.0Agency

0.0

0.2

0.4

0.6

0.8

1.0

Abi

lity

R [742] 11

R [746] 12R [747] 21

R [772] 12

R [753] 12

R [765] 16

R [775] 21

R [777] 19, 21, 21

HSC [777]21, 21, 21

HSC [267] 12

HSC [447] 16HSC [747] 19

HSC[217] 12

HSC[146] 11

HSC [771] 21

HSC [441] 12

Positive Ability

Negative Ability

Neutral Ability

Posi

tive

Self

Posi

tive

Self --

inte

rest

inte

rest

Neg

ativ

e Se

lfN

egat

ive

Self --

inte

rest

inte

rest

Negative Empathy

Neutral Empathy

Positive Empathy

0.0 0.2 0.4 0.6 0.8 1.0Agency

0.0

0.2

0.4

0.6

0.8

1.0

Abi

lity

R [742] 11

R [746] 12R [747] 21

R [772] 12

R [753] 12

R [765] 16

R [775] 21

R [777] 19, 21, 21

HSC [777]21, 21, 21

HSC [267] 12

HSC [447] 16HSC [747] 19

HSC[217] 12

HSC[146] 11

HSC [771] 21

HSC [441] 12

Positive Ability

Negative Ability

Neutral Ability

Posi

tive

Self

Posi

tive

Self --

inte

rest

inte

rest

Neg

ativ

e Se

lfN

egat

ive

Self --

inte

rest

inte

rest

Negative Empathy

Neutral Empathy

Positive Empathy

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

Empathy played an attenuating role. Without apositive ability assessment, positive empathy isnot likely. It appears that if ability and self-interestvalues are split�that is, one type of trust ispositive and the other is negative�then, consis-tent with its attenuating role, empathy doesn�tmatter. There are not enough data points to drawa definitive conclusion, but we speculate that ifempathy is neutral, it acts as a drag on the impactof those collaborative relationships.

Self-interest played an accentuating role. Itappears that self-interest cannot offset negativeassessments of the other party�s ability or em-pathy, but it can be the deciding factor whenability and empathy are neutral or positive.

The integrated trust model provided interestinginsights into the interrelationships of the different

types of trust on VCR performance. There was astrong relationship between all three types of trustbeing positive and VCR performance beingstrongly positive as well. In all eight cases whereone party�s assessment of the three types of trustwas all positive, the virtual collaborative relation-ship had a strongly positive impact on remote sitehealth care delivery. This relationship generallyheld when all the types of trust were either neutralor positive. In 11 out of the 12 cases where all thetypes of trust were either neutral or positive, thenimpact was positive as well.

There was also a strong relationship between trustand performance when at least one type of trustwas negative. In seven out of the eight caseswhere at least one type of trust was negative,there was no impact on remote site health caredelivery.



In the other cases where the three types of trustwere neutral or positive, it appeared that calcu-lative trust generally dominated, although in somecases, relational trust could overcome neutralcalculative trust values. In five of the six caseswhere one party�s assessment of the other party�sability was either neutral or positive, and theassessment of the other party�s empathy wasneutral, self-interest dominated. In two of thesesix cases where self-interest was neutral, therewas no impact on remote site health care delivery.In three of these six cases where ability was eitherneutral or positive, empathy was neutral, and self-interest was positive, then impact was positive aswell. In these six cases, there was only one casewhere the assessment of ability was positive,empathy was neutral, self-interest was positive,and there was no impact on remote site healthcare delivery. It seems that in this case, while itwas in one party�s self-interest to participate in thistelemedicine relationship, it was not strongenough to warrant participation.

In two of the three cases where ability andempathy were positive but self-interest wasneutral, the virtual collaborative relationship hadno impact on remote site health care delivery.This meant that in the one case where impact waspositive, a positive empathy value dominated aneutral self-interest value. Therefore, one partychose to participate in a telemedicine relationshipdespite it not being in their own interest becausethey identified with the other party.

Contributions to Researchand Practice

The results of this article extended prior studies byfinding that interpersonal trust is a primary deter-minant as to whether VCRs can address complexsituations. The results presented here confirm,extend, and apparently contradict prior studies ofinterpersonal trust.

Our finding of an association between perfor-mance and relational trust was consistent with andconfirmed McAllister�s (1995) finding of only partial

support for a relationship between affect-basedinterpersonal trust and performance. We alsofound support for an association between calcu-lative trust and performance.

Our research extended prior research byexamining the roles the different types of trustplay, and provided strong support for the hypo-thesis that there is a positive association betweenintegrated interpersonal trust and VCR perfor-mance. A neutral or positive assessment of theother party�s competence is a necessary but notsufficient condition if a VCR�s performance is to bepositive. Calculative trust plays an accentuatingrole, implying that it tends to sharpen the dif-ferentiation delineated by competence trust. Incontrast, relational trust plays an attenuating rolewhere it tends to temper the differentiationdelineated by the other types of trust.

Our research also extended prior research byexamining how the different types of trust interactand are interrelated. Our findings indicate thatpositive assessments of all three types of trust arenecessary if VCRs are to have a strongly positiveperformance. However, positive performance isstill possible if all three types of trust are non-negative; that is, if at least one type of trust ispositive and the others are neutral, then VCRperformance is likely to be positive as well, but notas positive as if each of the three types of trust areall positive. Our findings also establish that if anyone type of trust is negative, then it is very likelythat VCR performance will not be positive. Wealso found that the different types of trust areinterrelated. Self-interest dominates when abilityis either neutral or positive, and empathy isneutral�as long as one party perceives that it isstrongly enough in their own interest to participate.Empathy may dominate self-interest in caseswhere ability is positive but self-interest is neutral.

Our findings indicate that it is not one particulartype of interpersonal trust that is critical forsuccessful VCRs; rather, it is the integrated typesof interpersonal trust that matter. The differenttypes of interpersonal trust interact and are inter-dependent, and these interrelationships andinterdependencies must be taken into account by



researchers as they attempt to understand theimpact of trust on VCR performance. It alsomeans that managers of virtual collaborativeprocesses in both the project design phase andthe management of the project must not dependon a single type of interpersonal trust to enhancetheir adoption; instead, they should pay attentionto all of the types of interpersonal trust examinedin this research and realize that they are relatedand interdependent.

Our findings also apparently contradict priorresearch on integrated interpersonal trust andperformance. Unlike Zaheer et al. (1998), wefound a positive association between integratedinterpersonal trust and performance. We believethere is a methodological explanation for ourcontradictory findings. From a data analysisperspective, Zaheer et al. used structured equa-tion modeling in their study, while we used facettheory. The data assumptions in facet theory areless restrictive than those of structured equationmodeling in that the ordinal data are sufficient andthe relationships between variables need not belinear (Borg and Shye 1995; Shenkar et al. 1995;Shye et al. 1994), and are consistent with manyaspects of human behavior (Guttman 1944; Levyand Guttman 1985). Consistent with prior applica-tions of facet theory, the result here has been tofind relationships between variables that other,more restrictive methods have been unable to find(Canter 1985b; Shye et al. 1994). Thus, our con-tradictory findings may well be the result of thedifferent analysis techniques utilized.

This study also makes a methodological contri-bution by introducing facet theory to informationsystems research. The study demonstrated howfacet theory can be used in the study of complexsocial systems (both in the research design anddata analysis phases), and how it has the poten-tial to address many of the concerns and chal-lenges information systems researchers face inperforming field research.

This article also makes significant contributions topractice. For those organizations consideringdeploying VCRs, this research demonstratesinterpersonal trust factors must be addressed if

the benefits of these newer work relationships areto be reaped. Further, it is the integrated inter-personal trust factors that must be taken intoaccount. This research also benefits those healthcare delivery organizations utilizing or consideringutilizing VCRs as a means of more effectively andefficiently deploying health care resources byidentifying the social context in which such tele-medicine projects exist as a major contributingfactor in determining project impact. In doing so,it suggests new considerations in the ongoingconcerns about the failure of more extensiveutilization and adoption of telemedicine in general.This article has implications for information tech-nology professionals as well. Information techno-logy professionals need to be less obsessed withthe technology and the optimal configuration tosupport virtual collaboration activities, and insteadfocus more on the social context in which thetechnologies exist. In doing so, information tech-nology professionals must also understand that itis integrated interpersonal trust, and not any onespecific type of interpersonal trust, that matters.

Limitations and Suggestionsfor Future Research

This article makes important contributions to theunderstanding of interpersonal trust and its asso-ciation with VCR performance. However, trustresearch is sensitive to the context in which itoccurs (Rousseau et al. 1998), and this researchis subject to the same concerns about generaliza-bility as is all trust research. The use of facettheory design techniques facilitates replicatingthese findings in other contexts.

It may seem as though different types of tele-medicine activities require different degrees oftrust, and therefore the association between inter-personal trust and virtual collaborative relationshipperformance may be moderated by the type ofactivities in which the VCRs engaged. The dataindicate otherwise�that the type of telemedicineactivities in which the different relationships wereengaged did not impact the results. Additionally,



of the telemedicine relationships, 60 percent wereperceived as having a positive impact, while 40percent were not; of the teleconsultations, 60percent were perceived as having a positive im-pact while 40 percent were not; and of the dis-tance learning projects, 67 percent were per-ceived as having a positive impact, while 33 per-cent were not. Finally, of the teleradiology rela-tionships, 50 percent were found to have apositive impact, while 50 percent were not.

The findings that 50 percent of the teleradiologyrelationships were not perceived as having apositive impact was particularly supportive of theproposition that the type of telemedicine activitiesin which the relationship was engaged does notaffect whether the relationship positively impactedremote site health care delivery. Teleradiology isnot very sophisticated from a technology perspec-tive because it involves asynchronous file transfer.Further, digital radiography is well accepted in theradiology profession, and teleradiology is one ofthe few telemedicine activities for which the spe-cialists were reimbursed for participating. Finally,teleradiology involves the fewest process changesfrom the perspective of the radiologist, the primarycare provider, and the patient. These facts sug-gest that if there were a telemedicine activity forwhich interpersonal trust would matter the least, itwould be teleradiology. Yet one-half of the tele-radiology projects did not have a positive impactbecause of interpersonal trust values.

Trust is a messy concept and there are strongtheoretical and data analytical reasons why inte-grated trust should be included in the model.Numerous other researchers have theorized aboutthe concept of integrated trust. From a dataanalysis perspective, a strength of facet theory isthat it can assess the interaction effect of differentvariables by determining whether a facet orelement plays an attenuating or accentuating role.However, the integrated perspective of inter-personal trust is called integrated trust, but it maywell be a perspective of how these different typesof trust combine. Future research needs to deter-mine whether integrative trust is a new kind oftrust or just a mixture of the other types of trust.

Telemedicine is not a new type of medicine;rather, it is a different way of providing existingmedical care. Future research might investigatethe impact telemedicine and trust relationshipshave on conventional health care delivery. Futureresearch needs to address how the different typesof interpersonal trust interact and the temporalrelationships between these types of trust. Ourfindings are consistent with others� propositionthat calculative trust is needed if relational trust isto develop (Child 1998; Lewicki and Bunker 1996;Rousseau et al. 1998) and a collaborative rela-tionship is to have a positive impact, but thetimeline of when one type of trust may or may notbe more important was not studied. Futureresearch is needed to better understand theseinteractions and the temporal factors involved.

References

Allen, A., and Stein, S. �Cost Effectiveness ofTelemedicine,� Telemedicine Today (6:5),1998, pp. 10-15.

Anderson, E., and Weitz, B. �Determinants ofContinuity in Converntional Industrial ChannelDyads,� Marketing Science (8:4), 1989, pp.310-324.

Bachmann, R. �Conclusion: Trust�ConceptualAspects of a Complex Phenomenon,� in TrustWithin and Between Organizations: Concep-tual Issues and Empirical Applications, C. Laneand R. Bachmann (Eds.), Oxford UniversityPress, Oxford , 1998, pp. 298-322.

Bigley, G. A., and Pearce, J. L. �Straining forShared Meaning in Organizational Science:Problems of Trust and Distrust,� Academy ofManagement Review (23:3), 1998, pp. 405-421.

Borg, I., and Groenen, P. Modern Multidimen-sional Scaling: Theory and Applications,Springer-Verlag, New York, 1997.

Borg, I., and Lingoes, J. C. MultidimensionalSimilarity Structure Analysis, Springer, NewYork, 1987.

Borg, I., and Shye, S. Facet Theory: Form andContent, Sage, Thousand Oaks, CA, 1995.

Brown, J. �An Introduction to the Uses of FacetTheory,� in Facet Theory: Approaches to



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 Methodsin Psychology, G. M. Breakwell, S. Hammond,and C. Fife-Schaw (Eds.), Sage, ThousandOaks, CA:, 2000, pp. 105-118.

Butler, J. K. �Toward Understanding and Mea-suring Conditions of Trust: Evolution of a Con-ditions of Trust Inventory,� Journal of Manage-ment (17:3), 1991, pp. 643-663.

Butler, J. K., and Cantrell, R. S. �A BehavioralDecision Theory Approach to Modeling DyadicTrust in Superiors and Subordinates,� Psycho-logical-Reports (55:1), 1984, pp. 19-28.

Canter, D. Facet Theory: Approaches to SocialResearch, Springer-Verlag, New York, 1985a.

Canter, D. �How to Be a Facet Researcher,� inFacet 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 ofFactors,� Multivariate Behavioral Research(1:2), 1966, pp. 140-161.

Census Bureau. �1990 Census of Population andHousing, Summary Tape File 3a,� UnitedStates Census Bureau, 1990 (available onineat http://venus.census.gov/cdrom/lookup;accessed March 3, 1997).

Child, J. �Trust and International Strategic Alli-ances: The Case of Sino-Foreign Joint Ven-tures,� in Trust Within and Between Organiza-tions: Conceptual Issues and Empirical Appli-cations, C. Lane and R. Bachmann (Eds.),Oxford University Press, Oxford, 1998, pp. 241-272.

COC �Highway To Health: Transforming USHealth Care in the Information Age,� Council onCompetitiveness, Washington, DC, 1996.

Collins, T., and Bicknell, D. CRASH: Learningfrom the World�s Worst Computer Disasters,Simon and Schuster, London, 1998.

Debakey, M. E. �Telemedicine Has Now Come ofAge,� Telemedicine Journal (1:1), 1995, pp. 3-4.

Deutsch, M. �The Effect of Motivational Orienta-tion upon Trust and Suspicion,� Human Rela-tions (13), 1960, pp. 123-140.

Deutsch, M. �Trust and Suspicion,� Conflict Reso-lution (2:4), 1958, pp. 265-279.

DOC. �Telemedicine Report to Congress,� UnitedStates 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 andConstruct Validity,� Educational and Psycho-logical Measurement (53:2), 1993, pp. 351-369.

Eisenhardt, K. �Building Theories from CaseStudy Research,� Academy of ManagementReview (14), 1989, pp. 532-550.

Flaherty, R. D., Perednia, D. A., Singer, T., andOrbuch, P. The Western Governors� Asso-ciation Telemedicine Action Report, WesternGovernors� 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 Neededto Guide Investments, United States GeneralAccounting Office, United States Senate,Washington, DC, 1997.

Gefen, D. �E-Commerce: The Role of Familiarityand Trust,� Omega (28), 2000, pp. 725-737.

George, J. M., and Jones, G. R. �The Experienceand Evolution of Trust: Implications for Coop-eration and Teamwork,� Academy of Manage-ment Review (23:3), 1998, pp. 531-546.

Ghoshal, S., and Moran, P. �Bad for Practice: ACritique of the Transaction Cost Theory,�Academy of Management Review (21:1), 1996,pp. 35-48.

Grigsby, J. �Current Status of Domestic Tele-medicine,� Journal of Medical Systems (19:1),1995, pp. 19-27.

Guttman, L. �A Basis for Scaling QualitativeData,� American Sociological Review (9), 1944,pp. 139-150.

Guttman, L. �Facet Theory, Smallest SpaceAnalysis, and Factor Analysis,� Perceptual andMotor Skills (54), 1982a, pp. 491-493.

Guttman, L. �A General Nonmetric Technique forFinding the Smallest Coordinate Space for aConfiguration of Points,� Psychometrika (33:4),1968, pp. 469-506.

Guttman, L. �A Structural Theory for IntergroupBeliefs and Actions,� American SociologicalReview (24), 1959, pp. 318-328.



Guttman, L. ��What Is Not What� in TheoryConstruction,� 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. �FacetTheory: Its Development and Current Status,�European Psychologist (3:1), 1998, pp. 13-36.

Hagen, J. M., and Choe, S. �Trust in JapaneseInterfirm Relations: Institutional Sanctions Mat-ter,� Academy of Management Review (23:3),1998, pp. 601-620.

Handy, C. �Trust and the Virtual Organization,�Harvard Business Review (73), 1995, pp. 40-50.

IOM. Access To Health Care In America, Instituteof Medicine, National Academy Press, Wash-ington, DC, 1993.

IOM. Clinical Practice Guidelines: Directions fora New Program, Institute of Medicine, NationalAcademy Press, Washington, DC, 1990.

IOM. Primary Care: America�s Health in an NewEra, Institute of Medicine, National AcademyPress, Washington, DC, 1996a.

IOM. Telemedicine: A Guide to Assessing Tele-communications in Health Care, Institute ofMedicine, National Academy Press, Wash-ington, DC, 1996b.

Jarvenpaa, S. L., Knoll, K., and Leidner, D. E. �IsAnybody Out There? Antecedents of Trust inGlobal Virtual Teams,� Journal of ManagementInformation Systems (14:4), 1998, pp. 29-64.

Jarvenpaa, S. L., and Leidner, D. E. �Com-munication and Trust in Global Virtual Teams,�Organizational Science (10:6), 1999, pp. 791-815.

Kelle, U. Computer-Aided Qualitative Data Analy-sis: Theory Methods and Practice, Sage Publi-cations, London, 1995.

Kramer, R. M. �Trust and Distrust in Organi-zations: Emerging Perspectives, EnduringQuestions,� Annual Review of Psychology (50),1999, pp. 569-598.

Lane, C. �Introduction: Theories and Issues inthe Study of Trust,� in Trust Within and Be-tween Organizations: Conceptual Issues andEmpirical Applications, C. Lane and R. Bach-mann (Eds.), Oxford University Press, Oxford,1998, pp. 1-30.

Latane, B., Liu, J. H., Nowak, A., Bonevento, M.,and Zheng, L. �Distance Matters: PhysicalSpace and Social Impact,� Personality andSocial Psychology Bulletin (21:8), 1995, pp.795-895.

Levy, S. Louis Guttman on Theory and Methodo-logy: Selected Writings, Dartmouth Publishing,Brookfield, VT, 1994.

Levy, S., and Guttman, L. �The Partial-Order ofSeverity of Thyroid Cancer with the Prognosisof Survival,� in Analysis in Real Life Environ-ment: 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 andMaintaining Trust in Work Relationships,� inTrust in Organizations: Frontiers of Theory andResearch, R. Kramer and T. Tyler (Eds.), SagePublications, Thousand Oaks, CA, 1996, pp.114-139.

Lewicki, R. J., McAllister, D. J., and Bies, R. J.�Trust and Distrust: New Relationships andRealities,� Academy of Management Review(23:3), 1998, pp. 438-458.

Lewis, J. D., and Weigert, A. �Trust as a SocialReality,� Social Forces (63:4), 1985, pp. 967-985.

Luhmann, N. Trust and Power, Wiley,Chichester, UK, 1979.

Luhmann, N. �Familiarity, Confidence, Trust:Problems, and Alternatives,� in Trust: Makingand Breaking Cooperative Relations, D.Gambetta (Ed.) Basil Blackwell, New York,1988, pp. 94-108.

March, J., and Olsen, J. Rediscovering Institu-tions: The Organizational Basis for Politics,Free Press, New York, 1989.

Martin, P. Y., and Turner, B. A. �Grounded Theoryand Organizational Research,� Journal ofApplied Behavioral Science (22:2), 1986, pp.141-157.

Mason, R. O., and Mitroff, I. I. �A Program forResearch on Management Information Sys-tems,� 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.



McAllister, D. J. �Affect - and Cognition-BasedTrust as Foundations for Interpersonl Coopera-tion in Organizations,� Academy of Manage-ment Journal (38:1), 1995, pp. 24-59.

McKnight, D. H., Cummings, L. L., and Chervany,N. L. �Initial Trust Formation in New Organiza-tional Relationships,� Academy of ManagementReview (23:3), 1998, pp. 473-490.

Meyerson, D., Weick, K. E., and Kramer, R. M.�Swift Trust and Temporary Groups,� in Trust inOrganizations: Frontiers of Theory andResearch, R. Kramer and T. Tyler (Eds.), SagePublications, Thousand Oaks, CA, 1996, pp.166-195.

Mishra, A. K. �Organizational Responses toCrisis: The Centrality of Trust,� in Trust inOrganizations: Frontiers of Theory andResearch, R. Kramer and T. Tyler (Eds.), SagePublications, Thousand Oaks, CA, 1996, pp.261-287.

Morris, R. �Computerized Content Analysis inManagement Research: A Demonstration ofAdvantages and Limitations,� Journal ofManagement (20), 1994, pp. 903-932.

Newell, S., and Swan, J. �Trust and Inter-Organizational Networking,� Human Relations(53:10), 2000, pp. 1287-1328.

Nohria, N., and Eccles, R. G. �Face-to-Face:Making Network Organizations Work,� in Net-works and Organizations: Structure, Form, andAction, N. Nohria and R. G. Eccles (Eds.),Harvard Business School Press, Boston, 1992,pp. 288-308.

OHRA. Exploratory Evaluation of Rural Applica-tions of Telemedicine, Office of Rural HealthPolicy, United States Department of Health andHuman Services, Rockville, MD, 1997.

OTA. Bringing Health Care Online: The Role ofInformation Technologies, Office of TechnologyAssessment, United States GovernmentPrinting Office, Washington, DC, 1995.

Orlikowski, W. J. �CASE Tools as OrganizationalChange: Investigating Incremental and RadicalChanges in Systems Development,� MISQuarterly (17:3), 1993, pp. 309-340.

Paul, D. L., Pearlson, K. E., and McDaniel, R. R.�Technological Barriers to Telemedicine: Tech-nology Management Implications,� IEEETransactions on Engineering Management(6:2), 1999, pp. 279-288.

Ring, P. S. �Fragile and Resilient Trust and TheirRoles in Economic Exchange,� Business &Society (35:2), 1996,

Ring, P. S., and Van de Ven, A. H. �Develop-mental Processes of Cooperative Interorgani-zational Relationships,� Academy of Manage-ment Review (19:1), 1994, pp. 90-118.

Rousseau, D. M., Sitkin, S. B., Burt, R. S., andCamerer, C. �Not So Different After All: ACross-Discipline View of Trust,� Academy ofManagement Review:3), 1998, pp. 393-404.

Sackmann, S. A. Cultural Knowledge in Organi-zations: Exploring the Collective Mind, SagePublications, Newbury Park, CA, 1991.

Sako, M. �Does Trust Improve Business Per-formance?,� in Trust Within and BetweenOrganizations: Conceptual Issues and Empiri-cal Applications, C. Lane and R. Bachmann(Eds.), Oxford University Press, Oxford, 1998,pp. 88-117.

Sako, M. Prices, Quality, and Trust: Inter-firmRelations in Britain and Japan, Cambridge Uni-versity Press, Cambridge, UK, 1992.

Sako, M. �The Role of �Trust� in Japanese Buyer-Supplier Relationships,� Ricerche Economiche(45), 1991, pp. 449-474.

Shapira, Z. �A Facet Analysis of LeadershipStyles,� Journal of Applied Psychology (61),1976, pp. 136-139.

Shenkar, O., Aranya, N., and Almor, T. �ConstructDimensions in the Contingency Model: AnAnalysis Comparing Metric and Non-MetricMultivariate Instruments,� Human Relations(48), 1995, pp. 559-580.

Shepard, B., and Tuchinsky, M. �Micro-OB andthe Network Organization,� in Trust in Organi-zations: 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 UnknownDistance Function. Part I,� Psychometrika(27:2), 1962a, pp. 125-140.

Shepard, R. N. �The Analysis of Proximities:Multidimensional Scaling with an UnknownDistance Function. Part II,� Psychometrika(27:3), 1962b, pp. 219-246.

Shye, S. �Modern Facet Theory: Content Designand Measurement in Behavioral Research,�



European Journal of Psychological Assessment(14:2), 1998, pp. 160-171.Shye, S. Multiple Scaling: The Theory and Appli-

cation of Partial Order Scalogram Analysis,North-Holland, Amsterdam, 1985a.

Shye, S. �Nonmetric Multivariate Models forBehavioral 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 ScalogramAnalysis by Base Coordinates and LatticeMapping of the Items by Their ScalogramScales,� in Facet Theory: Approaches to SocialResearch, D. Canter (Ed.) Springer Verlag,New York, 1985, pp. 277-298.

Shye, S., Elizur, D., and Hoffman, M. Introductionto Facet Theory: Content Design and IntrinsicData Analysis in Behavioral Research, SagePublications, Thousand Oaks, CA, 1994.

Silberman, C. E. �What Is it like to Be a Patient inthe 1990s?,� in Technology and Health Care inan Era of Limits, Institute of Medicine, NationalAcademy Press, Washington, DC, 1992, pp.165-197.

Staw, B. M., Sandelands, L. E., and Dutton, J. E.�Threat-Rigidity Effects in Organizational Be-havior: A Multilevel Analysis,� AdministrativeScience Quarterly (26:4), 1981, pp. 501-524.

Strauss, A., and Corbin, J. Basics of QualitativeResearch: Grounded Theory Procedures andTechniques (2nd ed.), Sage Publications,Newbury Park, CA, 1998.

Tyler, T., and Kramer, R. M. �Whither Trust?,� inTrust in Organizations: Frontiers of Theory andResearch, R. Kramer and T. Tyler (Eds.), SagePublications, Thousand Oaks, CA, 1996, pp. 1-15.

Whitener, E. M., Brodt, S. E., Korsgaard, M. A.,and Werner, J. M. �Managers as Initiators ofTrust: An Exchange Relationship Frameworkfor Understanding Managerial TrustworthyBehavior,� Academy of Management Review(23:3), 1998, pp. 513-531.

Williamson, O. E. �Calculativeness, Trust, andEconomic Organization,� Journal of Law andEconomics (30), 1993, pp. 131-145.

Williamson, O. E. The Economic Institutions ofCapitalism, The Free Press, New York, 1985.

Williamson, O. E. �The Economics of Organiza-tion: The Transaction Cost Approach,� Ameri-can Journal of Sociology (87), 1981, pp. 548-577.

Wolfe, R., Gephart, R., and Johnson, T. �Com-puter Facilitated Qualitative Data Analysis:Potential Contributions to Management Re-search,� Journal of Management (19), 1993,pp. 637-660.

Yin, R. K. Case Study Research: Design andMethods (2nd ed.), Sage Publications, Thou-sand Oaks, CA, 1994.

Zaheer, A., McEvily, B., and Perrone, V. �DoesTrust Matter? Exploring the Effects of Inter-organizational and Interpersonal Trust onPerformance,� Organization Science (9:2),1998, pp. 141-159.

Zand, D. E. �Trust and Managerial ProblemSolving,� Administrative Science Quarterly (17),1972, pp. 229-239.

Zucker, L. G., Darby, M. R., Brewer, M. B., andPeng, Y. �Collaboration Structure and Infor-mation Dilemmas in Biotechnology: Organi-zational Boundaries as Trust Production,� inTrust in Organizations: Frontiers of Theory andResearch, R. Kramer and T. Tyler (Eds.), SagePublications, Thousand Oaks, CA, 1996, pp.90-113.

About the Authors

David L. Paul is an assistant professor in theDepartment of Information Technology and Elec-tronic Commerce at the Daniels College of Busi-ness of the University of Denver. He received aB.S. from the Wharton School of the University ofPennsylvania, an M.B.A. from the AndersonSchool at the University of California, LosAngeles, and a Ph.D. from the Graduate School ofBusiness at the University of Texas at Austin. DrPaul�s research interests include telemedicine,virtual collaboration, trust, and complex adaptivesystems. Dr. Paul has published articles in jour-nals such as IEEE Transactions on EngineeringManagement, International Journal of HealthcareTechnology and Management, and Computationaland Mathematical Organizational Theory, and he



has presented papers at the Hawaii InternationalConference on System Sciences and numerousINFORMS Conferences. He was the recipient ofthe 2000 International Conference on InformationSystems Best Dissertation Award.

Reuben R. McDaniel, Jr. holds the Charles andElizabeth Prothro Regents Chair in Health CareManagement in the McCombs School of Businessat The University of Texas at Austin and is aprofessor of Management Science and Informa-tion Systems. He received his Ed.D. from IndianaUniversity, his M.S. from Akron University, and his

B.S. from Drexel University. His present researchinterest is in management of complex adaptivesystems, with a particular interest in organizationaldesigns and information systems for more effec-tive sensemaking and decision-making in healthcare organizations. A partial list of journals he haspublished in includes The Academy of Manage-ment Journal, Decision Sciences, Health CareManagement Review, Health Services Research,The Journal of Applied Behavioral Sciences, TheJournal of the National Medical Association, Man-agement Science, and Organizational Behaviorand 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 addressthe concerns and challenges information systems researchers face in performing field research and whyit was an appropriate methodology for this study. Facet theory is a systematic approach to facilitating andintegrating research construction, design, and data analysis of complex social systems. It utilizes multi-dimensional data analysis guided by a theoretical framework (Borg and Shye 1995; Guttman and Green-baum 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 ofobservations and an aspect of the empirical structure of those observations, together with a rationale forsuch a hypothesis.�

Facet theory provides a system of concepts, definitions, and theorems on research design and dataanalysis (Borg and Shye 1995). The advantages of its research design concepts are that they providetechniques for defining the observations and constructing hypotheses that link features of the design withempirical 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 toolin the design phase. It is a verbal statement used to construct the formal definitional framework forresearch design and theory testing. In the mapping sentence, the theoretical constructs of the researchand the type of observations needed to test it are simultaneously identified and explicated, and it is fromthe mapping sentence that hypotheses are generated (Borg and Shye 1995; Brown 1985; Canter 1985b;Guttman and Greenbaum 1998).



Table A1. Facet Theory TerminologyTerm Definition

Accentuating A facet or element in an accentuating role partitions the POSAC item diagram withan inverted �L� shaped line and the facet or element�s weak monotonicity coefficientvalues should approach zero for the X axis and one for the Y axis. A facet or ele-ment in an accentuating role is structurally dependent on the base-coordinate (X orY) items and induces a finer division of the X and Y base coordinates by empha-sizing or sharpening whatever trend is indicated by the base coordinates. Thestructural dependence is such that values of an accentuating item tend to increasewith 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 (Borgand Shye 1995; Shye 1985a; Shye et al. 1994).

Attenuating An facet or element in an attenuating role partitions the POSAC item diagram withan �L� shaped line and the facet or element�s weak monotonicity coefficient valuesshould approach one for the X axis and zero for the Y axis. A facet or element in anattenuating role is structurally dependent on the base-coordinate (X or Y) items andinduces a finer division of the X and Y base coordinates by moderating whatevertrend is indicated by the base coordinates. The structural dependence is such thatvalues of an attenuating item tend to increase with an increase in the concurrentsimultaneous increase in values of both of the base coordinates, and its role is tomoderate the differentiation delineated by the base coordinates (Borg and Shye1995; Shye 1985a; Shye et al. 1994).

Coefficient ofAlienation

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

ComparableProfiles

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

ContentFacets

The �Q� in PQ!R, content facets represent the domain of interest of the researchin that they are the attributes or conditions by which the population of interest iscompared or classified (Borg and Shye 1995).

ContentStructuple

A subset of a score structuple, a content structuple is a definitional structuple con-sisting of element values for some but not all of the different content facets (Shye etal. 1994).

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



Table A1. Facet Theory Terminology (Continued)Term Definition

Elements Elements define the universe of a facet by specifying the exact values subjects maybe assigned within a particular facet. Elements within a facet are mutually exclusiveand jointly exhaustive in that the elements in a particular facet ideally make up all ofthe 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 obser-vation (Shye et al. 1994).

External Item An external item is a classification of the content facets and their elements specifiedin terms of the content facets� properties or behavior with respect to criterion externalto those content facets (Borg and Shye 1995). The external item enables thecomparison of an association between one or more of the internal content facets andan external concept.

External ItemDiagram

A variation of an item diagram, where the external item values for each profile aresuperimposed on the profiles in the POSAC diagram, which is then used to assesswhether there is a relationship between the content facets� elements and an externalitem (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 setthat together represent underlying conceptual and semantic components within acontent universe (Guttman and Greenbaum 1998).

Facet Roles Facets may play an axial, modular, polar, or joint role (Levy and Guttman 1985). Anaxial role is played by a facet whose elements are ordered but the order of theelements is uncorrelated with the order of the elements of the other facets. Axialfacets partition the SSA space into sections using horizontal or vertical lines. Amodular facet also consists of ordered elements, but it may be related to otherfacets. A modulating facet forms concentric circles radiating from a particular origin.A polar facet consists of unordered elements where the elements form a circulardistribution of points termed a circumplex geometric structure. Polar facets representa qualitative facet where no obvious beginning or end from which order among theelements exists (Levy and Guttman 1985). They partition space into wedge-shapedregions, with each region representing an element of the facet (Brown 1985). Afourth role is that of the joint role where two ordered facets act together to representone dimension of a model. Facets play a joint role when two or more of them havea common notion of order, and combinations of elements divide the SSA into con-ceptually consistent regions (Borg and Shye 1995, Shye 1998). Joint facets tend todivide the SSA space diagonally. Note that these partitions presented above areideal types; in practice, the partitioning of the SSA space will not be perfect (Shye1985a).




GeometricDataStructures

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

Guttman�smu2

Also called the weak monotonicity coefficient, it is a nonparametric measure ofassociation which makes the weakest possible assumptions about empiricalobservations in that data are assumed to have ordinal properties only so that itscalculation does not require the distribution of the variables to be known. Guttman�smu2 assesses the extent to which an increase in one variable is accompanied by anincrease (or no decrease) in the other. Like other measures of association, the valueof Guttman�s mu2 can vary between -1 to +1 (Canter 1985; Shye 1985a).

IncomparableProfiles

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

Intension The addition of new facets to an existing mapping sentence, usually as justified byempirical 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 eachcontent structuple.

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

Joint Axis (inPOSAC)

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

Lateral Axis(in POSAC)

The lateral axis obtained by rotating of the Y axis 45/, represents qualitative (dif-ferential) aspects of the observed phenomena and is utilized to order profiles thathave the same joint coordinate value and to determine the role the different contentfacets play (Shye et al. 1994).

MappingSentence

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 sen-tence serves to define a priori exactly what is being studied�the population, thecontent variables, and the range of possible responses serving as the definitionaland conceptual base of the problem to be studied (Borg and Shye 1995).




MonotonicRelationship

A monotonic relationship exists when there is a consistent upward or downward trendbetween 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 itselements (Levy and Guttman 1985).

Partial Order A partial order is a relation on a set of elements such that determining which elementrelative to another has a higher or lower order is not possible (Levy and Guttman1985).

Partial OrderScalogramAnalysis bybaseCoordinates(POSAC)

Partial order scalogram analysis by base coordinates (POSAC) is a procedure forfitting observed profiles into a two dimensional coordinate space representationsubject to the constraint that ordered relations, including incomparability, arepreserved. POSAC is used to investigate the structural characteristics of subject,enabling the comparison (or partial comparison) of the subjects (profiles) in thesample based on differences in their content facet element values. It also can beused 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).

POSACDiagram

A two dimensional partial order scalogram of an iteratively calculated configurationof points for a set of profiles based on some or all of the content facet values wherethe 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 populationof interest, Q the domain of interest, and R the response given the Cartesian productof 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 (Brownand 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 commonmeaning in such a way that, in all content facets, high numerical values indicatepresence (or absence) of the attribute (Shye et al. 1994).

RegionalHypotheses

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

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

ScoreStructuple

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




ShepardDiagram

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

SmallestSpaceAnalysis(SSA)

A method of multidimensional scaling in which a set of variables and their inter-correlations are geometrically portrayed in space while preserving the rank order ofrelations, where smallest space means the �space of lowest dimensionality� (Shyeet al. 1994). SSA aims at establishing the structural properties of variables and isoften used to test the hypotheses stated in the mapping sentence by relating theconceptual structure of observations on the universe of features of the empiricalstructure of observations on that universe (Guttman and Greenbaum 1998).

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

UnorderedFacet

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

WeakMonotonicityCoefficient

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) definesthe unit of analysis for a particular research problem. Q and R are two different types of facets, where afacet 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 comparedor classified (Borg and Shye 1995). In the mapping sentence, the items to the left of the arrow are theconditions of the observation�the subject (profile) and its attributes of interest (content facets). The itemto the right of the arrow is the range facet (R), which represents the range of possible observations usedto classify individual subjects within the population of interest (P).

Each facet consists of elements which specify the exact values subjects may be assigned within aparticular facet. Content facet elements may be either ordered or unordered, while range facet elementvalues must be ordered. Ordered facets are those facets whose elements represent quantitativedistinctions, 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 particularfacet 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 contentfacets and the range facets for a specific population, where the Cartesian product of facet sets P × Qrepresents the whole possible universe of the population asked all possible questions, and R representsthe whole possible universe of answers to those questions (Borg and Shye 1995). The mapping sentenceprovides 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



sentence, usually as justified by empirical observation�and extension�the addition of elements to existingfacets, usually as justified by empirical observation�provides the researcher with the flexibility of pursuingunexpected 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 �classificationof 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 itemenables the comparison of an association between one or more of the internal content facets and anexternal concept.

Data Analysis in Facet Theory

Facet theory data analysis methods involve two related multidimensional scaling (MDS) techniques. Thegoal is to determine whether a correspondence between the definitional system�the mappingsentence�and the observed data exists. Smallest space analysis (SSA) is used to evaluate the structureof the model, while partial order scalogram analysis by base coordinates (POSAC) is used to compare�orat least partially compare if direct comparison is not possible�differences in the sample profiles, and incases where an external item exists, how the different profiles or content facets are associated with theexternal item. Both techniques use a Euclidean space in which facet elements in SSA and profiles inPOSAC are presented as points to examine if they partition the space into contiguous regions. Bothtechniques are a form of discriminant analysis where regional hypotheses�how the data points partitiona geometric region into different spaces based on the different elements or profiles rather than theclustering of points within that space�are used to evaluate the structure of the model and to compare thedifferent sample profiles (Levy and Guttman 1985).

Data assumptions in facet theory require that the data need only be ordinal (Shenkar et al. 1995; Shye etal. 1994), and relationships between variables need not be linear (Borg and Shye 1995). Such assump-tions are appropriate in behavioral and social research because they are consistent with many aspects ofhuman behavior (Guttman 1944; Levy and Guttman 1985). Thus, while parametric measures ofassociation may be used, nonparametric measures of association are often used. An advantage of usingnonparametric measures of association in MDS is that with as few as seven data points the data analysisresults 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 observationsin that data are assumed to have ordinal properties only (Canter 1985b). The weak monotonicity coeffi-cient assesses the extent to which an increase in one variable is accompanied by an increase (or nodecrease) in the other. Like other measures of association, the value of the weak monotonicity coefficientcan vary between -1 to +1. However, its calculation does not require the distribution of the variables to beknown; as such, the order of interpoint distances is sufficient for determining a unique configuration ofpoints 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 andconstruct validity of the model. SSA comprises a class of MDS models that represent similarity (ordissimilarity) coefficients among a set of objects by distances in a multidimensional space (Borg andLingoes 1987). SSA of the content facets for different dimensional models is performed to determine theoptimal dimensionality of the model.



As in factor analysis, specific rules for determining the structure of a model and its appropriate dimen-sionality 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 varianceexplained 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 whichthe distance between pairs of points and a two dimensional space adhere to the rule regarding themonotonic relationship between input coefficients and output distances. A monotonic relationship existswhen 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 bya 0 value and the worst fit is given by the value 1 (Guttman 1968). As a general rule, a coefficient ofalienation 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 themodel. The Shepard diagram (Shepard 1962a, 1962b) is a scatterplot of distances between points in theMDS plot against the observed dissimilarities or similarities. A model�s goodness-of-fit is indicated by arelatively smooth negatively sloped monotone series of points for similarities or a positively slopedmonotone series of points for dissimilarities.

An approach similar to the scree test (Cattell 1966) in factor analysis is used to determine the appropriatedimensionality of a model in facet theory. Changes in the coefficient of alienation and in the proportion ofvariance explained when the model is tested in different dimensions are examined, and the model�sappropriate dimensionality is where adding another dimension would account for only small changes inthe 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 appropriatedimensionality. Construct validity is assessed through tests of regional hypotheses. SSA is a techniquefor 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, theresearcher 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 whichthe hypothesized model�represented by the mapping sentence�and the partitioning of the SSA space�ageometric 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 pointsto discriminate or divide the space into the different elements rather than the clustering of points within thatspace 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 facetand 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 facetsplayed in the conceptualized mapping sentence.

Facets may play an axial, modular, polar, or joint role (Levy and Guttman 1985). An axial facet is onewhose elements are ordered but the order of the elements is uncorrelated with the order of the elementsof the other facets. Axial facets partition the SSA space into sections using horizontal or vertical lines. Amodular 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 elementswhere the elements form a circular distribution of points termed a circumplex geometric structure. Polarfacets represent a qualitative facet where there is no obvious beginning or end from which order amongthe elements may be determined (Levy and Guttman 1985). They partition space into wedge-shapedregions, with each region representing an element of the facet (Brown 1985). A fourth role occurs whentwo ordered facets act together to represent one dimension of a model. Facets play a joint role when twoor more of them have a common notion of order, and combinations of elements divide the SSA into con-ceptually consistent regions (Borg and Shye 1995, Shye 1998). Joint facets tend to divide the SSA spacediagonally.

The facets, in their different roles, can be combined to form different structures of the data. Three of themost common structures are the duplex, radex, and cylindrex. A duplex structure is formed by combiningtwo axial facets, where one of the facets partitions the two-dimensional SSA space horizontally while theother 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, andhas a common origin for both the polar and modular facets. The cylindrex divides a three-dimensional SSAspace 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 willnot be perfect (Shye 1985a). The manner in which the SSA space is partitioned is guided by the need toreplicate the findings across different samples. Regional hypotheses are more likely to be replicated if theoriginal SSA space is partitioned by regular, simple lines. Therefore, while it is possible to partitionparticular samples with irregular lines or shapes, this is not desirable because it is less likely such irregularpartitions will be replicated. The more irregular lines or shapes utilized to partition a particular space, themore likely the regional hypothesis will not be empirically supported over time (Borg and Shye 1995, Shye1998).

Comparisons of Subjects and Their Attributes

Partial order scalogram analysis by base coordinates (POSAC) is used to compare�or partially com-pare�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�representedby different combinations of their content facets�and the external item. The subjects (p) are representedby their score structuples, which consist of element values for all the different content facets. The actualvalue 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 oftheir score structuples. The different subjects can be ordered or partially ordered by comparing score orcontent structuple values. A partial order is a relation on a set of elements such that determining whichelement relative to another has a higher or lower order is not possible (Borg and Lingoes 1987). Struc-tuples are comparable if all the structuple values of one structuple are equal to or greater than the corre-sponding structuple values of the other structuple, and is of a higher order than another if and only if it ishigher on at least one item and not lower on any other items. However, two structuples are not comparableif and only if one structuple is higher on at least one structuple while the other structuple is also higher onat 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 collectionof observed profiles in which each profile is represented as a point in a two-dimensional space and thecomparability between two profiles is represented by a line segment. POSAC is a procedure for fittingobserved profiles into a two-dimensional coordinate space subject to the constraint that ordered relations,



including incomparability, are preserved (Shye 1985a). A configuration of points for a set of profiles basedon some or all of the content facet values is iteratively calculated, and these points, labeled by their profileID, are used to graph a two-dimensional partial order scalogram such that the partial order is preservedas best as possible (Shye and Amar 1985). A two dimensional space representation (termed a POSACdiagram) is utilized because it is the smallest dimensionality that preserves the partial order (Levy andGuttman 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 amathematically optimal case of base axis for the empirical partial order (Levy and Guttman 1985). Thesecond set of coordinates is for the joint (J) and lateral (L) axis, which are the X and Y axis rotatedclockwise 45 degrees. The joint axis has a positive slope of one relative to the X and Y axis, and is usedto compare the order of different profiles in the sample. The lateral axis has a negative slope of onerelative to the X and Y axis, and is utilized to order profiles that have the same joint coordinate value andto determine the role the different content facets play (Borg and Shye 1995; Levy and Guttman 1985; Shyeand 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 byPOSAC. CORREP can have values between 0 and 1, with 1 meaning the model order is perfectlyrepresented (Shye and Amar 1985). The order represented in the POSAC diagram is based on the contentfacets only, where the order of a profile relative to other profiles is a function of their position relative to thejoint axis. Generally speaking, those profiles located in the upper right-hand corner of the POSAC diagramare 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 profileswhen only partial order exists. Not all profiles are comparable. Two methods to assess which profiles arecomparable or not comparable, and the order of comparable profiles, are available. The first method isbased on the slope of the line connecting two profile points. Two comparable score structuples are repre-sented by two points on a common line having a positive slope, with the points closer to the upper rightcorner having a higher order. Noncomparable score structuples have their points aligned along a negativeslope; that is, they are oriented toward a direction orthogonal to the joint direction, namely the lateraldirection (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 theother 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 intersectedby the two lines, while those profiles in the lower left-hand corner have profiles of a lower order than theprofile intersected by the two lines. The profiles in the remaining two quadrants�the upper left-hand andlower right-hand quadrants�are not comparable to the profiles intersected by the two lines (Levy andGuttman 1985; Shye and Amar 1985).

One of the most useful aspects of the POSAC analysis is the item diagram, which is the reproduction ofthe POSAC diagram with an item (content facet element) score replacing the profile name. This enablesthe comparison of the different profiles based on part or all of their score structuple values. Regionalhypotheses of the item diagram(s) are used to assess the role of that particular item. The POSAC spacefor the different item diagrams are partitioned based on the role the facets elements are hypothesized toplay. Facets or elements hypothesized to play either an axial or polar role should partition the POSAC itemdiagram with lines parallel to either the X or Y axis, and the weak monotonicity coefficient for the facet orelement should approach one and negative one for the X and Y axis (or vice versa). A facet or elementhypothesized to play an attenuating role partitions the POSAC item diagram with an �L� shaped line andthe facet or element�s weak monotonicity coefficient values should approach one for the X axis and zero



for the Y axis. Such a facet or element is structurally dependent on the base-coordinate (X or Y) itemssuch that values of an attenuating item tend to increase with an increase in the concurrent simultaneousincrease in values of both of the base coordinates, and its role is to moderate the differentiation delineatedby the base coordinates (Borg and Shye 1995; Shye 1985a; Shye et al. 1994). A facet or elementhypothesized to play an accentuating role partitions the POSAC item diagram with an inverted �L� shapedline and the facet or element�s weak monotonicity coefficient values should approach zero for the X axisand 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 basecoordinates (Borg and Shye 1995; Shye 1985a; Shye et al. 1994). Note that the partitions presentedabove 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 thereforethe 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 eachprofile in the POSAC diagram, are used to assess whether there is a relationship between the contentfacets� elements and an external item (Levy and Guttman 1985; Shye and Amar 1985). This relationshipis tested empirically by superimposing the external facet content structuple values of the profiles on theirrespective points in the POSAC diagram and calculating the external item�s weak monotonicity coefficientvalue with the joint axis and lateral axis. A content facet has a positive association with an external itemif the external item�s weak monotonicity coefficient value with the joint axis approaches one and its valuewith the lateral axis approaches zero (Shye 1985a). A positive relationship between the content facets andan external item is represented by the partitioning of the external item diagram of the POSAC space by aline parallel to the lateral axis (Levy and Guttman 1985). The hypothesis is empirically supported if theprofiles with the strongest association with the external item are located in the upper right-hand corner ofthe 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 discriminateor divide the space based on the different elements and not the clustering of points within that space is thebasis by which the acceptability of the partitioning is judged (Levy and Guttman 1985). Exact rules fordetermining a POSAC model�s ability to discriminate between the different facet elements, or in the caseof 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 mappingsentence and data collection methods, and the choice of facets and sample. Replication across differentsamples ultimately is the key criteria for determining the acceptability of the partitioning of the POSACspace as a means of understanding the role of the different content facets or for establishing a relationshipbetween the content facets and the external item (Borg and Shye 1995; Shye and Amar 1985). ThePOSAC diagram can be an effective tool for predicting the outcome/value of an external item given thevalues of content facet elements once the partitioning of the POSAC space has been determinedacceptable.

Facet theory is an appropriate methodology for addressing the challenges information systems researchersface because it provides a systematic approach to developing a research design and completing dataanalysis in the study of complex social systems. The development of a mapping sentence is one key tothe success of this methodology. The mapping sentence enables the modeling of the appropriate facetsand their relationships. Multidimensional scaling provides useful strategies for evaluating the structuresof the resulting models and for evaluating the hypotheses derived from these models.

Documents

A Field Study of the Effect of Interpersonal Trust on ...€¦ · MIS Quarterly Vol. 28 No. 2, pp. 183-227/June 2004 183 RESEARCH ARTICLE A FIELD STUDY OF THE EFFECT OF INTERPERSONAL