Take My Word for It: How organizational vocabularies foster trust across boundaries

Abstract

We examine how the presence of interpersonal trust relations within organizations is associated with the tendency of participants to share organizational vocabularies – sets of words that participants adopt to make sense of their organizational environment and negotiate its intersubjective meaning. In uncertain organizational settings, shared vocabularies offer a linguistic foundation for managing ambiguity, enabling trust to form when information on trustworthiness is absent or incomplete. We specify and estimate statistical models that capture the effect of shared organizational vocabularies on the likelihood of observing interpersonal trust relations among participants within and across the formal boundaries around intra-organizational units. We derive and test hypotheses regarding the association between organizational vocabularies and interpersonal trust relations among members of the top management team in an international corporation containing five semi-independent subsidiary companies. Our findings suggest a greater propensity for expressing trust among managers who share a common organizational vocabulary. However, the segregating effect of organizational structure confines this tendency within the formal boundaries of organizational units. Notably, this confinement weakens when participants share words aligned with the vocabulary of organizational identity, indicating that trust relations extend beyond formal boundaries when rooted in core organizational values. The study underscores that developing a shared understanding is not sufficient to sustain trust. Trust within hierarchical organizations is more likely to thrive when articulated through core elements of organizational identity, history and purpose.

Introduction

At least since its discovery as a central element in processes of institutionalization (Zucker, 1986), trust has been a theme of recurrent interest in contemporary research on organizations and organizing (De Jong, Kroon, & Schilke, 2017; Dirks & Ferrin, 2001; Mayer, Davis, & Schoorman, 1995; Rousseau, Sitkin, Burt, & Camerer, 1998; Zucker & Schilke, 2020). The basic argument behind this interest is that understanding the mechanisms and associations underlying the creation, stabilization and erosion of trust has implications for how organizations may function effectively as social and economic institutions (Bachmann & Inkpen, 2011; Zucker & Schilke, 2020).

Trust becomes particularly salient in conditions of uncertainty, when actors must rely on others without full information about their intentions, competencies or likely actions (Luhmann, 1979; Weick, 1979). In such environments, language plays a central role in facilitating shared understanding and reducing ambiguity (March & Simon, 1958; Thompson, 1967).

Consistent empirical evidence has accumulated in support of the hypotheses that trust is associated with the creation of social capital (Coleman, 1988), reduces task conflict (Simons & Peterson, 2000), speeds up the formation of temporary work groups (Meyerson, Weick, & Kramer, 1996) and shapes perceptions of risk (Molm, Takahashi, & Peterson, 2000). One stream of research emphasizes trust as embedded in one or more social networks in which the trustor (the ‘sender’ of a trust relation) and the trustee (the ‘receiver’) are embedded (Buskens, 1998; Lusher, Robins, Pattison, & Lomi, 2012; McEvily, Zaheer, & Soda, 2021; Uzzi, 1999). One important message of this relational view is that trust may be conceptualized as something that is ‘given’ and ‘received’ – and hence implicates social relations. Schilke, Reimann, and Cook (2021) take this argument one step further by suggesting that clear identification of the roles of trustor and trustee should be accompanied by an equally clear delineation of the object of trust. This view creates a conceptual bridge between contemporary studies of trust in organizations, and classic theories of cognitive balance (Heider, 1946, 1958; Newcomb, 1961).

Our approach adds a unique perspective to this body of research by focusing on shared language – a medium through which organizational members navigate interpretive ambiguity. We suggest that organizational vocabularies help reduce communicative uncertainty, making them an overlooked yet powerful resource for sustaining interpersonal trust.

We build on this line of research by examining the relational nature of trust within organizations more completely through the new empirical possibilities afforded by recent progress in the analysis of multilevel networks (Lazega & Wang, 2023; Lomi, Robins, & Tranmer, 2016; Wang et al., 2016a; Zappa & Lomi, 2015). Specifically, our work extends prior research in three ways.

First, we argue that trust relations are associated with shared understanding that organizational participants express of their work environment through the words they use to describe it. Specifically, we show that trust relations are more likely to link participants adopting the same organizational vocabulary to describe the overall organizational environment, as they experience it (Loewenstein, Ocasio, & Jones, 2012). Although prior research has linked shared organizational vocabularies to a variety of desirable outcomes such as improved coordination, collaboration and performance (Bechky, 2003; Collins & Smith, 2006; Lix, Goldberg, Srivastava, & Valentine, 2022; Weber & Camerer, 2003), it has rarely considered how shared vocabularies and trust relations shape one another within organizations. In uncertain organizational contexts, language serves as a key coordination device to reduce ambiguity (Weick, 1979), absorb complexity (March & Simon, 1958) and establish common expectations (Meyerson et al., 1996). A shared vocabulary thus not only facilitates understanding, but also enables organizational members to project reliability and interpret trustworthiness when other cues are ambiguous or unavailable. As such, shared language supports both the cognitive and affective bases of trust by anchoring sensemaking in environments where intentions and outcomes are uncertain (Luhmann, 1979; Thompson, 1967).

Second, we argue that interpreting the organizational world using the same organizational vocabulary is not sufficient to counterbalance the segregating effect of organizational design and support the development of trust relations between individuals across intra-organizational boundaries. We suggest that organizational vocabularies not only need to be shared, but must also be related to relevant dimensions of the overall organizational identity, institutionalized history and stated purpose. With few exceptions (Loewenstein et al., 2012), prior research has given only short shrift to the multilevel network mechanisms that this claim implicates (Tasselli, Zappa, & Lomi, 2020).

Finally, though extant research has frequently conceptualized trust as the outcome of intersecting networks of social relations (Granovetter, 1985; Uzzi, 1997; Uzzi & Lancaster, 2003), to date there has been relatively little research acknowledging that trust develops beyond social networks in the context of a differentiated multilevel relations connecting the parties involved not only to each other, but also to a broader system of symbolic, semantic and material objects (Basov, Breiger, & Hellsten, 2020; Fulmer, 2018; Lumineau & Schilke, 2018; Redhead, Maliti, Andrews, & Borgerhoff Mulder, 2023). We find that positions in these socio-material networks connecting organizational participants are associated with the likelihood of developing trust relations across intra-organizational boundaries.

Building on a major line of research on the relational basis of collaboration within organizations (Borgatti & Foster, 2003; Burt, Opper, & Holm, 2022; Lusher et al., 2012; Tasselli & Kilduff, 2018; Van de Bunt, Wittek, & de Klepper, 2005), we conceptualize trust broadly as the confidence and reliance that organizational participants place in one another when exchanging task-related advice. This definition encourages participants to bring their own understanding to bear as they nominate colleagues they trust for advice on work-related matters (Cross, Borgatti, & Parker, 2001). By allowing participants to define trust through their nominations, we capture their understanding of trust within the specific context of their organization. Trust, therefore, is understood as a relational social phenomenon rooted in shared experiences and interpretations of the organizational environment, particularly as expressed through shared organizational vocabularies (Nahapiet & Ghoshal, 1998; Tsai & Ghoshal, 1998). This participant-driven understanding of trust aligns with traditional definitions that emphasize trust as a psychological state comprising the intention to accept vulnerability based on positive expectations of another’s intentions or behaviour (Mayer et al., 1995; Rousseau et al., 1998). We highlight the role of shared organizational vocabularies in sustaining interpersonal trust relations within organizations. The approach we develop in this paper recognizes the presence of context-specific or situated dimensions of trustworthiness (Fulmer & Gelfand, 2012) affecting the distribution of trust relations across intra-organizational boundaries:

In this study, we clarify the multilevel mechanisms that concatenate to produce and reproduce relations of trust in organizations. The focus of our argument is on the conditions that allow these mechanisms to sustain trust across the boundaries of organizational units. The empirical context of the study, members of the top management team in an international industrial group containing five semi-independent production companies, allows us to document the relation between the vocabulary of organizational identity and trust relations across intra-organizational boundaries.

Theoretical background and hypotheses

Background

Organizational design – the set of solutions to the problem of interdependence of tasks, agents, decisions and activities within organizations (Puranam, Raveendran, & Knudsen, 2012; Thompson, 1967) – sets in motion conflicting social processes of segregation and blending. Segregation involves the creation and maintenance of formal boundaries around tasks, activities (sequences of tasks), people, teams and sites. Segregating processes are an organizational consequence of the intra-organizational division of labour (Puranam, Alexy, & Reitzig, 2014). Blending involves the erosion of the intra-organizational boundaries created by design. ¹ Blending processes are activated by the need to exchange information, transfer knowledge and coordinate activities across boundaries within organizations (Argote, Ingram, Levine, & Moreland, 2000).

Trust is essential to both segregation and blending processes. For example, trust is needed to establish productive relations within teams – a prime example of a well-defined subset of organizational participants with clearly identifiable boundaries (Costa, Fulmer, & Anderson, 2018). More generally, the establishment of formal boundaries creates shared social foci (Feld, 1981), and enhanced opportunities for interpersonal interaction and exchange. Shared social foci facilitate the development of social relations (Lomi, Lusher, Pattison, & Robins, 2014) and establish the conditions for the development of trust (Williams, 2001).

Classic research on intergroup behaviour instructs us that social boundaries placed around interacting individuals tend to induce a powerful sense of membership and collective identity (Tajfel & Turner, 1979). Under these conditions, trust among participants sharing membership in clearly bounded organizational sub-units is essential to establishing a system of reliable mutual expectations, ensuring that routine activities will run smoothly, interdependencies are managed efficiently and daily interactions will sustain productivity (De Jong & Elfring, 2010). Consistent with this view, organizational participants are more likely to express trust toward other members of their unit (Foddy, Platow, & Yamagishi, 2009). This trust will manifest itself in attitudes and behaviours that are highly idiosyncratic and context-specific, and hence difficult to explain to non-socialized outsiders. This is one of the most enduring insights in organizational behaviour research (Roethlisberger & Dickson, 1939: Part IV, sections XX–XXI; Roy, 1959). For this reason, we will regard this as a default expectation undergirding the hypotheses tested in the empirical part of the study.

In general, segregating processes triggered by boundary formation tend to induce resonant social identities; that is, identities that are not only distinct, but also ‘capture or activate powerful distinctions’ (Baron, 2004, p. 11). According to this view, boundaries within organizations not only institutionalize differences between participants, but strengthen, activate, maintain – and occasionally amplify – those differences.

Trust is critical when the need of organizational participants to collaborate and exchange information or resources is accompanied by limited possibilities of monitoring or verifying the actions of others working in separate units (De Jong et al., 2017). Trust is essential also when exchange takes place in the context of open structures where the reputation and insurance effects of network closure are typically unreliable (Burt, 2005; Coleman, 1988). Social separation and lack of network closure are the typical conditions that define situations in which organizational participants must rely on others located in different – and possibly distant – units separated by formal organizational boundaries. Under conditions of segregation, trust across units is typically supported by reciprocity (Molm et al., 2000), homophily (Ertug, Brennecke, Kovács, & Zou, 2022), repeated exchange (Kirchler, Fehr, & Evans, 1996) and comparable mechanisms that may be relied upon to provide partial insurance against the risks related to potential opportunism (Williamson, 1993). Trust relations are rarely observed across intra-organizational boundaries in the absence of any form of social insurance against the risk of making oneself vulnerable to strangers (Mayer et al., 1995).

To summarize our argument: some level of trust may be needed for organizational participants to work together productively in segregated organizational units. Some level of trust may also be needed to support the blending processes required for coordination across units. This view of trust as fuelling opposite processes of segregation and blending within organizations involves a paradox: the factors that are likely to strengthen trust within the boundaries of organizational units (and hence make segregating processes more efficient) are precisely the same factors that make trust important (to make blending processes more effective), but also particularly difficult to establish across those boundaries. When is trust more likely to be granted to information and advice coming from outside sources? More generally, we ask: when is ‘outside’ advice is more likely to be accepted and ‘get in’ (Gibson, 2008)?

Hypotheses

Given the purposeful segregating effects produced by organizational design, we focus on the common factors that can be relied upon to foster, encourage and sustain trust relations between participants across intra-organizational boundaries. We frame the answer to our orienting question in terms of the results produced by organizational research that has emphasized the role of superordinate identities in making intra-organizational boundaries more permeable to blending processes, while preventing erosion of individual unit-specific identities (Argote, 2024; Gaertner et al., 2000; Kane, 2010). This framing gives prominence to the multilevel hierarchical nature of organizations as distinct forms of collective action (Simon, 1962). Specifically, we emphasize language as the main vessel through which superordinate identities are expressed (Van Swol & Kane, 2019), and practically shared with all organizational participants. Language is also the means through which participants are reminded of their belonging to a broader and more inclusive organizational community (Argote & Kane, 2009).

Ample empirical evidence is available to support the theoretical prediction that awareness and recognition of (and identification with) a shared superordinate social identity encourages blending processes by providing a common target for identification and reducing in-group favouritism and out-group bias (Brewer, 2000; Gaertner & Dovidio, 2000; Hornsey & Hogg, 2000). The evidence produced by research in social psychology has encouraged attempts to replicate these results ‘in the wild’; for example, in real and simulated organizational settings (Dokko, Kane, & Tortoriello, 2014; Hinds & Mortensen, 2005; Kane, 2010; Lomi et al., 2014). The results of this research are convergent: recognition of and identification with a superordinate organizational identity may enhance the capacity of organization participants to understand and appreciate the knowledge, competences and contributions of others located in different sub-units separated by organizational boundaries (Kane, 2010). What the present study adds to this result is a more detailed account of the specific mechanisms that may be responsible for converting an individual sense of belonging in a superordinate organizational identity to expressions of trust within and across intra-organizational boundaries.

The segregating and blending processes that we have discussed contribute to making trust directly relevant for a wide range of desirable organizational outcomes (Kramer, Hanna, Su, & Wei, 2001). They are clearly detectable in the organizational vocabulary adopted by organizational participants to describe and classify their work experience. This is the case because the segregating process triggered by organizational design not only produces units structured around specific sub-goals (March & Simon, 1958), but also induces ‘zones of meaning that are linguistically circumscribed’ as originally proposed by Berger and Luckmann (1966, p. 56), according to whom

the sum of linguistic objectifications pertaining to my occupation constitutes another semantic field, which meaningfully orders all the routine events I encounter in my daily work. Within the semantic fields thus built up, it is possible for both biographical and historical experience to be objectified, retained, and accumulated.

One possible implication of this constructivist view is that developing a relation of trust may be easier and more natural among participants who are able to rely on a common organizational vocabulary to make sense of, verbalize and share their experience of membership in a superordinate organizational entity that extends beyond the physical and sensory boundaries of their local work environment.

An organizational vocabulary may be defined as a ‘socially constructed set of linguistic categories used by social groups, both to make sense of organizational reality and to bring attention to available sources of organizing activities’ (Ocasio & Joseph, 2005, p. 165). It is through shared vocabularies that organizational members may be able to develop shared meanings that define their workplace as they understand it (Bechky, 2003; Loewenstein et al., 2012). The words used to describe their organization may influence the ability of organizational members to think about the organization, and hence to communicate with other participants (Gartner, 1993, p. 231). One implication of this argument is that the relation of trust needed to negotiate the intersubjective meaning of individual experiences may be difficult to develop in the absence of a shared organizational vocabulary. This view is summarized in our first hypothesis:

• Hypothesis 1 (H1): A relation of trust is more likely to be observed between participants expressing a similar understanding of their superordinate organizational environment.

As modular social systems with an explicit hierarchical structure (Simon, 1962), organizations are internally differentiated into multiple semantic fields (Berger & Luckmann, 1966). The boundaries of organizational sub-units (divisions, functions, subsidiary companies and the like) are typically designed around specific sub-goals, functions, objectives or the boundaries of professional families (Brock, 2006; March & Simon, 1958; Rhoades, 2023).

We predict that the segregating effects of intra-organizational boundaries will be stronger than the blending effects associated with sharing a vocabulary to describe the personal experience of being a member of a superordinate organizational entity. This happens also because the reduced opportunities of interaction with members of different units will be likely to prevent, or greatly limit, the contact needed for negotiating and ultimately sharing the meaning of being a member of a superordinate organizational entity:

• Hypothesis 2 (H2): A relation of trust is less likely to be observed between participants in different organizational sub-units despite a similar understanding of their superordinate organizational environment.

Hypothesis 2 postulates that the segregating effects of formal boundaries defined around organizational units will be stronger than the blending effects potentially determined by similarities in the organizational vocabularies adopted by participants. Yet rarely can organizations be represented accurately as disconnected ‘islands of trust’ within units floating in a ‘sea of distrust’ between units. More commonly, at least some relation of trust will crosscut the boundaries of organizational units. When is this permeation process more likely to be observed (Gibson, 2008)? The theoretical argument that we have developed emphasizes the role played by formal superordinate organizational identities. But how are superordinate identities communicated to organizational participants? What is the material basis that supports the individuals’ interpretation of the superordinate’s identity? These questions are important in the development of our theoretical argument because symbolic elements in organizations cannot be interpreted without reference to their material basis (Mohr & White, 2008).

In the typical case, organizations summarize and formalize their identity claim in official documents for the purpose of communicating such claim to their internal and external constituencies. For example, the mission statement emphasizes the uniqueness of the organizational identity and its distinctive elements (Alegre, Berbegal-Mirabent, Guerrero, & Mas-Machuca, 2018). Similarly, public documents containing narratives about the company history, values and purpose provide the material basis that sustains the participants’ attempts to make sense of what it means to derive aspects of their own identities from their membership in the superordinate organizational entity.

The vocabularies that participants share to make sense of their experience of belonging to a superordinate organizational entity (Ocasio & Joseph, 2005) may or may not overlap with the vocabulary adopted by the organizational identity claim to communicate its history, values and purpose to its constituencies. When it does, however, overlap will signal some degree of alignment of individuals’ understanding of the superordinate organizational environment with concepts recognized, institutionalized and officially adopted by the organization. When the vocabulary of personal experience does not overlap with the vocabulary of organizational identity, participants may not converge on a consistent interpretation of the object of trust (Schilke et al., 2021). In consequence, their relationship may be unbalanced, and they will be less likely to develop a personal connection based on trust (Heider, 1946). This argument is summarized in our last hypothesis:

• Hypothesis 3 (H3): A relation of trust is more likely to be observed between participants working in different organizational sub-units when their understanding of the organizational environment is expressed in terms of core elements of the superordinate organizational identity, history, and purpose.

These hypotheses can be represented by sub-network configurations of a multilevel network that consists of: (a) organization participants and their interpersonal trust ties as a micro-level network; (b) the organization vocabularies that participants nominated to describe their experience working under the superordinate organizational entity – this forms a bipartite meso-level participants-by-words network; and (c) a macro-level network defined among the organization vocabularies based on co-appearances in the organizational identity claim about its history, values and purpose. It is important to clarify that the term ‘level’ in this context pertains specifically to distinguishing participants from vocabularies, rather than referring to other organizational hierarchies. Additionally, the formation of the macro-level word network, established from organizational identity claims, operates independently from the creation of the participant-vocabulary meso-level network, which is derived from surveys. These hypotheses address distinct aspects of trust in organizations. Hypothesis 1 examines how sharing a similar understanding of the organizational environment represented by nominating the same organization vocabularies may promote trust among individuals, focusing on the alignment of perspectives. In contrast, Hypothesis 2 focuses on the segregating effect of organizational unit boundaries on interpersonal trust. It highlights the effect of structural factors on interpersonal trust across unit boundaries. Hypothesis 3 adds a new layer of complexity to the analysis by considering the inter-words network within the organizational vocabulary. It explores how the dependence among words, generated by their co-occurrence in the organizational identity statement, can impact trust. Unlike Hypothesis 2, which focuses on the individual words in isolation, Hypothesis 3 emphasizes the connections among these words and their collective impact on trust. This hypothesis underscores the importance of linguistic alignment with the organizational identity claim, suggesting that trust is bolstered when individuals’ understanding of the superordinate organizational environment reflects the broader organizational identity.

This multilevel network conceptualization enables us to assess the relationship between organizational vocabulary network and interpersonal trust relations using exponential random graph models (ERGMs) for multilevel networks (Parker, Lomi, & Zappa, 2024; Wang, Robins, & Matous, 2016a; Wang et al., 2016c; Wang, Robins, Pattison, & Lazega, 2013; Zappa & Lomi, 2015) described in greater detail in the following section.Material and methods

Setting and data

We examine the effect of organizational vocabularies on how participants develop trust within and across formal intra-organizational boundaries in the context of data we have collected on members of the top management team of a multi-unit industrial group. The corporate group – the common legal and financial term typically used to identify a set of subsidiary companies under the same ownership (Squire, 2011; Witting, 2018) – contains five semi-independent subsidiary companies involved in the design, manufacturing and sale of high-end products in the international market for leisure motor yachts. We refer to this as ‘the group’ or simply ‘corporate’, and to its companies as ‘subsidiaries’ or ‘sub-units’.

The central subsidiary also function as the corporate headquarter. The four other subsidiaries operate semi-independently, manufacturing products with distinct brand identities. More importantly for our purposes, they have non-overlapping management teams. Additional information on the organization is available in Lomi et al. (2014) and Lusher et al. (2012).

As part of our fieldwork, we asked the corporate president and chief executive officer (CEO) to identify organizational members they would consider the key decision makers across all subsidiaries. The final list comprised 47 individuals, including corporate executives (president, CEO, chief operating officer (COO), chief financial officer (CFO) and human resources (HR) director), functional heads (engineering, design, marketing and sales) and five external consultants involved in new product development.

Following best recommended practice in organizational social network research (Borgatti, Everett, & Johnson, 2018; Borgatti & Molina, 2005; Robins, 2015) and on collection of social network data (Kilduff & Krackhardt, 2008), we used a roster-style network questionnaire to collect information on task-related trust relations. Respondents were asked to indicate the names of others whose advice on work-related matters they would trust.

The outcome of this questionnaire was a micro-level interpersonal trust network represented by a binary adjacency matrix. The response rate was 100%. On average, each participant expressed trust in 10 others. The density of the trust network was approximately 23%, meaning that about one-fifth of all trust ties were present.

Our research design implements an emic approach to studying trust in organizations, allowing respondents to define ‘trust‘ in the context of their own advice-seeking experience (Morris, Leung, Ames, & Lickel, 1999). We do not measure trust directly, or aim to validate specific constructs, as an etic approach would require. We have, however, explored the extent to which expressed trust network overlaps with other common network contents (Burt & Schøtt, 1985; Marsden, 1990). ² Although establishing an analytical differentiation between trust and related concepts is important in an etic approach to social networks, it is not our focus in this study. Both emic and etic approaches have been used in social network research and are complementary to some extent (Laumann et al., 1983; Marsden, 1990). We do not impose any analytical reconstructions of ‘trust‘ on participants. Instead, our analysis is based on data derived from their contextual understanding as shaped by their personal work experience - a research design strategy clearly inspired by emic, rather than etic principles (Headland, Pike, & Harris, 1990; Morris et al., 1999; Pike, 1966, 1967).

Clearly, an emic approach avoids – but is not in itself sufficient to resolve – problems related to a more general and abstract definition of trust. ³ For this reason, during the last 40 years or so, research on the analysis of social networks has evolved questionnaire design and administration practices that integrate emic and etic perspective in the practice of data collection (Bernard, Killworth, Kronenfeld, & Sailer, 1984; Borgatti, Mehra, Brass, & Labianca, 2009; Freeman, 1992; Marsden, 2005, 2011; Prell, 2012, chapters 3 and 4; Ross & Redhead, 2021). Best recommended practice suggests that when relational contents are reconstructed in terms of abstract cognitive categories (e.g. friendship, trust or advice), contextually meaningful examples of what these generic categories could mean and entail for respondents inhabiting the specific social or cultural space being studied. This practice involves significant fieldwork to build a reliable understanding of participants’ situated experience, understanding and expectations.

In the collection of social network data that we examine in the empirical part of the study, we followed closely these best recommended practices. Systematic observation allowed us to identify practical examples of ‘trusted advice’ relations among the managers included in the sample (Lusher et al., 2012). Examples of topics on which members of the top management team claimed to seek and exchange advice included the solvency of potential clients, logistic and transportation problems, flexibility with terms of payment, and communication of cost overruns and delivery delays to clients. The questionnaire specifically associated examples of these issues to exemplify and situate ‘trusted advice’ in the daily organizational life of respondents. Although we think that relying on a situated understanding of trust is crucial in organizational settings, we are aware of the risks and limitations entailed in an excessively idiosyncratic and context-specific definition of trust. In our specific case, however, the existence of advice and knowledge transfer and exchange relations in which trust is embedded is a general feature of organizations (Cross et al., 2001; Gibbons, 2004; Phelps, Heidl, & Wadhwa, 2012). This makes our approach easily transposable to other contexts.

To reconstruct the organizational vocabularies, we focused on the association between managers and keywords they would adopt to describe their experience within a larger corporate entity (Burke, 1989; Hirsch, 1986; Mills, 1940; Mohr, 1994; Suddaby & Greenwood, 2005). Figure 1 shows the network linking managers (red nodes), words (blue nodes) and interpersonal trust relations (grey edges).

Figure 1.

Association between people (red circle nodes) and words (blue square nodes). Meso-level people-by-words network ties are in light grey. Micro-level interpersonal trust and macro-level ties among words are in darker grey.

We began by interviewing subsidiary CEOs to identify commonly used words. Examples included ‘technology’, ‘luxury’, ‘team’ and ‘chaos.’ We reviewed the resulting list with the president and CEO to finalize a consistent vocabulary. Presenting participants with a pre-selected word list facilitated interpersonal comparison. We note that no participants understood the words as irrelevant, inappropriate or surprising. Each respondent indicated, on average, four words. Each of the 30 words was selected by about 7 participants.

Participants rated each word on a 1–5 Likert scale according to how well it captured their experience of the corporate group. These ratings formed a two-mode ‘participant-by-words’ network.

For analysis, we dichotomized the matrix, treating scores of 5 as strong endorsements $(X_{ik}=1).$ We chose this cut-off value because we wanted to emphasize the words that participants considered more resonant and distinctive, and about which they expressed unambiguous sentiments. Each respondent nominated on average five words. Table 1 presents descriptive statistics for the trust and word-nomination networks.

Table 1.

Descriptive network statistics.

Trusted advice network		Word nomination network
Statistics	Value	Statistics	Value
Density	0.23	Density	0.15
Mean degree	10.55	Mean degree (person)	4.51
SD (in-degree)	7.01	SD (person)	5.54
Skewness (in-degree)	1.09	Skewness (person)	1.79
SD (out-degree)	8.39	Mean degree (word)	7.07
Skewness (out-degree)	1.70	SD (word)	5.15
Global clustering	0.47	Skewness (word)	1.10
		Global clustering	0.45

Note: SD: standard deviation. SD and skewness are calculated based on the corresponding degree distributions.

The final step involved building the network among the words based on the official corporate identity document summarizing the corporate and the social history of the group since its very beginning as a small family firm in 1968, commercializing motorboats for an American producer. The founders of the company built their first motorboat (a ‘motor-sailer’) in 1971. The text explains how the company came into being, and how the core guiding values and visions of its founders drove its growth and development through a series of acquisitions of yachting companies owning renowned brands in the luxury segment of the global market for motor yachts. The document, available online, was about 2000 words in length. We identified that 12 of the 30 words in our list were contained in the corporate identity claim. We considered this subset of words as directly connected to construct the macro-level network of words forming the last component of our multilevel network. The effectiveness and credibility of alternative strategies for reconstructing the relations between words is likely to depend on many factors related to the length of the corporate identity claim, its internal organization and its purpose. Given the document’s brevity and coherence, we treated it as a single unit for identifying co-occurrences.

Empirical model specification estimation and evaluation

ERGMs are statistical models for social networks, aiming at identifying local network processes driving network tie formation and unveiling key structural signatures represented by local network configurations (Lusher, Koskinen, & Robins, 2013). ERGMs are a recognized approach for analysing cross-sectional networks within and between organizations (Amati, Lomi, & Mira, 2018). The following section provides an overview of multilevel ERGM constructs and specifications implemented in the empirical part of the study. ⁴

ERGMs treat network ties as outcome variables, and explicitly account for their interdependence. Given a network with $n$ nodes, let $X_{ij}$ denote a network tie between node $i$ and node $j$ , such that $X_{ij}=1$ if a tie exists, 0 otherwise. The network as a random variable $X$ can be seen as a collection of $n\left(n-1\right)$ network tie variables $.$ Let $Y$ denote possible covariates, such as age or gender of individual nodes $(Y_i)$ , or dyadic attributes $(Y_{ij})$ representing the strength of network ties or other networks such as the ‘boss of’ network where $Y_{ij}=1$ if $i$ is a manager of $j$ .

ERGMs assign probabilities to networks using the exponential family, while treating other variables as covariates:

$$\mathrm{Pr}\left(X=x|Y=y\right)=\frac{1}{\kappa }exp{\displaystyle \sum }_Q{\theta }_Q{z}_Q(X,Y)$$

where:

• $Q$ represents various local network configurations inside which all network tie variables and covariates are considered interdependent. Using a network triangle as an example configuration representing network closure, the three network tie variables involved in a closed triad are considered interdependent.

• $z_Q\left(X,Y\right)$ is a graph statistic that counts the number of configurations of type $Q$ .

• ${\theta }_Q$ is the parameter associated with $z_Q$ , governing the prominence of the corresponding configuration and the underlying social process. Estimated parameter values represent conditional log odds on the likelihood of observing the corresponding configuration conditioning on the rest of the model. $\kappa$ is a normalizing constant over all possible networks for a given size.

The model estimates are based on MPNet software (Wang et al., 2016b) which implements MCMC algorithm (Snijders, 2002). The statistical significance of the estimated parameters is indicated by two-sided test p-values based on the ratio of the effect size to its estimated standard error.

Variables and measures

Dependent variable

The dependent variable is the presence of a trust relation between members of the top management team, conditional on a series of incidental factors. Three of these factors are directly linked to the hypotheses of the study. The others enter the empirical model specification to control for the effects of endogenous mechanisms that are known to shape the structure of social networks, and exogenous node-specific differences, respectively. We describe these various components of the empirical model specification next.

Hypotheses as network effects

Hypotheses H1 to H3 are implemented as ERGM configurations detailed in Table 2. To test H1, we assess whether trust is more likely between participants who nominated the same organizational vocabulary words. We expect the corresponding effect to be positive; that is, we expect association with the same words to increase the probability of observing the presence of a trust relation between organizational participants. For H2, we include an interaction effect to test whether shared vocabulary still supports trust across organizational units. A significant negative interaction would indicate the segregating effect of unit boundaries. For H3, we test whether trust across units is more likely when shared words also appear in the organizational identity claim.

Table 2.

Exponential random graph model (ERGM) effects related to hypotheses.

Hypotheses	Graphical representation	Mathematical definition
Baseline (default) expectation: A relation of trust is more likely to be observed between participants working in the same organizational unit.		${\displaystyle \sum }_{i,j}{A}_{ij}I(Y_i,Y_j),I\left(Y_i,Y_j\right)=\{\begin{array}{c}1,if{Y}_i=Y_j\\ 0,otherwise\end{array}$
H1. A relation of trust is more likely to be observed between participants expressing a similar understanding of their superordinate corporate environment.		${\displaystyle \sum }_{i,j,k}{A}_{ij}{X}_{ik}{X}_{jk}$
H2. A relation of trust is less likely to be observed between participants in different organizational sub-units even when they express a similar understanding of their superordinate corporate environment.		${\displaystyle \sum }_{i,j}{A}_{ij}{X}_{ik}{X}_{jk}\left[I\left(Y_i,Y_j\right)-1\right]$
H3. A relation of trust is more likely to be observed between participants working in different organizational sub-units when their understanding of the corporate environment is expressed in terms of core elements of the superordinate corporate identity, history and purpose.		${\displaystyle \sum }_{i,j,k,l}{A}_{ij}{B}_{kl}{X}_{ik}{X}_{jl}\left[I\left(Y_i,Y_j\right)-1\right]$

Note: Red circles are organizational participants; blue square are words in the organizational vocabulary. Edges linking participants depict the existence of a trust relation, labelled by network tie variable ( $A_{ij}$ ). Edges linking words represent the joint presence of the words in the official corporate group’ identity claim, labelled by variable ( $B_{kl}$ ). Edges linking participants and words, labelled by ( $X_{ik}$ ), represent participant’s nominations of words associated with qualities of the general organizational environment. Grey squares record membership of participants in organizational units (subsidiary companies) represented by categorical covariate ( $Y_i$ ). $I\left(Y_i,Y_j\right)$ is an indicator function checking whether $i$ and $j$ belong to the same unit.

Endogenous network mechanisms

For the trust network, we use the Robins, Pattison, and Wang (2009) specification. Table 3 summarizes the endogenous and covariate network mechanisms following Wang et al. (2013). Each configuration is labelled by the social process they represent, followed by a label from the modelling software MPNet. Effects directly related to hypothesis testing are labelled by the hypothesis numbers in bold.

Table 3.

Exponential random graph model (ERGM) specification.

Trust
	(a) Density [Arc]		(b) Reciprocation [Reciprocity]	(c) Popularity [AinS]	(d) Expansiveness [AoutS]
	(e) Transitive two-paths [A2P-T]		(f) Transitive closure [AT-T]		(g) Popularity closure [AT-D]
	(h) Sender	(i) Receiver	(j) Homophily[Match]		(k) Heterophily[Mismatch]or [Difference]	(l) Dyadic
People by words
	(m) Two-mode density[XEdge]		(n) Expressiveness[XStar2A]	(o) Word popularity[XStar2B]	(p) Shared popularity[X4Cycle]
Cross level interactons
	(q) Expansiveness association[In2StarAX]		(r) Popularity association[Out2StarAX]	(s) Trust association[TXAX arc] (H1)	(t) Cross-unit trust association[TXAX mismatch arc] (H2)
	(u) Cross-level entrainment [C4AXB entrainment]			(v) Cross-unit cross-level entrainment [C4AXB mismatch entrainment] (H3)

Note: MPNet labels shown in square brackets. Effects related to the hypotheses labelled in bold font in round brackets.

The empirical model specification includes effects for network density (Table 3(a)), or the baseline propensity to form trust ties; reciprocation (Table 3(b)), or the tendency to form reciprocal trust between pairs of nodes; popularity (Table 3(c)) and expansiveness (Table 3(d)); they are effects capturing the tendency for trust relations to be centralized on a few highly popular or expressive individuals.

The transitive two-path configuration (Table 3(e)) represents a brokerage structure or the pre-condition for network closure. The transitive closure (Table 3(f)) closes the transitive two-path to form local network closure. It may also represent a form of hierarchical structure where an expressive individual (at the bottom of the hierarchy) trusts a few others who in turn trust the popular individual at the top of the hierarchy. The popularity closure (Table 3(g)), on the other hand, is a form of closure based on structure equivalence where popular individuals among a common group of others also trust one another.

For the meso-level people to words nomination network, we use Wang et al. (2009) specification, including the density effect (Table 3(m)) as the baseline propensity to nominate words; the expressiveness effect (Table 3(n)) captures the tendency for certain individuals who nominated more words than others, and the word popularity effect (Table 3(o)) captures potential popular words that attracted nominations by more people than other words. The shared popularity effect (Table 3(p)) tests whether individuals tend to nominate a group of common words representing shared understanding of the organizational environment.

For the cross-level interactions, we follow Wang et al. (2013) and Wang et al., (2016c), including effects for expansiveness association (Table 3(q)) to test whether people nominated more words are also likely to trust more others; the popularity association (Table 3(r)) effect tests whether nominating more words attracted more trust; the trust association configuration (Table 3(s)) tests whether people who nominated the same words are more likely to trust one another (H1); and the cross-unit trust association (Table 3(t)) tests the association between nominating the same words and trust between people from different units (H2); the cross-level entrainment effect (Table 3(u)) tests whether trust relation is associated with nominating related words that appeared in the corporate identity claim, and the cross-unit cross-level entrainment configuration (Table 3(v)) tests whether nominating related words are associated with trust relations between people from different units more specifically (H3). Other effects improve model fit by accounting for known dependencies in multilevel structures.

Exogenous covariates

Control covariates include nationality, job function, interval measures of their age group, education level, seniorities in their subsidiaries and in the overall corporate, and binary indicators for male managers and whether a person is a consultant. There are five main job functions in this corporate where marketing has the smallest number of managers (8.5%), whereas production has the largest number (25.5%). Most respondents are male (85.1%) and aged 31–45 (60%).

We follow Robins et al. (2001) for attribute interactions, which includes the sender (Table 3(h)) and receiver (Table 3(i)) configuration testing whether individuals with a particular attribute or having a higher attribute value tend to send or receive more trust nominations than expected. The homophily effect (Table 3(j)) tests whether individuals with the same or similar attribute values tend to trust one another, whereas the heterophily effect (Table 3(k)) tests the opposite. Using male covariate as an example, sender effects test whether male managers expressed more trust ties; the receiver effects assess whether males are more often trusted; homophily effects test whether males tend to trust other males.

For the interval measures, ERGMs use sender (Table 3(h)) and receiver (Table 3(i)) effects to test whether managers in a higher interval tend to send or receive more trust nominations; and heterophily (Table 3(k)) effect that calculates the difference between the intervals of the two manages in a dyad to test whether a greater difference in the interval measures is associated with the formation of trust relations.

The dyadic attribute effect (Table 3(l)), in our case the ‘Boss of’ network, tests whether trust is aligned with the formal hierarchy of the organization. For the categorical covariates, ERGMs can test effects of tendency for trust ties to occur within the same category (Table 3(j)), matching category) or across different categories (Table 3(k)), mismatching categories). Using unit membership as an example, a positive matching unit membership effect suggests trust relations are more likely to occur among managers within the same unit. These effects can also be interpreted as homophily (matching category) or heterophily (mismatching categories), respectively.

Results

Analysis

We estimated a sequence of four models reported in an increasing order of complexity of their empirical specification to reach a robust final model. The first model, serving as a baseline, included no network structural effects. Subsequent models added endogenous effects for both interpersonal trust and word nominations. The final model incorporates all hypothesized effects and cross-level interactions. The model development, comparisons and goodnesses-of-fit (GOF) are reported in the supplementary material along with a discussion of the importance of accounting for interdependence across network levels.

The final model has 38 parameters, of which 22 are statistically significant at the 5% level and 4 at the 10% level. Non-significant effects either serve as simpler components building towards more complex interactions or enhance the model’s GOF. We organize the results into three parts: (a) effects testing our hypotheses (Table 4(a)); (b) effects for interpersonal trust networks (Table 4(b)); and (c) effects for the word nomination network and its interaction with trust (Table 4(c)).

Table 4(a).

Effects extracted from the overall exponential random graph model (ERGM) related to the hypotheses.

Effects	MPNet labels	Graphical representation	para.	SE	p-value
Within unit homophily	Match		0.907	0.125	<.001*
(H1) Trust association	TXAX arc		0.063	0.035	.075+
(H2) Cross-unit trust association	TXAX mismatch arc		−0.085	0.048	.073+
(H3) Cross-unit cross-level entrainment	C4AXB mismatch entrainment		0.022	0.011	.044*

‘*’ indicates significance at the 5% level, and ‘+’ indicates significance at the 10% level.

Table 4(b).

Effects for interpersonal trust network extracted from the overall exponential random graph model (ERGM).

Effects	MPNet labels	para.	SE	p-value
Density	Arc	−3.391	1.371	.013*
Reciprocation	Reciprocity	0.872	0.212	<.001*
Popularity	AinS	−1.579	0.590	.007*
Expansiveness	AoutS	−0.811	0.310	.009*
Transitive closure	AT-T	1.506	0.214	<.001*
Popularity closure	AT-D	0.539	0.164	.001*
Transitive path	A2P-T	−0.018	0.016	.274
Sender	Male	0.477	0.678	.482
Receiver	Male	0.309	0.328	.346
Homophily	Male	−0.645	0.711	.364
Sender	Consultant	−0.766	0.271	.005*
Receiver	Consultant	1.206	0.385	.002*
Homophily	Consultant	−1.236	1.253	.324
Sender	Age group	0.449	0.095	<.001*
Receiver	Age group	−0.214	0.120	.074+
Heterophily	Age group	−0.253	0.082	.002*
Sender	Unit seniority	0.139	0.067	.038*
Receiver	Unit seniority	−0.068	0.092	.461
Heterophily	Unit seniority	−0.006	0.060	.923
Sender	Corporate seniority	−0.400	0.079	<.001*
Receiver	Corporate seniority	0.364	0.094	<.001*
Heterophily	Corporate seniority	0.060	0.073	.408
Sender	Education	−0.038	0.072	.597
Receiver	Education	0.101	0.087	.248
Heterophily	Education	−0.115	0.056	.040*
Homophily	Nationality	0.298	0.110	.007*
Homophily	Function	0.361	0.104	<.001*
Dyadic	Boss of	0.436	0.356	.220

‘*’ indicates significance at the 5% level, and ‘+’ indicates significance at the 10% level.

Table 4(c).

Effects for cross-level word nomination two-mode network, and associations between trust and word nominations extracted from the overall exponential random graph models (ERGM).

Effects	MPNet labels	para.	SE	p-value
Density	XEdge	−3.250	0.199	<.001*
Expressiveness	XStar2A	0.206	0.015	<.001*
Word popularity	XStar2B	0.085	0.020	<.001*
Shared word popularity	X4Cycle	−0.013	0.004	.004*
Popularity association	In2StarAX	−0.022	0.012	.072+
Expansiveness association	Out2StarAX	0.008	0.008	.303
Cross-level entrainment	C4AXB entrainment	−0.007	0.004	.070+

‘*’ indicates significance at the 5% level, and ‘+’ indicates significance at the 10% level.

Hypotheses

Table 4(a) shows the effects directly related to our hypotheses. Our default expectation – that trust relations tend to operate more strongly within intra-organizational boundaries – is confirmed. The estimates indicate a positive and significant (at the 5% level) within-unit homophily effect on trust. The odds of having a trust relation between managers in the same subsidiary unit relative to trust relation in general increase by a factor of $(e^{0.907}=2.48)$ with a binary probability of $(e^{0.907}/\left[1+e^{0.907}\right]=71.26\%)$ . In other words, trust relations are nearly 2.5 times more likely to be observed within units than across them.

For Hypothesis 1 (H1), the trust-association effect (TXAX arc = 0.063) is positive and significant at the 10% level, and indicates a positive association between common word nominations and interpersonal trust. From the trust relation’s perspective, sharing one extra word between a pair of managers is associated with a 7% increase in the odds of forming a trust tie between them $(e^{0.063}=1.07).$ Hypothesis 1 is supported: a relation of trust is more likely to be observed between participants using the same words to describe their understanding of the superordinate corporate organization.

The cross-unit trust association (TXAX mismatch arc = −0.085) effect is negative and significant at the 10% level. This result supports Hypothesis 2 (H2) that trust relations are less likely to be observed between participants in different organizational units even when they express similar understanding of their superordinate corporate environment. The organizational unit boundary is associated with a reduction in the odds of trust formation by a factor of $(e^{-0.085}=0.92).$ Managers separated by intra-organizational boundaries are approximately 8% less likely to have a relation of trust, despite adopting similar words to describe their experience as members of a superordinate corporate entity. Sharing the same understanding of the corporate environment does not necessarily positively associate with intra-organizational trust.

Hypothesis 3 (H3) involves the cross-level association between intra-organizational trust and related words that appeared in the corporate identity claim. The positive cross-unit cross-level entrainment (C4AXB mismatch entrainment = 0.022) effect is significant at the 5% level, indicating that trust relations between managers from different units are more common when their nominated words appear in the corporate identity claim. This result is consistent with H3: a relation of trust is more likely to be observed between participants working in different organizational units when their understanding of the organizational environment is expressed in terms of core elements of the organizational identity, history and purpose. By choosing words that co-appeared in the corporate identity claim, the odds of forming trust relations between managers from different units increases by a factor of $(e^{0.022}=1.02)$ .

This effect is detectable over and above a negative and significant (at the 10% level) cross-level entrainment (C4AXB entrainment = −0.007) effect shown in Table 4(c). The C4AXB entrainment effect is a general cross-level effect that does not take into consideration unit boundaries. The combined interpretation of the negative C4AXB entrainment and positive C4AXB mismatch entrainment suggests that a shared understanding of the organizational environment described by words that appeared in the corporate identity claim does not consistently promote interpersonal trust, except when trust relations span different organizational units.

Additional model-based findings

The empirical results that we have reported are not only consistent with the hypotheses we have proposed, but also reveal additional features of the multilevel network generated by the multiple relationships that we have observed between managers and words.

Table 4(b) presents the parameter estimates for the interpersonal trust network. Most of the significant effects are at the 5% level, but the receiver effect for age group is significant at the 10% level. The negative density effect suggests trust ties are sparse. Confirming both intuitions, as well as prior research (e.g. Caimo & Lomi, 2015), the positive reciprocity effect (0.872) suggests that it is $(e^{0.872}=2.39)$ times more likely to observe a reciprocated trust relation than an unreciprocated (i.e. asymmetric) one. Both the popularity and expansiveness effects are negative, indicating a truncated in- and out-degree distribution. In other words, people nominated and received a similar number of trust ties, and trust relations are not centralized on a few popular or expressive individuals. The positive transitive (AT-T) and popularity closure (AT-D) effects suggest trust also tends to form local closures. The AT-T effect also indicates a possible hierarchical trust structure where one person received more whereas the other sent more nominations. The positive ATA-D effect suggests structurally equivalent popular individuals are also likely to trust one another.

The interaction effects between trust and nodal attributes show that consultants are approximately $(1-e^{-0.766}=54\%)$ less likely to nominate trust relations, but they are $(e^{1.206}=3.34)$ times more likely to receive trust nominations given their advisory role. Older managers are $(e^{0.449}=1.57)$ times more likely to trust the adjacent younger age group, but they are $(1-e^{-0.214}=19\%)$ less likely to be trusted by the younger group. There is also a negative heterophily effect on age groups, suggesting people in similar age groups are more likely to trust each other. In terms of people’s seniority levels within their own organizational units, managers at a more senior level are $(e^{0.139}=1.15)$ times more likely to trust others for each decrease in seniority interval, based on the positive sender effect. Conversely, people who are more senior in the overall corporate group are $(1-e^{-0.400}=33\%)$ less likely to trust others. However, they are $(e^{0.364}=1.44)$ or 44% more likely to receive trust from more junior colleagues in adjacent intervals, as suggested by the positive receiver effect. Education heterophily is negative, and homophily on nationality and function is positive. It is also worth noting that the ‘boss of’ dyadic covariate effect is not significant, suggesting that the structure of trust network does not necessarily replicate the formal hierarchy.

In Table 4(c), the word nomination network shows there are tie centralization effects on both people and words based on the positive and significant star-effects for expressiveness (XStar2A) and word popularity (XStar2B). In other words, a limited number of individuals nominated more words than any other individual, and a limited number of words received more nominations than any other words. These centralization effects suggest greater variations on the nomination activities by the people, and on the popularities of words. The negative shared word popularity (X4Cycle) effect suggests people tend to nominate distinct words rather than sharing the common nominations, reflecting a rather diverse understanding of the organizational environment. Combining the cross-level tie centralization effects (XStar2A and XStar2B) with the negative (X4Cycle) effect, the word nomination network can be seen as being segmented by active individuals and popular words, and the active individuals may not nominate the same set of popular words.

For the interaction effects between the cross-level word nomination network and the trust relations, the model identified a negative and significant (at the 10% level) popularity association (In2StarAX) effect, suggesting a negative correlation between the number of words nominated by an individual and the number of trust relations received by that individual: the higher the number of words a person nominates, the fewer the relations of trust that person is likely to receive.

Discussion

Framing trust as a social relation linking senders, recipients and objects of trust (Schilke et al., 2021) facilitates the formation of testable hypotheses about the specific multilevel network mechanisms that may be associated with trust in organizations. It also enables exploration of how trust relations emerge within organizations, not only in terms of the attributional characteristics of the trustor (the sender of a trust relation), the trustee (the recipient) and the object of trust (the material basis for trust as a symbolic social relation), but also in terms of multiple configurations of ties linking these elementary components of trust.

That shared understanding contributes to trust is particularly important under conditions of uncertainty, when organizational members cannot easily predict or verify others’ behaviour and must rely on interpretive cues (Luhmann, 1979; Thomas, Zolin, & Hartman, 2009). In such situations, shared language provides a foundation for projecting and recognizing trustworthy intent across boundaries, particularly when task-based interactions leave room for ambiguity or misinterpretation (Weick, 1979).

We believe that the results of the study add novel elements to our understanding of the mechanisms that generate and sustain interpersonal trust within organizations. We have learned that organizational participants who express a similar understanding of their organizational environment are more likely to develop relations of trust. This highlights a new role for organizational vocabularies in shaping trust within organizations (Loewenstein et al., 2012). Extant research has shown that similarity in linguistic style likely encourages the development of social relations (Gonzales, Hancock, & Pennebaker, 2010; Ireland et al., 2011; Kovács & Kleinbaum, 2020). Yet available research on organizational vocabularies has not linked trust to cognitive similarity as revealed by linguistic similarity. According to Loewenstein et al. (2012, p. 31), shared vocabularies serve as ‘common ground’ for communication and coordination. Our results provide empirical support that such common ground helps facilitate trust among organizational members.

We have also learned that the blending effect of similarity in how participants express their understanding of the superordinate organizational environment on the presence of trust relations is constrained by the segregating effect of formal intra-organizational boundaries. We have shown that similarity in organizational vocabularies based on the subjective understanding of the organizational environment is not sufficient to sustain a relation of trust between organizational participants. These boundaries often encircle not only people and activities but also distinct ‘zones of meaning’ (Berger & Luckmann, 1966, p. 55). The segregating effect of administrative boundaries is too strong for shared understanding alone to support trust across them.

Our findings suggest that the segregating effect of formal intra-organizational boundaries on interpersonal trust relations may be offset by the tendency of organizational participants to express their understanding of superordinate organizational structures in terms of words associated with resonant dimensions of organizational identity (Baron, 2004). We found that trust relations are more likely to crosscut the ‘zones of meaning’ represented in our study by the subsidiary companies within the corporate group, when participants express their understanding of their superordinate organizational environment using words that are also used in the official organizational narrative about the corporate identity, history and purpose. Prior studies have associated blending processes of the kind we have documented in this study with a variety of desirable outcomes for organizations, organizational units and teams (Argote & Ingram, 2000; Hansen, 1999; Reagans & McEvily, 2003; Reagans & Zuckerman, 2001).

Implications

Our study provides clear evidence that shared organizational vocabularies may play a critical role in the formation of trust within organizations. This finding extends existing theories that emphasize cognitive and affective bases of trust (McAllister, 1995) by adding a linguistic dimension to the understanding of trust development. By showing that trust is associated with shared language and meanings, we advance the concept of the cognitive dimension of social capital (Nahapiet & Ghoshal, 1998). These insights suggest that fostering shared organizational vocabularies may be a valuable strategy for enhancing trust across unit boundaries, thus improving coordination and collaboration. The study also extends current research on social networks and organizational vocabularies (Tasselli et al., 2020) by linking trust more explicitly to social cognition and language within organizations. Our findings contribute to a growing recognition that trust in organizations is not only a response to stable expectations, but also a mechanism for navigating communicative and relational uncertainty (Schilke et al., 2021; Luhmann, 1979). By showing that trust tends to form when actors use similar vocabularies, especially those aligned with official identity narratives, we highlight how language can serve as an anchor when more objective markers of reliability are absent or unclear.

Our findings have broad implications for organizational research which we invite future studies to explore. At least two sets of implications deserve mention in this concluding section because they may contribute to advance two major lines of contemporary organizational research. The first set of implications concerns the potential role that superordinate entities and identities may play in supporting trust within organizations – particularly across intra-organizational boundaries (Argote & Kane, 2009). According to Argote (2024, p. 423) a superordinate organizational identity ‘increases the likelihood that units will be motivated to transfer knowledge and to consider information contributed by others very thoroughly. These conditions create a supportive context for knowledge transfer.’ We have found that achieving this desirable goal may depend, in part, on the extent to which the vocabularies of organizational participants and vocabularies of organizational identity intersect. This result adds important detail to available empirical evidence on the effects of superordinate identities on knowledge sharing within organizations (Kane, Argote, & Levine, 2005). The central result of our study does not dispute that language similarity may reveal a deeper cognitive similarity which, in turn, facilitates social selection (Kovács & Kleinbaum, 2020) and the establishment of trust among network partners (Ertug et al., 2022). However, when this general tendency is situated in actual organizational settings, our study demonstrates that language similarity alone may not be sufficient to overcome the powerful segregating effects of formal administrative boundaries on interpersonal trust.

Additional research is needed on the conditions under which superordinate institutional identities may be able to trigger the desegregation or blending processes needed for trust to flow freely within organizations unencumbered by internal social identities produced by organizational design. In this way, we can begin to address fundamental research questions about how institutions produce and reproduce trust across multiple levels of organizational analysis (Zucker & Schilke, 2020).

Second, our results support a new understanding of an organization’s culture of trust as both an outcome and antecedent of networks of interdependent micro-level associations linking organizational members to subjectively meaningful symbolic entities (like, for example, the words examined in our study). These individual associations crystallize into regular meso-level structures that enable and constrain a variety of organizational processes. The exchange of trusted advice across internal organizational boundaries that we have examined in this study is but one example of such processes. We are not the first to note the connection between organizational cultures and the multiple dualities (Breiger, 1974) connecting individuals to symbolic entities, material objects and concrete practices in organizations (Breiger, 2000; Mohr, 1994; Mohr & Duquenne, 1997; Mohr & White, 2008). We are also not the first to discover the connection between language similarity and social networks (Kovács & Kleinbaum, 2020). However, we are not aware of studies that have identified the specific micro-mechanisms linking organizational vocabularies and social networks of trust relations within and across organizational boundaries and levels (Fulmer & Gelfand, 2012). Future research may take our experience as a starting point for developing rigorous studies on the mechanisms linking ‘trust’ to ‘organizational culture’ through networks linking participants with a variety of organizational artefacts such as, for example, language, material objects, sites, categories and practices (Mohr & Lee, 2000).

Beyond these implications we are positioned to offer a grounded conjecture on how our findings may extend beyond the specific empirical scope of the study. We have studied a relatively conventional setting where the members of a small top-management team know one another personally and have frequent opportunities for face-to-face interaction, discussion and informal exchange of advice. The importance of trust in this context is, perhaps, unsurprising. But would the relational view of trust we proposed be meaningful when transposed to larger organizational environments undergirded by virtual relations supported by technology-mediated communication and platform-based interaction? Why, after all, should interpersonal trust be affected by membership in different groups when no actual boundaries exist to impair communication between groups and units? Evidence is mounting that the growing virtualization of interpersonal relations and the pervasive presence of interaction mediated by technological platforms with social media functionalities intensify, rather than alleviate, the polarizing effect of membership in different communities on interpersonal trust and knowledge exchange across communities (Bail et al., 2018). This happens because social media technologies tend to reduce, rather than increase, users’ exposure to diverse information (Iandoli, Primario, & Zollo, 2021). The result is the polarizing effect of virtual echo-chambers that has been repeatedly demonstrated in experimental (Hobolt, Lawall, & Tilley, 2023) and observational (Barberá, Jost, Nagler, Tucker, & Bonneau, 2015) studies.

To summarize our conjecture, and to conclude: if ‘what firms do better than markets is the sharing and transfer of the knowledge of individuals and groups within an organization’ (Kogut & Zander, 1992, p. 383), then it matters greatly how trust within organizations facilitates or precludes the sharing and transfer of knowledge.

Limitations

We believe that the limitations of the study reveal clear directions for future research. First, the way in which organizational vocabularies can be generated is not unique. The strategy we adopted has all the advantages and some of the disadvantages of simplicity (Breiger, Wagner-Pacifici, & Mohr, 2018). In the context of our study, we believe that the fieldwork leading to the selection of the words in the organizational vocabulary was very instructive and allowed us to learn more about the companies and their business and the individual members of the top management team, including the president and CEO of the industrial group. Extracting keywords from more structured text might have added a sense of objectivity to the process but might also have sacrificed the phenomenological and ethnographical richness of the research experience. A similar argument holds for the master document that we have used to define the ‘corporate vocabulary’. More documents could have been considered, and a greater variety of textual sources could have been consulted. This might have produced a richer and more comprehensive textual basis. Technological advances in natural language processing and text analysis offer exciting possibilities for students of organizations interested in examining organizational vocabularies. We hope to see future research coupling the sophistication of the statistical analysis that we have implemented in this study with a comparable sophistication in the collection and structuring of textual data to support creative and consequential organizational research on language and trust.

Second, measuring trust is notoriously difficult, and no agreement has yet been reached on the single best way (McEvily & Tortoriello, 2011). Our approach based on best recommended practice in social network research (Cronin et al., 2021; Marsden, 2011) may strike non-network researchers as exceedingly simplistic. However, our objective in this paper was to demonstrate the insight afforded by framing trust as a social relation, and hence a generator of interdependence. Following established best practice in empirical studies inspired by relational conceptions of trust (van de Bunt et al., 2005) allowed us to link data on the interpersonal trust relations among managers to powerful statistical models for multilevel networks that are now well understood and widely adopted in empirical organizational research (Amati et al., 2018; Wang et al., 2013; Zappa & Lomi, 2015). We would be hard pressed to think of alternative analytical frameworks that might have afforded a more direct connection between our hypotheses and their empirical translation.

Third, the results we have reported are clearly not based on a random sample, but on a specific organizational setting selected, at least in part, for its convenient structural features. For example, managers in our sample were members only of a single organizational unit (subsidiary company), which made the identification of intra-organizational boundaries unambiguous – a situation that is becoming increasingly less common in contemporary organizations (Mortensen & Haas, 2018). Consequently, the potential replicability of the results we have reported, and their empirical extension must be considered carefully when assessing the value of the study.

Finally, can we conclude that sharing a similar organizational vocabulary causes the formation of trust relations within organizations? No, we cannot. The purpose of the study was not to test causal claims about how organizational vocabularies and trust are related. It is generally understood that social selection and social influence mechanisms are typically confounded in observational studies of social and other networks (Shalizi & Thomas, 2011; Steglich, Snijders, & Pearson, 2010). This is particularly evident in studies based on cross-sectional research designs (Amati et al., 2018; Amati et al., 2021). Testing causal statements is not logically impossible but requires the development of additional assumptions about equilibrium processes (Butts, 2024), or community structure (McFowland & Shalizi, 2023). Different research designs, data structures and statistical models are needed for a study whose purpose is to disentangle homophily-based social selection (i.e. the tendency of individuals adopting a similar organizational vocabulary to develop trust relations) from social influence or social contagion (i.e. the tendency of individuals connected by a trust relations to adopt a similar vocabulary). Recent studies are beginning to find experimental evidence that language and social relations co-evolve (Kovács & Kleinbaum, 2020) – a result that casts a shadow of doubt over the intellectual and practical value of framing the relation between behavioural similarity and social connectivity in terms of unidirectional causal processes.

Although there is little doubt that the empirical setting of the study has several idiosyncratic features, and the research design constrains the scope of the conclusions supported by the statistical analysis, the questions we have posed and the hypotheses we have derived about the relation between trust and organizational vocabularies remain general and can be replicated on any data consistent with a multilevel network design. We invite scholars interested in relational theories and models of trust to replicate the research design implemented in this paper, and to collect data on multilevel networks that may yield results directly comparable to those we have reported in this study.

Conclusions

Relations of trust tend to be more prevalent between managers who share a common organizational vocabulary. This tendency appears stronger within organizational sub-units and significantly weaker across them. This pattern may be explained by the fact that the same cognitive and structural forces that foster relations of trust within bounded groups within organizations also tend to weaken trust across group boundaries. However, this tendency is mitigated when individuals from different sub-units use language that reflects core organizational values. Such shared values may help facilitate trust across internal boundaries by promoting mutual understanding and reducing communication barriers.

More broadly, the results of the study demonstrate the value of anchoring abstract notions of trust into concrete social relations that connect individuals within organizations. Taken together, these findings suggest that language and shared meanings play an important role in cultivating trust in uncertain organizational environments. As technological mediation, virtual teams and cross-unit collaborations become more common so too does the challenge of interpreting others’ intentions under conditions of limited information. In such settings, shared vocabularies act as cognitive shortcuts, allowing individuals to align their sensemaking frameworks and more confidently assess the reliability of their counterparts (Weick, 1979). In this way, our study shows that trust in organizations can be built not by avoiding uncertainty, but by using shared language to make uncertainty easier to understand and manage. We believe that the study’s findings suggest broader relevance beyond this specific empirical setting, providing a foundation for future research on how language and shared meanings are associated with trust within organizations.

Supplemental Material

sj-docx-1-oss-10.1177_01708406251349316 – Supplemental material for Take My Word for it: How organizational vocabularies foster trust across boundaries

Supplemental material, sj-docx-1-oss-10.1177_01708406251349316 for Take My Word for it: How organizational vocabularies foster trust across boundaries by Peng Wang and Alessandro Lomi in Organization Studies