Acknowledgments: The authors wish to thank three anonymous reviewers from the Organizational Behavior division of the 1996 meeting of the Academy of Management.
Researchers and practitioners have shown sufficient interest in trust that it has become a widely-studied and highly-heralded concept. But what does the word 'trust' mean? Since the term trust is so frequently used, the business practitioner literature tends to assume everyone knows what trust means. Therefore, the word is rarely defined in most practitioner articles. While most scholars provide a definition of trust, they are dissatisfied regarding their own lack of consensus about what trust is. Researchers have used a wide variety of trust definitions, leading to a plethora of in-use meanings of the concept. This has resulted in confusion regarding how to compare one trust research result to another.
In order to move trust research forward in a cumulative manner, the questions about the conceptual meaning of trust must be resolved. To move toward this objective, the authors propose: (a) a classification system for types of trust and (b) definitions of six related trust types. The classification system shows analytically the types of trust that have been studied and compares these types to what people mean by 'trust' in everyday language. Our six trust construct model addresses the more common and important types of trust in use and provides a manageable way to understand trust-related phenomena for both theoretical, empirical, and practical purposes. Some implications of the model are drawn for management.
Scholars and practitioners widely acknowledge trust's importance. Trust makes cooperative endeavors happen (e.g., Arrow, 1974; Deutsch, 1973; Gambetta, 1988). It is a key to positive interpersonal relationships in various settings (e.g., Fox, 1974; Lewis & Weigert, 1985a) because it is central to how we interact with others (e.g., Berscheid, 1994; Golembiewski & McConkie, 1975). Trust becomes even more central and critical during periods of uncertainty due to organizational crisis (Mishra, 1996; Weick & Roberts, 1993). In the organizational "restructuring" crisis of the 1990s, trust has emerged as a central and key strategic asset for organizations (e.g., Mayer, Davis & Schoorman, 1995; Mishra, 1996). Trust is a central component in effective working relationships (Gabarro, 1978). Practitioners acknowledge the importance of trust as much as do scholars (e.g., Bartlett & Ghoshal, 1995; Covey, 1989; Peters, 1992). For example, a book on partnering recently quoted one business person as saying, "...'there are a lot of issues in partnering,...but trust is truly the key. Everything else has to be based on it. Without trust, there is no basis for partnering. It's the bottom line.'" (Rackham, Friedman & Ruff, 1996: 75) The same authors reported, "We heard the same sentiment over and over."
Because trust is considered so vital, it has been studied extensively in many research disciplines (e.g., Golembiewski & McConkie, 1975; Kramer & Tyler, 1996). Scientific study should, over time, lead to some level of consensus on a topic (Kuhn, 1962). But while agreement is rising concerning the positive effects of trust (e.g., Kramer & Tyler, 1996), little consensus has formed on what trust means (Kee & Knox, 1970; Taylor, 1989; Yamagishi & Yamagishi, 1994). Rather, significantly diverse meanings of trust continue to be used in the interdisciplinary research literature. This pattern is primarily driven by empirical studies that typically define trust in very specific, narrow ways. Wrightsman's (1991: 411) commentary on future research directions is insightful: "...the general concept of trust deserves much more theoretical analysis. Measurement has advanced more rapidly than conceptual clarification..."
To continue to make progress in any scientific field, researchers need to be able to clearly summarize the state of that progress. This is difficult to do for trust research because of the widely divergent ways in which trust has been defined. "Efforts to measure trust...are so variegated that the results of any two or more studies are not necessarily comparable." (Golembiewski & McConkie, 1975: 132) In order for researchers to make sense of the empirical and theoretical literature on trust, ways of categorizing each type of empirical and theoretical trust construct should be developed. In this way, trust research can move forward appropriately--and its progress can more easily be judged. Having some measure of agreement on specific definitions of trust types should make a discussion of the antecedents and consequents of trust more meaningful and specific. Definitional clarity represents a greatly neglected topic in the trust literature. Yet it is a key to long-term progress. Schwab (1980) argued that researchers have overemphasized covariation between constructs (substantive research) and underemphasized construct validation. While these are both important endeavors, our reading of the trust literature found that both empirical construct validation and substantive research have been overemphasized while integrative conceptualization has been underemphasized (Kee & Knox, 1970; Lewicki & Bunker, 1995; Taylor, 1989; Williamson, 1993). Effective conceptualization is critical to the success of construct validation and substantive research. Schwab (1980) endorsed James and Jones's (1974) assessment that adequately defining the conceptual meaning boundaries of a construct should take priority over construct measurement and construct validation. Without clear conceptual definitions, the overall nomological network will still be fraught with puzzling gaps and overlaps. Schwab (1980: 6) outlined a related challenge: "constructs are of interest only if they are connected to other constructs." Similarly, Wrightsman said, "...research is needed on the relationships among the several recent measures of trust." (1991: 411) Hence, differentiating several types of trust is only effective if these types can be shown to relate to each other, and to other constructs, in some sensible manner. This intensifies the importance of building a model of trust constructs that fit together in meaningful ways--this paper's primary objective.
We believe a cross-disciplinary conceptualization of trust constructs would be valuable. By creating a cross-disciplinary set of trust concepts, researchers in various fields could study types of trust that could be related to each other. Then work by researchers in one field could be compared to work in other fields. In this way, researchers will make cumulative progress on trust. However, this approach is not without its potential pitfalls. Van de Ven & Ferry (1980) applied cross-disciplinary concepts to their organizational assessment framework. In so doing, they found that the potential danger of this strategy lies in "using concepts incorrectly because they were abstracted from their parent disciplines...[O]ne may lose sight of the paradigm origin or base [and develop] an eclectic conceptual model that is without 'roots.'" (1980: 376) To guard against these problems, they defined their concepts "as clearly as possible," grounded them in the originating literature, and evaluated their framework for "logical validity and consistency" (1980: 376). This paper attempts to follow Van de Ven & Ferry's example.
Van de Ven & Ferry (1980) also pointed out the high payoff to researchers and practitioners when such an endeavor is successful: "The value in taking these risks lies in the potential for conceptual advances that are present when concepts from different paradigm origins are juxtaposed to create a new paradigm" (Ibid.). To the extent that this paper's trust typology is useful for the management discipline, its usefulness is largely due to its cross-disciplinary nature. This is because an interdisciplinary approach enables a richer, more well-balanced view of a phenomenon. This seems especially important for phenomena in organizations, in which complex factors are present. The resulting theory is therefore likely to be more useful for practice. "Nothing is quite so practical as a good theory" (Van de Ven, 1989: 486).
Although definitional diversity can lead to confusion, we do not assert that diversity of trust definitions is completely wrong or improper. Rather, we argue that trust is appropriately difficult to define narrowly. This paper analyzes the trust literature and common usage (dictionary) trust meanings to further understand why trust definitions are so difficult to specify. We will show that trust refers to a relatively broad set of constructs, both in terms of the trust research literature and in terms of everyday usage of the term. We will argue that narrow definitions of trust do not accurately depict this concept's rich set of meanings. Hence, we suggest that trust be characterized as a set of inter-related constructs.
In Section 1, we discuss the issue of divergent trust definitions. To begin to address this issue, this paper creates two kinds of conceptual typologies: typology type (a)--a classification system for types or kinds of trust; typology type (b)--a set of six related types of trust constructs resulting from the analysis of the classification system (Tiryakian, 1968). By classification system, we mean a sensible method of differentiating one conceptual type from another. In Section 2, we propose a classification system [typology type (a)] for trust constructs to better understand the definitional problem. By a set of related types of a concept, we mean a group of constructs that are conceptually distinguishable, but which relate to each other in specified ways. In Section 3, guided by the classification system, we define six related types of trust [conceptual typology type (b)], and briefly relate them to each other. These six definitions:
The relations between the six trust types are more completely explained elsewhere (McKnight & Chervany, 1995; McKnight, Cummings & Chervany, 1996). We will argue that understanding the meanings of trust in both everyday and scientific usage is important to creating a better definition of trust for scientific purposes.
The paper adds value in two ways. First, it develops a way to classify types of trust, benefiting our understanding of how to evaluate the trust literature. Second, through synthesizing existing research- and common use meanings of trust, it provides a set of interrelated trust constructs that are useful for research and practice. Research implications and management applications of the model are included in the Conclusion.
In 1967, Giffin said that trust "has been viewed as a somewhat mystical and intangible factor, probably defying careful definition" (1967: 104). Twenty-two years later, Taylor (1989: 85) found that if one examined the available library research on trust, "a bewildering array of meanings and connotations for trust would be provided." Today, researchers are still far from a consensus on what trust means. In fact, they remark that trust definitions have become a "confusing potpourri," (Shapiro, 1987a: 625), a "conceptual confusion" (Lewis & Weigert, 1985a: 975), even a "conceptual morass" (Barber, 1983: 1; Carnevale & Wechsler, 1992: 473). Trust is an "elusive" concept (Gambetta, 1988: ix; Yamagishi & Yamagishi, 1994: 130).
After reviewing the trust literature, we find that these statements are scarcely exaggerations. Table 1 summarizes the divergent ways in which trust has been defined in sixty research articles or books. Eighteen of the sources come from the management  /communication-related literatures, nineteen come from sociology/economics/political science, and twenty-three from psychology or social psychology. This range of literature was chosen to reflect the breadth of major research efforts about trust both from the management discipline and from disciplines supporting the management discipline. A particular work on the list was selected either because it is cited by others in the literature, or because it represents a unique view about the definition of trust. Note in Table 1 that we have momentarily excluded the important recent work of several researchers (Bromiley & Cummings, 1995; Dobing, 1993; Mayer, Davis & Schoorman, 1995; McAllister, 1995; Mishra, 1996). Their work will be included in the discussion later. Many other useful and important articles on trust were excluded from Table 1 either because they adopted someone else's definition of trust instead of creating their own, or because they never stated what trust means (e.g., Boyle & Bonacich, 1970; Conviser, 1983; Granovetter, 1985; Kramer, 1994; Lewicki & Bunker, 1995; Shapiro, Sheppard & Cheraskin, 1992).
Personal /Interpersonal _______________________________________________ Impersonal Affective State Cognitive State ________________________ Structural Dispositional Attitude Feeling Expectancy Belief Intention Behavior Article/Book Anderson & Narus, 1990 X x Baier, 1986 x x X Barber, 1983 X x X Blakeney, 1986 X Bonoma, 1976 x X Bradach & Eccles, 1989 X Carnevale et al., 1982 X Coleman, 1990 X Currall & Judge, 1995 X X Dasgupta, 1988 X Deutsch, 1973 X X X Driscoll, 1978 X Dunn, 1988 X X X X Erikson, 1968 X Fox, 1974 X X X Gabarro, 1978 X Gaines, 1980 X Gambetta, 1988 X X Garfinkel, 1967 X X Giffin, 1967 x X Good, 1988 X Heimovics, 1984 X X Holmes, 1991 x X Hoy & Kupersmith, 1989 X Husted, 1990 X X Johnson & Johnson, 1989 X X Johnson-George & Swap, 1982 X X Kasperson et al., 1992 X Kee & Knox, 1970 X X X Kegan & Rubenstein, 1973 X Koller, 1988 X x x x Krackhardt & Stern, 1988 X Lewis & Weigert,1985ab X X X X Lindskold, 1978 X Lorenz, 1988 X Luhmann, 1991 X X X Luhmann, 1988 X x McGregor, 1967 X X X McLain & Hackman, 1995 x X X Orbell et al., 1994 X X Rempel et al., 1985 X X Riker, 1971 X X X Ring & Van de Ven, 1994 X Roberts & O'Reilly, 1974 X Rotter, 1967 X X Sato, 1988 X X Scanzoni, 1979 X X Schlenker et al., 1973 X Shapiro, 1987a X Sitkin & Roth, 1993 Solomon, 1960 X X Taylor, 1989 X X Thorslund, 1976 X X Williams, 1988 X Worchel, 1979 X X Yamagishi et al., 1994 X x Zaheer et al., 1993 X Zaltman et al., 1988 X Zand, 1972 X Zucker, 1986 X X _____ _____ ____ ____ _____ ____ _____ _____ TOTAL 7 5 3 19 23 27 9 20 % of Total 6% 4% 3% 17% 20% 24% 8% 18%
1. A large 'X' denotes authors' primary definition of trust. A small 'x' denotes other than primary definition.
2. Feeling includes specific affective states (besides Attitude), such as confidence and security.
3. Expectancy includes expectations.
4. Belief includes other cognitive perceptions or assessments.
5. Intention includes cognitive choices and behavioral estimates.
The top of Table 1 shows categories of construct types. The three major categories are Impersonal/Structural, Dispositional, and Personal/Interpersonal. Researchers have characterized trust as anything from an impersonal structural construct to an interpersonal behavior construct.  Impersonal/Structural means that trust is founded upon social or institutional structures in the situation, not on personal attributes of the trusting or trusted parties (Lewis & Weigert, 1985b). Further, Impersonal/Structural refers to those definitions of trust that differentiate it from being a property or state of a person or persons. Rather, it is an institutional property, either in terms of the natural (Garfinkel, 1967) or social/organizational (Shapiro, 1987a) situation. By Dispositional, we mean that trust is based in the personality attributes of the trusting party. That is, the trustor has a general tendency to trust others across situations (Rotter, 1967, 1971), or has a general faith in human nature (Rosenberg, 1957; Wrightsman, 1991). Erikson describes Dispositional trust as "a sense of basic trust, which is a pervasive attitude toward oneself and the world," an "essential trustfulness of others as well as a fundamental sense of one's own trustworthiness" (Erikson, 1968: 96). By Personal, we mean that one person trusts another person, persons, or thing(s) in the situation. That is, the trusting entity is one person, and trust is directed to another party or parties. By Interpersonal, we mean that two or more people (or groups) trust each other in the situation. That is, the trusting entity is at least two persons.
Our categories of trust constructs implicitly include one other important consideration (Johnson-George & Swap, 1982): the situation. The situation is implicitly included because Dispositional means cross-situational, while Personal/Interpersonal and Impersonal/Structural are, by definition, situation-specific (see category definitions above). Since trust is often defined as a verb,  the direct object of the verb trust also seems important to include in our categories. Table 1 does not explicitly contain the object of trust. Implicitly, however, the object of Personal or Interpersonal trust is likely to be a person (e.g., to trust a specific person), while in the Impersonal/Structural category, trust is more likely to refer to an institutional structure (e.g., to trust that the judicial system will uphold contract law). The research literature generally uses people as the object of Dispositional trust (e.g., to generally trust other people).
Note in Table 1 that an attempt was made to break down research definitions to the lowest level possible. Hence, for example, we only placed an 'X' in the attitude column when the researcher explicitly called trust an attitude without breaking the attitude down further into beliefs, intentions, and affect (cf. Cummings & Bromiley, 1996).
Table 1 reveals several things about the trust literature. First, trust is most often defined in terms of expectancies or beliefs. Expectancies and expectations reflect the future orientation of trust. Beliefs reflect the critical role perceptions about the other party play in trust. Second, many definitions include affective, or cognitive/affective, aspects. These definitions of trust typically include a phrase about feelings of security about, or confidence in, the trusted party (e.g., Rempel, Holmes & Zanna, 1985: 97 discuss "emotional security"). Third, note that a large number of definitions refer to trust as a behavior.
Also, note the considerable breadth of coverage of the types of definitions. If this were a test of consensus on trust definitions, then researchers would receive a very low consensus rating. By reading across the rows, we found that thirty-six of the sixty articles or books (sixty percent) define trust in more than one conceptual category. On average, these researchers used 1.9 categories. Hence, most of these individual researchers feel trust has more than one meaning. This is additional evidence of the breadth of the trust concept.
But trust definitions diverge not only based on the type of construct, as Table 1 shows, but also diverge regarding the perceived attributes of the trusted party. Table 2 shows the attribute dimension of trust conceptualizations. Thirty of the sixty articles or books in Table 1 state that trust refers to a perceived attribute, or set of attributes, of the person trusted. These sources generally define trust in terms of beliefs or expectations about the other person (e.g., P believes O is competent; P expects O to act benevolently). These definitions have proliferated into a large number of attributes that we clustered into sixteen groupings.  The notes at the end of Table 2 explain which definitions are grouped into the categories shown. Some researchers mention only one attribute (e.g., Husted, 1990), while others list as many as six or seven (Blakeney, 1986; Giffin, 1967). On average, they list 2.7 attributes. Certain attributes are mentioned more frequently, such as benevolence/caring/concern (14), competence (10), good will/good intentions (10), and honesty (7). In spite of these four categories comprising just over half of the eighty entries, Table 2 shows a rather divergent set of attributes of the person trusted. This is often due to the specific context being studied. For example, Giffin (1967) addressed trust (credibility) in speech communication. Hence, appearance-related attributes like dynamism and personal attraction were more important to his research than were benevolence or honesty of the person trusted. On the other hand, Gaines (1980) studied relationships between subordinates and superiors. Here, the superior's benevolence was the key to the subordinate's trust. When Table 2 is considered along with Table 1, a good view of the diversity of trust definitions can be seen. Not only do trust definitions vary significantly in terms of type of construct (Table 1), but they also vary in terms of the attributes of the person trusted--the belief or expectancy referent (Table 2).
e s m y y s n s y y y y d e g n Article/Book Anderson & x x Narus, 1990 Baier, 1986 x x Barber, 1983 x x Blakeney, 1986 x x x x x x Bonoma, 1976 x x x x Dunn, 1988 x Gabarro, 1978 x x x x x Gaines, 1980 x Giffin, 1967 x x x x x x x Heimovics,1984 x x x x Holmes, 1991 x x Husted, 1990 x Johnson-George & Swap, 1982 x x x x x Kasperson, et al, 1992 x x x Kee & Knox, 1970 x x Koller, 1988 x x x x Krackhardt & Stern, 1988 x Lindskold, 1978 x x McGregor, 1967 x McLain & Hack- man, 1995 x x Rempel et al, 1985 x x x x x Ring & Van de Ven, 1994 x x Sato, 1988 x x Sitkin & Roth, 1993 x Solomon, 1960 x Thorslund, 1976 x x x Worchel, 1979 x x Yamagishi & Yam- agishi, 1994 x Zaheer & Venkat- ramen, 1993 x x Zaltman & Moor- man, 1988 x x x ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ ___ TOTAL 10 3 3 6 6 10 14 4 7 1 6 5 2 2 0 1 % of Total 12% 4% 4% 8% 8% 12% 18% 5% 9% 1% 8% 6% 2% 2% - 1%
1. Benevolent/Caring/Concern includes not acting opportunistically or manipulatively, and favorable motives
2. The definitions of Baier (1986), Giffin (1967), and Husted (1990) are stated in terms of behaviors or actions, not beliefs in personal attributes
3. Honesty includes integrity and sincerity
4. Competence includes ability, capability, good judgment
5. Predictability includes consistency
6. Careful/safe includes keeping confidences
We believe that Table 2's conceptual propagation has occurred for a reason: what is important to trust varies from situation to situation. Therefore, in order to instill trust in the other person, the attributes the trusted person must possess differ from relationship to relationship and from situation to situation. For example, a client trusts her physician to diagnose and treat her malady properly. This means that the client believes the doctor desires to help and is capable of helping. But the patient doesn't care whether the doctor is motivated by a desire to make money or by the goodness of her heart (benevolence). The patient does need the doctor to be competent, however, or the illness will be improperly diagnosed or ineffectively treated. Therefore, in this situation, the doctor's competence is vital, while her benevolence is not as important. By contrast, in the parent/child relationship, the child is more likely to be concerned with benevolence, since the child cannot readily choose a new parent based on competence anyway. However, this emphasis will vary from situation to situation. In fact, benevolence is important in a physician/patient relationship when there arise interpretive treatment choices, in which the physician has a monetary incentive to prefer one treatment over another--perhaps at the patient's expense. When a child is failing in mathematics, the parent's mathematical competence may be more important to the child than is the parent's benevolence toward the child. Since different situations require different attributes of the other person, some proliferation of these attributes is natural and proper as researchers look across a range of situations in which to apply trust.
Overall, then, trust has been defined by researchers in many different ways. This situation, called homonymy, means that one label encompasses more than one construct (Smith, 1990). The severity of homonymy in trust definitions is such that Lewicki & Bunker (1995) compared it to the story of the six blind men and an elephant. Each man perceived the elephant ("trust") to be something different (e.g., a rope, a wall, a tree), because of the narrow portion of the elephant which they blindly felt (e.g., the tail, the side, the leg, respectively). They each thought the elephant was what they felt because they were unable to see the big picture of what an entire elephant is like. Based on Tables 1 and 2, the story of the blind man and the elephant probably exaggerates the situation. Many trust researchers have examined more than one type of trust. Still, individual researchers have conceptualized trust in relatively narrow ways, given the breadth of meanings trust has accumulated across researchers. On the whole, researchers have tended "to deal with two or a few variables at a time" (Golembiewski & McConkie, 1975: 149).
In addition, the definitions of trust in the literature tend to reflect the paradigms of the particular academic discipline of the researcher (Lewicki & Bunker, 1995; Lewis & Weigert, 1985b). For example, while sociologists tend to see trust as structural in nature (e.g., Garfinkel, 1967; Lewis & Weigert, 1985b; Shapiro, 1987a), some psychologists have viewed trust as a personal attribute (e.g., Erikson, 1968; Rotter, 1967). Social psychologists are more likely to view trust as an interpersonal phenomenon (e.g., Deutsch, 1973; Holmes, 1991). Economists are more inclined to view trust as a rational choice mechanism (see Williamson, 1993).
Not all studies have approached trust in a narrow way. Some researchers have analyzed more than one or two of the dimensions of trust. Examples include: Barber (1983), Butler (1991), Lewis & Weigert (1985a,b), Luhmann (1991), and Zucker (1986). Note that, with the exception of Butler (1991), these pieces were theoretical in nature. In an empirical article, researchers have a norm to include only the particular portion of a theory they are testing (Sutton & Staw, 1995). Much of the trust research has been empirical--in which each study measures a particular aspect of trust. Hence, many studies provide unique definitions of trust.
This raises the question asked two decades ago by Golembiewski & McConkie (1975): If one researcher defines trust differently from another researcher, how can the theoretical formulations and the empirical results of researchers build on each other? The work on trust will not be convincingly cumulative unless researchers can compare the results of one study with the results of another study. This will not adequately occur unless scholars understand the range of trust's conceptual meanings and then purposely conduct research in a specific portion of the conceptual range.1.2 Suggestion for Individual Researchers: Carefully Broaden One's Theoretical Conceptualization of Trust
In the aggregate (Tables 1 and 2), trust definitions have become too broad (Golembiewski & McConkie, 1975). On the other hand, individual studies tend to define trust too narrowly. In fact, in a given paper, the definition of trust is typically very narrow compared to the aggregate diversity of definitions (Kee & Knox, 1970; see Tables 1 and 2).
The issue of divergent trust definitions is similar to the problem of having many diverse theories explaining the same dependent variable. The latter problem was addressed by the editor of the 1989 Academy of Management Review special issue on theory building. Van de Ven (1989: 487) warned that when theories on a topic widely diverge, the advocates "for each theory engage in activities to make their theory better by increasing its internal consistency, often at the expense of limiting its scope...[S]uch impeccable micro logic is creating macro nonsense!" We found evidence that this is happening in terms of definitions in the trust literature. For example, Lewis and Weigert (1985b) depicted the differences in trust definitions used by psychologists and sociologists in incommensurable terms. That is, trust definitions cannot be compared because they have different bases in psychology and sociology. Lewis and Weigert did not allow the possibility that trust could be an individual characteristic. To counter such tendencies, Van de Ven recommended that scholars should "deal with the tensions, debates, and forced choices between overly narrow competing theories" by using these "inconsistencies and contradictions between theories...to develop better and more encompassing theories." (1989: 488) Similarly, Luhmann (1991: 3) warned against devaluing the "traditional range of meaning" of the word trust. This counsel suggests that individual researchers should broaden their conceptual views of what trust is.
Note that this is not the proper advice for researchers as a whole. Tables 1 and 2 show more than enough divergence. In fact, the suggestion we have for the trust literature as a whole is to try to narrow its view of what trust is. That is, we suggest that the need exists to narrow the collective perspective on trust conceptualizations at the same time as researchers broaden their individual perspective on trust conceptualizations. Similarly, practitioners need to come to a consensus on how to define trust, so that they can communicate--with each other and with researchers--on this important topic.
This suggestion can now be framed in terms of what the rest of this paper will do. The divergence of narrow trust definitions should motivate individual researchers to broaden--and thus improve--the theoretical basis of the study of trust. Overall, researchers and practitioners need to build consensus toward a manageable number of meaningful types of trust. If several types of trust really exist, what are they, and how do they fit together? We propose that the primary challenge is twofold: (a) to understand the nature of the various types of trust; (b) to synthesize and define a broad, parsimonious, and cohesive set of useful trust types. The resulting set of constructs should effectively serve theoretical, empirical, and practical objectives. Section 2 will address (a) by attempting to expand understanding of the meanings of trust through use of a classification system--a type (a) trust typology. Section 3 will address (b) by defining and then briefly relating six types of trust--a type (b) typology.
Tables 1 and 2 showed that social scientists have used widely diverging meanings of the term trust. In Section 2.1, we show that trust has numerous and diverse meanings in everyday usage as well. An analysis of common usages of the word trust supports the suggestion to broaden trust conceptualizations. Studying a single narrow type of trust does not adequately capture the breadth of meaning assigned to the word trust in everyday usage.
Social scientists like Harold H. Kelley (Kelley, 1992) and trust researchers in particular (e.g., Barber, 1983; Bromiley & Cummings, 1995; Strickland, Cafferty, Allen, Klecka & Silver, 1968), have noted the importance of relating scientific terms to their everyday usage counterparts. This technique is especially important for building good conceptual theory. In social psychology, Berscheid & Meyers (1996) gave as an example, "The extraordinary theoretical fruits that Heider's (1958) systematic mining of 'common-sense psychology' produced for social psychology." In the trust literature, Gabarro (1978: 295) purposely "ignored" literature definitions of trust and "probed...for the person's own conceptions of trust." Gabarro's study yielded a large number of useful trust dimensions. Deutsch (1973) and Coleman (1990) each went through four examples of everyday trust situations. To improve his own trust conceptualization, Fox (1974) compared dictionary definitions of trust with those of Zand (1972) and Deutsch (e.g., 1973), and noted that the scientific definitions were similar to those in the dictionary.
To be effective, then, scientists should start with, or at least be grounded in, common terms (such as trust). Writing to members of his discipline, Kelley (1992) called these "common-sense psychology" terms. Then, per Kelley, scientists should try "to extract from [common-sense psychology] the essence of everyday terms that lend themselves to [scientific psychology] uses." (Kelley, 1992: 11) Finally, the terms used in scientific models should be compared back to common terms to see how well they approximate the meaning, and range of meaning, the everyday terms connote (Sagasti & Mitroff, 1973). In the preface to his timeless treatise on trust, Luhmann (1991) suggested that sociologists form a theory of trust and "then enter a dialogue with the everyday understanding of the social world" (1991: 3). Since the term trust is used so frequently in everyday language, Kelley's and Luhmann's advice seems particularly appropriate. This dialectical interplay between common terms and scientific terms improves the effectiveness of science; it improves the practical applicability of the scientific, and renders researchable the common.
Using common-sense terms to help create scientific definitions has its potential drawbacks, however. Berscheid & Meyers (1996) point out that using common-use examples to help define scientific terms may only provide a "toehold," since ordinary language will not provide particularly novel definitions. Rather, common language will supply ordinary meanings. Kelley (1992) and Barber (1983) pointed out that common-use definitions need to be sharpened for scientific use.
But using everyday meanings of terms like trust has large potential benefits. In fields of applied research, where concepts and theories are communicated back to practitioners for potential use, it is important that scientists create and use concepts that practitioners can understand, and definitions with which practitioners can agree. In order for this to occur, the everyday meanings of the term 'trust' should be properly understood by trust researchers (cf. Berscheid & Meyers, 1996). This seems implicit in Cook & Wall's (1980: 39) comment that "Trust as a common word in ordinary language retains much of that meaning when employed as a concept in social science." Unless this happens, researchers will create and study concepts that diverge significantly from lay use. When this occurs, they will be unable to convince practitioners that their terms have the meaning they claim they have in real world situations. In fact, researchers and practitioners will be unable to communicate effectively, because their views of the same concept are so different. An example of this has occurred in research on love (Berscheid & Meyers, 1996). Fehr & Russell (1991: 435) point out that "Scientists doing research on love are probably subscribing to a much narrower concept of love than are their subjects." The problem with this is that "The natural language concept of love guides people's official and unofficial interpretation of some of life's major and minor events, and that concept must be understood--as it is." (Fehr, 1991: 436) Too narrow a view of a concept like trust will not necessarily prove helpful as an everyday guide for people. Berscheid & Meyers (1996) cite Heider's (1958) observation that people are guided by what they believe to be true--including their implicit definitions of concepts. If people are guided by their own conceptualization of a term like love or trust, their conceptualization will likely guide their actions much more strongly than a scientific conceptualization to which they have trouble relating. While scientific uses of concepts will diverge from those of practice, part of the job of social scientific inquiry is to develop ways to relate scientific terms to real world use (Sagasti & Mitroff, 1973). In so doing, scientists can improve their own set of definitions of important concepts (Berscheid & Meyers, 1996).
In the trust domain, one way to conduct a dialectical analysis between common and scientific definitions is to compare scientific definitions of trust to dictionary definitions of trust. This is already common practice in the trust literature (e.g., Barber, 1983; Giffin, 1983; Taylor, 1989). Dictionaries attempt to stay abreast of the common usages of words as the words dynamically change over time. For example, Stein (1971: v) says, "If modern man is to function well in his society, one of his necessities, surely, is to keep pace with the dynamic growth of his language. To meet such a need The Random House Dictionary of the English Language has been prepared." Because dictionaries try to reflect current common usage of words, we can see how broadly or narrowly people use the word trust--and its particular everyday meanings-- by an examination of the dictionary meanings of trust.
A look at three specific dictionaries' trust definitions provides one indication of the breadth of usage of the word trust. Table 3 compares the number of definitions and the length of entry for trust and five other terms. Trust has more (and longer) definitions than the words that Mayer, Davis & Schoorman (1995) differentiate from trust: cooperation, confidence, and predictability. But trust generally has somewhat fewer and shorter definitions than the very vague interpersonal terms, love and like. Hence, (a) trust is broader than Mayer, Davis & Schoorman's (1995) three similar terms; (b) trust approaches the breadth of the terms love and liking--both of which are very broad and difficult to define terms (Brehm, 1992).  Taylor (1989) apparently conducted, but did not specifically report, a similar review. Taylor (1989: 85) said that if someone were to compare the definition lengths of trust and of other concepts, "[they] would most likely conclude--and quite correctly--that trust is not a singular or simple concept."
Random House Webster's Oxford (1971) (1981) (1989) Concept Counts Cooperation #Def'ns 6 3 2 #Lines 15 14 75 Confidence #Def'ns 8 6 13 #Lines 27 41 234 Predictable #Def'ns 2 1 1 #Lines 27 4 21 Trust #Def'ns 24 9 18 #Lines 57 112 633 Love #Def'ns 24 17 28 #Lines 54 82 1670 Like #Def'ns 32 31 40 #Lines 74 274 1515
1. #Lines refers to the number of lines devoted to the definition of the word.
2. #Def'ns means the number of ordinally numbered definitions, regardless of their number of subheadings (e.g., 2a 2b = 1 #Def'n; 1a 1b 1c = 1 #Def'n)
Even more telling is an analysis of the conceptual range of these dictionary trust definitions. Table 4 shows the conceptual range of the Random House (Stein, 1971) trust definitions plotted across the dimensions Tables 1 and 2 used to map the research literature. Table 4 uses Concept Types from Table 1 across the top and Attribute types from Table 2 down the left hand side. The left hand side also includes Behavior and Miscellaneous referent types. By referent, we mean an explanatory category. Just as the Attribute Types explain what characteristic one believes or expects in the other person, so Miscellaneous Types generally explain what is the object of the (verb) trust. Sometimes the object of trust is a person; sometimes it is a thing. The Behavior Types simply further categorize the behaviors into subtypes, indicating reliance, dependence, or commitment. Table 4 shows the relatively broad range of trust meanings found in one dictionary (Stein, 1971).
Personal /Interpersonal _______________________________________________ Impersonal Affective State Cognitive State ______________________________ Structural Dispositional Attitude Feeling Expectancy Belief Intention Behavior Article/Book Referent Attribute Types Competence X Expertness X Dynamism Predictability Goodness/Morality Good Will/Intentions Benevolent/Care/Concrn Responsiveness Honesty X Credibility Reliability Dependability Openness/minded Careful/Safe X Shared Understanding Personal Attraction Behavior Types -Rely on X -Depend on X -Commit to X -Commission X Misc. Types Other person X X Something X X X X Something/someone X Oblig/Responsibility X Not specified X X X
Though the Random House dictionary focuses primarily on behavior (Table 4), its range of trust definitions is still rather large. It portrays trust as a reliance, a dependence, a competent-, expert-, honest- or safe- behavior. It defines trust as a belief, confidence, or expectation (without specifying a referent). It defines trust across the horizontal axis as a structural concept, an affective concept, a cognitive state, and a behavior. Hence, Stein portrays trust as a broad, hard-to-narrowly-define concept. Since the Random House dictionary attempts to closely match actual word usage, we may infer that trust is a concept with many everyday meanings. We believe that for frequently used terms, it is important that scientific concepts reflect common usage. Hence, this discussion supports the suggestion that research views of the trust concept should be broadened.
In essence, Table 4 presents a classification system (type [a] typology) for the term trust. Across the top, it shows the types of concepts (e.g., belief). On the left hand side, it shows the specific referent of the concept type. For example, the belief concept has the other person's attribute as a referent. Similarly, a behavior concept has a type of behavior as a referent.2.2 Scientific Usages of the Term Trust
The dictionary tries to present the meanings of a word from a broad range of common-use sources. It may or may not give the source. But it does categorize uses into types. We have tried to capture these types of trust definitions in the Table 4 categories. The same thing could be done for scientific-use definitions. From the articles and books in Tables 1 and 2, we can see that the research definitions plotted against Table 4 would fill up a large fraction of the cells of the matrix. From this, we can see that trust is a very broad concept to the scientific community. In fact, trust definitions have probably proliferated more in scientific use than in common use. This is somewhat deceptive, however, in that several of the columns in Table 4 are more useful to scientific practice than to everyday language. Also, the dictionary would only include uses of the word trust that are frequently encountered.2.3 Past Efforts to Broaden Trust Conceptualizations
Some researchers have tried to incorporate aspects of the broad common meanings of trust by looking at dictionary definitions (e.g., Barber, 1983; Dobing, 1993; Fox, 1974; Giffin, 1967; Good, 1988; Lindskold, 1978). However, each researcher typically uses only a small portion of the range of trust definitions the dictionary contains. For example, compare Barber's own definitions of trust (Table 5, second column) to his list of three definitions of trust from his dictionary (Gove, 1981) (Table 5, first column).
Trust Definitions in Webster's Barber's Trust Definitions in Third New Internat'l Dictionary The Logic and Limits of Trust 1. assumed reliance on some 1.the expectation of technical- person or thing; a confident ly competent role performance dependence on the character, (e.g., expert knowledge, tech- ability, strength, or truth nical facility, or everyday of someone or something routine performance) 2. dependence on something 2.expectations that the natural future or contingent; order--both physical and confident anticipation... biological--and the moral social order will persist and be more or less realized 3. a charge or duty imposed 3. expectations of fiduciary in faith or confidence or obligation and responsibility, as a condition of some that is, that some others in relationship our social relationships have moral obligations and respon- sibility to demonstrate a special concern for other's interests above their own
Barber's third definition corresponds closely to the third dictionary definition. Barber's first definition corresponds closely to the first dictionary definition. However, Barber made his first and third definitions significantly less vague and overlapping than those of the dictionary. In the process, Barber's first definition also became much more narrow. Barber's dictionary said trust was a reliance on a person or thing; Barber limited his definition to reliance on a person only. The dictionary referred to a confident dependence on a person's or thing's attributes of character, ability, strength or truth. Barber narrowed the four dictionary attributes (character, ability, truth or strength) into one attribute (role competence). He eliminated the dictionary's dimensions of confidence ("confident dependence") and assumption ("assumed reliance"). The conceptual definition type (e.g., belief, behavior) changed from a behavior (Webster's) to an expectation (Barber), following Luhmann (1991). Barber also changed the rather vague dictionary definition 2. (dependence on something) into a specific expectation, basing that expectation on work by Garfinkel (1967). In the process, he left behind the dictionary's concepts of dependence and confident anticipation.We should note that Barber was citing only Webster's trust definitions 1a, 2a, and 5a(1). What were the other dictionary definitions?  A look at Webster's (Gove, 1981) shows several more trust meanings. Barber probably cited the most appropriate of Webster's definitions, since: definitions 2b, 3a and 3b refer to financial trusts, definition 4 is archaic, and 5a(2), 5b, and 5c refer again to a fiduciary obligation. But Barber forfeited several conceptual meanings by not using the following of Webster's noun definitions of trust:
From the above analysis, we constructed Table 6, which shows the contrast in coverage between Barber's definitions and those of his dictionary. Webster's does cover a significant amount of conceptual space, and is conceptually similar to the definitions from the Random House dictionary (see Table 4). Notice especially how thoroughly Webster's covers the behavior and feeling (confidence/security) columns in Table 6.
Personal /Interpersonal _______________________________________________ Impersonal Affective State Cognitive State ______________________________ Structural Dispositional Attitude Feeling Expectancy Belief Intention Behavior Article/Book Referent Attribute Types Competence * WB Expertness Dynamism WB WB Predictability Goodness/Morality WB * WB Good Will/Intentions Benevolent/Care/Concrn Responsiveness Honesty WB WB Credibility Reliability Dependability Openness/minded Careful/Safe Shared Understanding Personal Attraction Behavior Types -Rely on WB WB -Depend on WB WB -Commit to -Commission Misc. Types Other person WB WB WB Something Something/someone Oblig/Responsibility Not specified
Barber made the trust concept more manageable and closer to common usage by consulting a dictionary. This was intentional on his part, since he had noticed with dismay his own, and others', vague use of the word trust in an earlier work: "I saw on reflection that I had not defined trust or made clear just what I meant by it...I looked to see how others deal with it, only to discover that my own shortcoming often occurs in distinguished quarters." (Barber, 1983: 2-3) But while Barber's trust conceptualization is manageable, a significant number of common usages were left out of Barber's conceptualization (compare with Table 6's row and column headings):
In spite of this, we believe Barber made an outstanding contribution toward defining trust in a way that was useful both to scientists and to practitioners. Barber's book on trust is a classic.2.4 Recent Efforts to Broaden/Clarify Trust Conceptualizations
What has happened since Barber? Scholars continue to struggle with how trust should be defined. For example, rather than using existing trust conceptualizations, five recent studies developed new conceptual and/or empirical views of trust (Bromiley & Cummings, 1995/Cummings & Bromiley, 1996; Dobing, 1993; Mayer, Davis & Schoorman, 1995; McAllister, 1994; Mishra, 1996). Each of these studies used a multi-dimensional view of trust. This action underscores the felt need to conceptualize trust more broadly. Table 7 shows where these five studies fit in the classification typology.
Personal /Interpersonal _______________________________________________________________ Impersonal Affective State Cognitive State _________________________________ REFERENT Structural Dispositional Attitude Feeling Expectancy Belief Intention Behavior Attribute Types Competence D/mds/Mc/M Expertness Dynamism Predictability Goodness/Morality Good Will/Intentions Benevolent/Care/Concrn BC/Mc BC/mds/M BC Responsiveness Honesty BC BC/mds/D BC Credibility Reliability Mc/M Dependability Mc Openness/minded M Careful/Safe D Shared Understanding D Personal Attraction Behavior Types -Rely on -Depend on -Commit to -Commission Miscellaneous Types Other person mds Mc mds D D/mds/M Something Something/someone Oblig/Responsibility D
Bromiley & Cummings (1995) conceptualized trust in terms of three belief dimensions. They operationalized trust into a three-by-three grid with nine cells (Cummings & Bromiley, 1995). One side of their three-by-three matrix has the three belief dimensions of trust (keeps commitments, negotiates honestly, avoids taking excessive advantage). The first two of these have been plotted in the honesty row in Table 7's Attribute Types. Both refer to honesty, but differ in terms of time frame (negotiating commitments precedes keeping commitments). The other side of their three-by-three matrix has three components of belief types (affective state, cognition, and intended behavior). We have mapped these to the feeling, belief, and intention columns. Dobing's (1993) view of trust included willingness to depend (an intention), trusting beliefs, and several situation-specific trusting behaviors. Dobing's definitions of trusting behaviors were tailored to the domain he studied--the relationship between an information system user and the system analyst during system development. In Table 7, we have mapped Dobing's definitions in the belief, intention, and behavior columns.
Mayer, Davis & Schoorman (1995) focused on trust as a willingness to be vulnerable to another. Their trust construct is based on two types of antecedents of trust: (a) a propensity to trust (similar to a personal disposition to trust), and (b) a set of three perceptions regarding the other person's trustworthy attributes--ability, benevolence and integrity. Their model also includes risk as a moderator of the relationship between trust and risk taking. Table 7 shows their trust constructs in the dispositional, belief, and intention columns.
McAllister (1994) developed conceptual and empirical versions of trust that differentiated trust's cognitive and emotional aspects. McAllister's study found evidence for a clear distinction between affect-based and cognition-based trust, both in terms of factor separation, and in terms of distinct relationships with other concepts. He concluded: "Thus, affect-based trust and cognition-based trust represent distinct forms of interpersonal trust." (1995: 49) Table 7 shows two McAllister symbols in the Feeling column, and three in the Belief column.
Mishra (1996) defines trust as a party's willingness to be vulnerable to another party based on the belief that the other party is competent, open, concerned, and reliable. Mishra argues that trust is an overall construct that is made up of the combination of the four belief dimensions. That is, trust is a multi-dimensional construct formed by the these beliefs. Mishra also describes these beliefs as combining in a multiplicative way. "That is, a low level of trust in terms of any of the dimensions offsets high levels of trust in terms of the other dimensions." (1996: 269) We plotted Mishra's definitions in the intention column (once) and in the belief column (four times).
Each of these five recent trust conceptualizations is based on literature reviews. The researchers overlap on only six (33%) of the eighteen trust cells filled in Table 7. Taken together, these definitions cover a large portion of the trust classification typology. However, as Table 7 shows, each covers a different subset of trust constructs. Researchers in these recent studies, by choosing to reconceptualize trust, seem to be expressing dissatisfaction with existing conceptualizations. Their collective efforts to conceptualize trust multidimensionally indicate a felt need to reflect more of trust's meanings. Yet individually, none of the four recent conceptualizations of trust covers the entire range of trust meanings in the research literature or in common usage. Collectively, they cover all of the columns except impersonal structural, and ten of the twenty-five rows. Still, each provides well-grounded and useful definitions. Each defines several important and valid meanings of trust. However, if we assume that each individual definition is valid, we must conclude that trust is a broader concept than any of the five portray it to be. This means that a cohesive set of trust definitions that approximates the range of meanings of the word trust in research and everyday settings probably does not exist.
Having such a set of trust constructs is important to the objective that trust research should be cumulative. To effectively research something multi-dimensional, one must understand all its dimensions. Otherwise, one study may unintentionally overlap another. Research gaps may go undiscovered. Worse still, one study's results may contradict another's with little hope of understanding why. The same problem has been experienced in other domains. For example, Berscheid & Meyers (1996) quoted Rubin about the lack of progress caused by confusion regarding the concept of love:
We suggest that the matrix in Table 7 provides a set of 'translation rules' for trust concepts. But knowing how to categorize a trust construct is not our only concern. We also want to provide a set of consistent constructs for research and practical use. For this to be accomplished, knowing how each type of trust construct relates to the other trust constructs is critical. Tiryakian said, 
To Tiryakian, then, a good conceptual typology produces common, related entities. In the social sciences, this means that one construct in a type (b) typology relates to another. For example, beliefs, intentions and behaviors may fit together in a meaningful way if they are defined to be cohesive constructs (e.g., Fishbein & Ajzen, 1975). Triandis (1979), Fishbein & Ajzen (1975) and others (e.g., Davis, 1989) have demonstrated that related constructs can fit together such that one construct leads to, or predicts, another. For example, Davis (1989) showed that a person's belief in the usefulness of an information system predicted behavioral intention to use the system, which in turn predicted actual system use. Dobing (1993) and Mayer, Davis & Schoorman (1995) have each conceptualized trust such that their respective constructs relate to each other. Without such linkages, scientists may lose some of the predictive and/or explanatory power possible in their models of trust.
From our analysis of the literature's trust definitions, we have developed a type (a) typology--a system for categorizing (Tiryakian, 1968) the definitions of trust. This typology consists of the N by M chart shown in Tables 4, 6, and 7. In Section 1, we used this typology to show how narrow individual researcher definitions of trust are, compared with the broad totality of common use and scientific trust definitions. Section 2 used the typology to better understand research and everyday definitions of trust. In Section 3, the classification typology will guide a trust type (b) typology--a parsimonious, yet wide-ranging set of six trust construct definitions.
In the past, trust researchers have not clearly distinguished between trust and its antecedents and consequences. This "has hindered previous research on trust." (Mayer, Davis & Schoorman, 1995). In order to develop an acceptable set of trust constructs, we decided to use these criteria:
Directed by these criteria and the trust classification typology, we define six trust constructs that extend across impersonal, dispositional and interpersonal construct dimensions and Table 10's broad attributes. We will show that they are structured to be compatible with, and supportive of, each other (Figure 1). By compatible, we mean that the constructs do not overlap each other and lie at the same level of analysis. By supportive, we mean that one type of trust construct may be considered an antecedent of another type of trust construct. (Other research has also viewed trust as the relation between more than one construct--e.g., Dobing, 1993; Mayer, Davis & Schoorman, 1995; Zucker, 1986). The six constructs are: Trusting Intention, Trusting Behavior, Trusting Beliefs, System Trust, Dispositional Trust, and Situational Decision to Trust. Following what Table 1 showed about the importance of certain construct types, we include both cognitive and affective components in Trusting Beliefs, Trusting Intention, and Trusting Behavior. We will also show that the six constructs cover the more common of the dictionary definitions of trust. This multi-dimensional view of trust provides a parsimonious way to organize measurable trust types, while clearly distinguishing one type from another.
This set of six constructs is not fully comprehensive; it does not cover every type of trust in the literature. The types of trust we define were selected to cover what the authors consider the more important forms of trust, such that a broad, cohesive, yet parsimonious group of constructs emerges. The concept types covered are structural, dispositional, feeling, belief, intention, and behavior. Figure 1 shows how these constructs fit together. The detailed relationships between these trust constructs are discussed elsewhere (McKnight, Cummings, & Chervany, 1996). In general, their relationships follow the pattern of the Fishbein & Ajzen (1975) theory of reasoned action. That is, beliefs/attitudes (in our case Trusting Beliefs) lead to intentions (Trusting Intention), which, in turn, readily become manifest in behaviors (Trusting Behavior). The logic here is simple. When one has Trusting Beliefs about another, one will be willing to depend on that person (Trusting Intention). If one intends to depend on the person, then one will behave in ways that manifest that intention to depend (Trusting Behavior) (cf. Currall & Judge, 1995; Dobing, 1993). Table 8 compares the range of this set of definitions with that of the five recent studies shown in Table 7. These six constructs cover a somewhat greater definitional range than any one of the five recent studies. They cover a somewhat smaller range than the collective constructs in the five studies. As each is defined below, we will discuss where the six trust constructs lie on Table 8. The six trust constructs will be defined and distinguished from each other largely based on the dimensions shown in Table 9. Each definition implies that there are different empirical levels of each trust construct, such that each construct could be measured with a level-differentiating (e.g., Likert) scale.
Personal /Interpersonal _______________________________________________________________ Impersonal Affective State Cognitive State _________________________________ REFERENT Structural Dispositional Attitude Feeling Expectancy Belief Intention Behavior Attribute Types Competence �9 D/mds/Mc/M/**10 Expertness Dynamism Predictability �11 �12 Goodness/Morality Good Will/Intentions Benevolent/Care/Concrn BC/Mc**13 BC/ mds/M**14 BC Responsiveness Honesty BC/D**15 BC/mds/D**16 BC Credibility Reliability Mc/M Dependability Mc Openness/minded M Careful/Safe D Shared Understanding D Personal Attraction Behavior Types -Rely on -Depend on **8 **5 -Commit to -Commission Misc. Types Other person mds/**19 Mc/**2/**6 mds D D/mds/M /**1 **3 Something **17 **18 **20 Something/someone Oblig/Responsibility **7 D/**4
DIMENSIONS _________________________________________________________ CONTEXT PERSONS CONSTRUCT ________________________ __________________ ________________________________________ Situation Cross- Person- Cross- Impersonal Personal / Interpersonal Trust Type -Specific Situational Specific Personal Structural Dispos'l Feeling Belief Intention Behavior Trusting Intention X - X - - - X - X - Trusting Behavior X - X - - - X - - X Trusting Beliefs X - X - - - X X - - System Trust X - - X X - - X - - Dispos'l Trust - X - X - X - - - - Situat'l Decision X - - X - - - - X - To Trust
We define Trusting Intention as follows: the extent to which one party is willing to depend on the other party in a given situation with a feeling of relative security, even though negative consequences are possible. Trusting Intention is a situation-specific construct (Table 9): one is willing to depend on the other party in a specific situation. Trusting Intention is an intentional state: the person is ready to depend on the other in the situation. It is personal (originating in a person) and (one-way) directional: one person is willing to depend on the other. We define Trusting Intention at an individual level of analysis, as opposed to some higher level like a group or society. The individual level of analysis was chosen because it is the simplest, most elemental unit of a relationship. By elemental, we mean that two or more of these units may together form at a higher level of analysis. For example, if P has Trusting Intention toward O, and if O has Trusting Intention toward P, this forms a reciprocal dyadic Trusting Intention relationship. Obviously, additional combinations of persons could form higher levels of analysis, such as triads or groups.
Our definition of Trusting Intention embodies five essential elements synthesized from the trust literature: (a) potential negative consequences, (b) dependence, (c) feelings of security, (d) a situation-specific context, and (e) lack of reliance on control.
In Table 8, we placed a Trusting Intention mark (�1) at the intersection of Intention (top of matrix) and Other person (left side of matrix). This is because one is willing to depend on the other person. We put a second mark (�2) reflecting felt security in one's willingness to depend at the intersection of Feeling and Other person.
Figure 1 shows Trusting Intention supporting Trusting Behavior. Willingness to depend leads one to actually depend (behaviorally) on the other party. Trusting Behavior means the extent to which one person voluntarily (Lewis & Weigert, 1985a) depends on another person in a specific situation with a feeling of relative security, even though negative consequences are possible. Depends is a behavioral term, which distinguishes Trusting Behavior from the intentional term Trusting Intention (willingness to depend). When one depends on another, one confers upon the other person a fiduciary obligation to act in one's behalf (cf. Barber, 1983). We do not define the fiduciary responsibility itself as trust. Rather, Trusting Behavior means giving the other person a fiduciary obligation. In effect, the first person gives the second person some measure of power over him/her. When you give another power over you, that means you have placed yourself in a situation of risk. Hence, our Trusting Behavior construct implies acceptance of risk, much as Mayer, Davis & Schoorman (1995) argue. Table 8 shows Trusting Behavior at the intersection of Behavior and: (a) Other Person (�3), (b) Obligation/Responsibility (fiduciary) (�4), and (c) Behavior Types: -Depend on (�5). Corresponding marks have been placed in the Feeling column (�6)( �7)( �8), to reflect the feeling of security associated with depending on the other.
We may discuss Trusting Behavior in terms of many of the same ideas already discussed in Trusting Intention (Table 9). Trusting Behavior takes place in a given situation, with a specific other person. This definition of trust as a dependence on another ties both to dictionary definitions (Table 4) and the trust literature (e.g., Baier, 1986; Bonoma, 1976; Giffin, 1967; Riker, 1971; Wheeless, 1978). Trusting Behavior takes place in the presence of little or no use of control. One relies on trust, not on control (Lorenz, 1993; Ring & Van de Ven, 1994; Zand, 1972). The primary difference (Figure 1) between Trusting Behavior and Trusting Intention is that the latter is a cognitive-based construct (willing to depend) and the former is an behavior-based construct (depends). That is, with Trusting Behavior, one takes action on one's willingness to depend. We believe trust as a behavior can be directly measured in the form in which we define it here (cf. Cummings & Bromiley, 1996).
Researchers can also treat Trusting Behavior as a latent construct with measurable indicators (Riker, 1971). Specifically, we distinguish Trusting Behavior from other constructs like information sharing openness, which have sometimes been called trust (e.g., Mishra, 1993), but which are really indicators of trust. By this method, Trusting Behavior acts like a latent construct, with measurable indicators of its presence. In game theory research, for example, the act of not defecting has been referred to as trust (e.g., Deutsch, 1958). But it has also been called "cooperation" or "cooperative behavior." Using Trusting Behavior as a latent construct, the act of not defecting may be designated as an empirical indicator of the latent construct Trusting Behavior. This overcomes two weaknesses of many behavioral trust definitions: that the indicator selected (e.g., cooperation) (a) is really a consequent of trust and (b) is only one of a number of behaviors that demonstrate trust. It would be difficult to develop a list of all the types of behaviors that people believe are manifestations of trust. Arguably, any act of dependence or increasing dependence on another would fit. However, we found these in the literature:
Each of these may be used as an indicator of the latent construct Trusting Behavior. Therefore, researchers can distinguish such behaviors from Trusting Behavior by treating them as manifestations of the latent construct Trusting Behavior.
Trusting Intention, as is typical of intentions (Fishbein & Ajzen, 1975), is primarily based upon the person's cognitive beliefs about the other person (Bromiley & Cummings, 1995; Dobing, 1993; Gabarro, 1978). Yamagishi & Yamagishi (1994), for example, said that trust is based on inference about the other person's traits and intentions. We also posit that Trusting Intention is based on the person's confidence in those beliefs (McKnight, Cummings & Chervany, 1996). Together, these cognitive beliefs and belief-related confidence form a construct we call Trusting Beliefs.
Trusting Beliefs means the extent to which one believes (and feels confident in believing) that the other person is trustworthy in the situation. Lest this be viewed as using the word trust to define trust, we hasten to define the term 'trustworthy.' Trustworthy means one is willing and able to act in the other person's best interests (Driver, Russell, Cafferty, & Allen, 1968; McLain & Hackman, 1995). The Trusting Beliefs construct is shown as person- and situation-specific (Table 9). Based on an early 1995 trust literature review (McKnight & Chervany, 1995), the most prevalent (and probably the most important) trusting beliefs in the literature involve benevolence, honesty, competence, and predictability.  Benevolence means one cares about the welfare of the other person and is therefore motivated to act in the other person's interest. A benevolent person does not act opportunistically toward the other person. Honesty means one makes good faith agreements (cf. Bromiley & Cummings, 1995), tells the truth, and fulfills any promises made. Competence means one has the ability to do for the other person what the other person needs to have done. So the essence of competence is efficacy. Predictability means one's actions are consistent enough that another can forecast what one will do in a given situation.
Table 10 shows how most of Table 2's attributes can be categorized under these four broad belief categories. We constructed Table 10 by counting the times each belief attribute (e.g., dynamism, credibility, etc.) was used in Table 2's definitions of trust. Then we grouped those individual belief attributes that seemed appropriate into our trusting belief super-categories (competence, benevolence, predictability, honesty). 93.8% of the Table 2 literature's belief counts fall within these four belief categories. Hence, nearly all of the attributes found in Table 2's articles can be considered either subconstructs of, or equivalent to, the four Trusting Beliefs. In the situation, then, one with Trusting Beliefs believes that the other is benevolent, honest, competent, or/and predictable.
Table 2 Table 2 Attribute Column Broad Category Attribute Count % COMPETENCE: 1 Competence 10 2 Expertness 3 3 Dynamism 3 Subtotal 16 20.0 PREDICTABILITY: 4 Predictability 6 7.5 BENEVOLENCE: 5 Goodness/Morality 6 6 Good Will/Intentions 10 7 Benev'nt/Caring/Concern 14 8 Responsiveness 4 Subtotal 34 42.5 HONESTY: 9 Honesty 7 10 Credibility 1 11 Reliability 6 12 Dependability 5 Subtotal 19 23.8 OTHER: 13 Openness/Openminded 2 14 Careful/Safe 2 15 Shared Understanding 0 16 Personal Attraction 1 Subtotal 5 6.2 Grand Total 80 100.0%
Note the important relationship between these four beliefs and the 'willing and able' definition of trustworthiness. Benevolence is the essence of willingness to serve another's interests. Competence is the essence of ability to serve another's interests. One with honesty will prove one's willingness by making and fulfilling agreements to do so. Predictability embodies an element of temporal continuity that can be related to the other trusting beliefs. Having predictability means that one's willingness and ability to serve another's interest does not vary or change over time. As articulated elsewhere (Mayer, Davis & Schoorman, 1995; McKnight, Cummings & Chervany, 1996), the combination of the four beliefs provides a firm foundation for trusting intention and trusting behavior. That is, if one is consistently (predictably) proven to be willing (benevolent) and able (competent) to serve the trustor's interests in an (honest) manner, then one is worthy of trust indeed. We propose that such beliefs form a core cognitive construct we call Trusting Beliefs, that finds expression via Trusting Intention, which in turn is acted upon through Trusting Behavior (Figure 1).
We don't claim that these four beliefs are the only ones on which Trusting Intention is based. In a given situation, a different belief or set of beliefs than these may be important. However, one or more of these four beliefs will be salient in most organizational situations. For example, in many situations, one must rely on the other's competence. This would include whenever we require the services of a professional (e.g., doctor, lawyer, airline pilot) or an organizational co-worker or team member to do for us things we cannot do for ourselves. For those close to us, with whom we share confidential things, we depend on their benevolence to assure us certain information won't get in the wrong hands. In sales transactions in which the other party has information we don't, we depend on their honesty and benevolence to assure that they do not take advantage of us. When timing or sequence or repeatability is an issue, we depend on the other person's predictability (e.g., for delivering us something we need on time on a consistent periodic basis). Having these Trusting Beliefs helps one to be willing to depend on the other person (Trusting Intention). Hence, Trusting Beliefs is an antecedent of Trusting Intention.
The Trusting Beliefs construct encompasses belief-related confidence. This means the level of confidence one possesses in one's beliefs about the other person. By confidence, we mean a feeling of certainty regarding a belief one holds (cf. Gove, 1981). Using this construct ties closely to both dictionary and research trust definitions (e.g., Cook & Wall, 1980; Giffin, 1967; Lindskold, 1978). Table 8 reflects Trusting Beliefs at the intersection of the four attribute types (competence, predictability, benevolence, and honesty) and the Feeling and Belief columns (�9-�16).
System Trust means the extent to which one believes that proper impersonal structures are in place to enable one to anticipate a successful future endeavor (Luhmann, 1991; Lewis & Weigert, 1985a; Shapiro, 1987b). Personal attributes of the other are not at issue with System Trust (Table 9). Hence, it does not support Trusting Beliefs about the other (Figure 1); but it does support Trusting Intention. Two types of impersonal structures can be differentiated: (a) structural assurances, and (b) situational normality. Structural assurances include such safeguards as regulations, guarantees, or contracts (Shapiro, 1987a; Zucker, 1986). Situational normality may include one's own role and others' roles in the situation (Baier, 1986). Both types of System Trust relate to a specific situation (Table 9). Situational normality (type [b]) System Trust is based on the appearance that things are normal (Baier, 1986; Garfinkel, 1967) or in "proper order" (Lewis & Weigert, 1985a: 974). System Trust supports Trusting Intention in that it makes it feel safe to depend on that person because of type (a) System Trust's safeguards, which act like a 'safety net,' or type (b) System Trust's reduction of uncertainty, which enables one to feel more secure in taking risks with other people.
Our definitions revises system trust's impersonal focus by making it into a belief. This was done to make it more compatible with the other trust constructs. In our depiction, the structures themselves do not constitute a trust construct. Rather, the person's beliefs about those structures forms a trust construct. In Table 8, System Trust is found: (a) at the intersection of Impersonal/Structural and Something (�17); and (b) at the intersection of Belief and Something (�18). The "something" refers, in this context, to the impersonal structures. System Trust might be regarded as personal in that a person holds a belief; but it is impersonal in that the referent of the belief is not a person or the person's attributes.
To this point, we have treated trust as a situational construct. But it can also be viewed (Table 9) as a cross-situational, cross-personal construct (e.g., Harnett & Cummings, 1980; Wrightsman, 1991), which we call Dispositional Trust. This construct recognizes that people develop, over the course of their lives (Erikson, 1968), generalized expectations (Rotter, 1967, 1971) about the trustworthiness of other people. By our definition, a person has Dispositional Trust to the extent that s/he has a consistent tendency to trust across a broad spectrum of situations and persons.
There are two types of reasoning that underlie Dispositional Trust. With reasoning Type A, one assumes that others are generally trustworthy people--hence, one should almost always trust others. With reasoning Type B, one assumes that irrespective of whether people are good or bad, one will obtain better outcomes by trusting them--hence, one should trust them. In either case, Dispositional Trust is directed toward people (i.e., that others generally should be trusted). However, Type A supports Trusting Beliefs (Kramer, 1994), while Type B supports Trusting Intention directly (Figure 1). That is, if one believes that others are generally trustworthy (Type A), this supports Trusting Beliefs (which in turn support Trusting Intention). But if one solely believes that better outcomes are achieved through trusting others (Type B), Trusting Beliefs are not affected. However, a Type B Dispositional Trust does encourage one to be willing to depend on specific others; so Type B supports Trusting Intention directly. Table 8 places Dispositional Trust at the intersection of Dispositional and Other person (�19).
Situational Decision to Trust means the extent to which one intends to depend on a non-specific other party in a given situation. It means that one has formed an intention to trust every time a particular situation arises, irrespective of one's beliefs about the attributes of the other party in the situation (Riker, 1971). Like Dispositional Trust, it means that one has decided to trust without regard to the specific persons involved (Table 9)--because the benefits of trusting in this situation outweigh the possible negative outcomes of trusting. Kee & Knox suggested that this may occur "when there is much to gain from trusting..., but little attendant risk." (1970: 360). Situational Decision to Trust differs from Dispositional Trust in that it is an intentional construct and relates only to specific situations (Table 9), not across situations generally. It differs from System Trust in that it does not imply structural safeguards. It is simply an individual, situational strategy. Because it does not concern specific individuals, Situational Decision to Trust does not support Trusting Beliefs about specific individuals. But because it encourages a willingness to depend on others in the situation, Situational Decision to Trust supports Trusting Intention directly. Because of this construct's impersonal nature, Table 8 shows Situational Decision to Trust at the intersection of Intention and Something (�20).
Given the complexity of the trust concept and the wide range of existing scientific definitions (Tables 1 and 2), we consider our set of constructs to be parsimonious. Parsimony is important because it makes conducting empirical research more practical (Mayer, Davis & Schoorman, 1995). However, parsimony should be viewed as a relative term. The more complex a concept is, the less parsimonious its dimensions may appear. We join Hirschman (1984) in arguing that one must loosen the strictest bounds of parsimony in order to increase conceptual understanding and relevance of trust constructs. These six constructs extend across the impersonal, dispositional, and interpersonal dimensions of Table 8. They cover both cognitions, behaviors, and related emotions.
Behaviors are important to include because both scientists (Table 1) and practitioners (Tables 4 and 6) have frequently viewed trust as a behavior. For scientific use, cognitions and emotions are important. The preponderant use of both cognitions (beliefs, expectancies) and emotions (felt security, confidence) in past research (Table 1) provides support for including both in our set of trust constructs. We depict Trusting Beliefs as beliefs instead of expectations because: (a) beliefs are more often found in the research literature, and (b) beliefs relate more closely than expectations to such construct-cohesive models as that of Fishbein and Ajzen (1975). Beliefs are too distal from behaviors to always predict well (e.g., Fishbein & Ajzen, 1975). Hence, an intentional construct was selected. Being willing to depend was selected as the intention because it is grounded in scientific use (e.g., Dobing, 1993; Scanzoni, 1979) and because it is related to the concept of dependence, which is widely seen as a behavioral form of trust (e.g., Riker, 1971). We excluded attitudes because of their infrequent use in the trust literature (Table 1) and because they have not proven to be as highly predictive as beliefs and intentions in some studies (e.g., Davis, 1989). The use of dual (cognitive/emotional) constructs like our Trusting Intention and Trusting Beliefs is well-supported in research practice (e.g., Cummings & Bromiley, 1996; Rempel, Holmes & Zanna, 1985). Our combination of constructs that reflect institutional phenomena (System Trust) and molar behavior (Trusting Behavior) facilitate middle and higher levels of analysis views of trust-related phenomena. However, the cognitive/emotional (Trusting Beliefs) and personality (Dispositional Trust) constructs provide person-specific detail explanatory mechanism level of analysis. Prediction is enabled by two intentional constructs (Trusting Intention and Situational Decision to Trust), which together will relate strongly to Trusting Behavior. In total, the constructs have a significant synergistic effect. This result would probably not have occurred if this typology had not been created using an interdisciplinary approach. The interdisciplinary nature of the constructs is apparent: System Trust comes from sociology, Situational Decision to Trust comes from economics, Dispositional Trust from psychology, while Trusting Beliefs and Trusting Intention, and Trusting Behavior reflect research in several disciplines.
The four beliefs we chose cover most of the beliefs found in the trust literature. Overall, these six trust constructs provide adequate coverage. They tie both to common usage and the research literature. Though geared to a psychological framework, they extend across disciplines into both personality, interpersonal, and structural concerns. They are manageable for scientific purposes. Further, these constructs fit together (Figure 1), meeting one of Tiryakian's (1968) criteria for a good type (b) typology. As a quick 'reality test' of these constructs' usefulness in actual work relationships, we checked them against John Gabarro's (1978, 1987) work on trust development between new division or company presidents and their subordinates. Gabarro used qualitative techniques that in order to be careful not to 'force' theory on his respondents. Hence, his results are a test for what attributes were important to trust in practice. Gabarro (1987) lists integrity as the most important (compare our honesty), then motives (similar to our benevolence), and then consistency of behavior (predictability) and competency (same term as ours). Note that he also mentioned openness and discretion. But the first four he emphasizes were the same ones we cover. This provides some assurance that our four trusting beliefs are useful in actual work relationships. Perhaps most important to researchers, these constructs are amenable to empirical work. They are defined in sufficient detail to enable researchers to develop instruments that can correspond closely with these definitions, in order to demonstrate construct validity. More importantly, these constructs can be distinguished from each other and from closely related constructs (e.g., control), such that research results on trust can be compared, analyzed, and cumulated. The cognitive and behavioral portions of the constructs can be measured through level distinguishing (e.g., Likert) scales. The confidence and security portions of the constructs can be measured through secondary scales that tap one's feelings about the cognitive scale response just made. For example, the respondent would be asked how they secure they felt about their response to the Trusting Intention question they just answered.
Lewis & Weigert (1985a) called trust a highly complex and multi-dimensional phenomenon. Our classification typology helps us understand the nature of existing trust meanings. Our set of six trust definitions helps address conceptual confusion by representing trust as a broad, but coherent set of constructs. One benefit of this type of broad depiction of trust is that it has heuristic value (Kaplan, 1964) by being generative of research possibilities. Another benefit is that it presents a set of specifically defined trust types that enables practitioners and scholars to know what each means when they discuss this important topic. Implications for Research
This paper integrates and synthesizes the research literature on trust. Several researchers have suggested that the trust concept needs considerable integrative effort (Barber, 1983; Kramer, 1994; Lewis & Weigert, 1985a; Sitkin & Roth, 1993). These two typologies ([a]--classification system; and [b] set of related types) make progress toward achieving Tiryakian's target results for a typology: "[it] creates order out of the potential chaos..." (1968: 178) This paper provides a way to make the results of trust research more cumulative. The type (a) typology can be used to classify past trust research efforts in order to analyze evidence regarding trust. For example, empirical studies about trust's antecedents can now be classified and cumulated by the dimensions of Table 4 (trust construct type and referent). The trust type (b) typology can guide future research, making research results more comparable between researchers (even in different disciplines).
To build on this foundation for cumulative trust research, future operationalizations should correspond as closely as possible to these conceptualizations. Reliable and valid instruments should be developed. These should be tested against what people mean when they use the term trust in everyday life. The interrelationships between the Figure 1 constructs should be tested empirically in a number of situational contexts, using various methods. As Kee & Knox (1970: 365) recommended, research should measure several types of trust concurrently to gain greater explanatory power.
Sutton & Staw (1995) commented that editors need to more readily accept articles that present a broad theory but only test a portion of that theory. Since trust is not unidimensional, we recommend that this advice be applied in the trust research domain. In this way, empirical evidence on trust will more rapidly and reliably accumulate.
First, the size of the boxes in Figure 1 is meant to imply that Trusting Beliefs is the most important (but not the sole) determinant of Trusting Intention, and therefore Trusting Behavior. If a boss's subordinates, for example, do not believe that the boss is predictably competent, honest, and benevolent, then the chances are low that the subordinate will be willing to depend on the boss. Managers need to first try to be the type of person that others feel is trustworthy. But since Trusting Intention is based primarily on perceptions (beliefs) of trustworthiness, considerable attention needs to be paid to presenting oneself in ways that are consistent with trusting beliefs. Perception management alone, however, is not likely to work over time, because experience with a superior gives the subordinates a history of how well one's actions match one's self-presentation. Hypocrisy is not hard to detect over time.
Second, practitioners should note the affective component of Trusting Beliefs and Trusting Intention. The affective component (confidence, security) raises the issue of a set of subtle feelings about the trusted person. Feelings are hard to separate from each other, so one who likes the boss is more likely to also feel secure in the belief that the boss is benevolent towards her/him. Hence, bosses should devote time to develop an appropriate measure of positive personal relationships with their subordinates, so that subordinates can feel comfortable, confident and secure with them (Gabarro, 1987).
Third, one can develop a good relationship with another person by gradually increasing trusting behavior, while at the same time decreasing control measures, directed at the other person. Decreasing controls includes less monitoring of the other person. It also includes moving from a formal relationship to a more personal, informal relationship. This indicates to the other person that they are 'okay,' that they can be trusted, and therefore can have positive effects on the trusted person's self-esteem. In fact, trust is often a motivator. Control measures tend to express to the subordinate that they are not okay and are not trusted. Being controlled is very demotivating because it feels demeaning (Kohn, 1993).
Fourth, this model points out the relationship between trust and power. Superiors have positional power. But they should recognize (and generally should not discourage) that their employees have power over the boss by the boss's extending them trust--by the boss's depending on them. This levels out what is often an asymmetrical power situation between employees and bosses, in which the boss dominates. In a dominating situation, the employee will feel insecure around the boss. Therefore, the employee will withhold or distort information given to the boss. They will also accept less influence from the boss. This will typically lead to the boss's use of control mechanisms, leading to a downward spiral of distrust (e.g., Zand, 1972).
Fifth, the boss needs to consider the employee's individual makeup. If the employee has dispositional distrust, s/he may interpret messages from the boss more cynically. Hence, the boss should take additional steps to overcome these tendencies by developing a personal relationship with these individuals and by communicating very clearly the positive intent behind directions.
Sixth, the superior can take advantage of System Trust in several ways. Management should make sure employees know about the effectiveness of company safety nets (such as grievance procedures). This will make employees feel more secure in the company's environment, leading them to be better able to trust others within the environment. Management should also be sure that roles are clearly defined in the organization, such that people feel comfortable in their roles. Also, major changes like restructuring should be handled carefully, and layoffs should be minimized in order to provide workers a feeling that the environment is safe for them.
Seventh, since trust is built or destroyed through iterative reciprocal interaction, the initial period of the relationship is crucial. To get off 'on the right foot,' managers should begin each new work relationship by choosing to 'trust until' (Situational Decision to Trust). In other words, managers should take small initial risks with their people to signal a desire to have a trusting relationship. This gives the relationship a chance to move forward on the trusting cycle instead backward on the distrusting cycle. At the same time, we caution that initial trust results should be monitored to assure that they are warranted in the future with larger risks.