Abstract
Starting with the premise that subteam psychological safety (PS) microclimates are vitally important to team behavior yet invisible to team-level PS concepts and measures, we introduce a multilevel theory and model of PS within work teams. We first demonstrate the inevitability and influence of distinct PS microclimates in teams and highlight the limits of current team PS approaches, and then develop a multilevel PS theory using social network methods. We introduce multilevel PS measures and theorize their influence on specific aspects of team and subteam learning and performance outcomes. These include new applications of traditional network metrics (e.g., team PS density, member-only PS density, subteam PS density, and leader PS centrality) and a newly developed multilevel team PS index (mPSi). The mPSi measure synthesizes multilevel leader and member PS influences in a single number to better predict outcomes in teams that engage in multilevel (subteam and intact-team) activity to meet work demands. We employ the new metrics to examine four archetypal team PS structures, contrasting new and current approaches and illuminating the implications of incongruity between subteam and intact-team safety climates. We propose that this multilevel theory extends the team PS literature, effecting far greater understanding and prediction of team outcomes and development, while increasing the number of team PS studies that reach publication.
In this article we introduce the idea that intact-team and subteam psychological safety (PS) dynamics coexist, interact, and may often misalign—with important implications for team learning, performance and development, and organizational team research. Edmondson (1999: 354) describes team PS as an overarching team “climate characterized by interpersonal trust and mutual respect in which people are comfortable being themselves.” Employees perceiving PS in their team environment are more likely to express authentic behaviors and directly communicate in the pursuit of team goals (e.g., Edmondson, 2004). While PS has been considered as both an individual perception of a particular work environment (e.g., Carmeli, Brueller, & Dutton, 2008; Kahn, 1990; Kark & Carmeli, 2009; O’Neill, 2009; Schein, 1985) and as existing in dyadic relationships (Tynan, 2005), the conceptualization of PS as an intact team-level variable (i.e., assessing the general team environment), constructed by aggregating high agreement individual perceptions into a team average, dominates the organizational literature (e.g., Baer & Frese, 2003; Bstieler & Hemmert, 2010; Edmondson, 1999, 2002, 2003; Walumbwa & Schaubroeck, 2009). This intact-team PS conceptualization is related to a wide range of positive learning and performance outcomes in organizational teams (e.g., Baer & Frese, 2003; Brueller & Carmeli, 2011; Bstieler & Hemmert, 2010; Edmondson, 1999, 2002, 2003; Edmondson, Bohmer, & Pisano, 2001; O’Neill, 2009; Schaubroeck, Peng, & Lam, 2011; Schein, 1993; Tucker, Nembhard, & Edmondson, 2007; Walumbwa & Schaubroeck, 2009).
Despite theoretical models and empirical evidence of the existence and importance of subteam dynamics within intact teams (e.g., Aubert & Kelsey, 2003; Carton & Cummings, 2012; Lau & Murnighan, 2005; Liden, Sparrowe, & Wayne, 1997; Nirmala & Vemuri, 2009), the extant team PS literature acknowledges neither the actuality of subteam structures within intact teams nor the possibility of subteam PS dynamics that are predictive of learning and performance outcomes. Subteams, whether intentional or incidental in their formation, often possess characteristics distinct from the greater team and can have either positive or deleterious effects on overall team functioning, ranging from serving as a specialized unit of production within the team (Aubert & Kelsey, 2003; Nirmala & Vemuri, 2009) to initiating a harmful fracturing of the larger team (Lau & Murnighan, 1998) and stimulating intact-team learning behavior (Gibson & Vermeulen, 2003). Despite their importance, subteams and their sometimes “hidden” PS microclimates are rendered invisible by current intact-team PS conceptualization and measures.
To address this gap, we introduce a multilevel theory and model of team PS and theorize its implications for learning, performance, and team development in complex teams. In the following sections we first develop the theoretical need and rationale for a multilevel PS model and then explicate the model using social network theory and analytical tools. We then introduce a range of multilevel PS structures and measures, including a new multilevel team PS index (mPSi), and new applications of traditional network measures including member-only PS density, subteam PS, and leader PS centrality, in each case proposing how the measure uniquely predicts or illuminates aspects of multilevel team behavior. To demonstrate the applicability and potential of the model for team research, we next interpret exemplar multilevel team structures and contrast new and existing methods of PS conceptualization and measurement. Finally, we discuss implications of incongruence between subteam and intact-team PS dynamics and offer specific guidance to team scholars in designing research.
Literature Review
The Intact-Team Approach to PS Conceptualization and Measurement
Team PS, as pioneered by Edmondson (1999) and commonly featured in contemporary organizational research (e.g., Leroy et al., 2012; Roussin & Webber, 2012), is an emergent quality of the intact work team, describing (Edmondson, 1999: 350) “a shared belief held by members of a team that the team is safe for interpersonal risk taking.” Employees in high PS teams do not fear negative consequences from the expression of novel ideas, authentic personal behaviors, or respectful disagreement with team members or leaders. As a result, employees in high PS teams more freely exchange task-relevant information, facilitating coordination, learning, and progress toward team goals. Team PS can be improved (or damaged) over time, dependent on the actions and interactions of a team’s leader and influential members (e.g., Roussin, 2008; Schulte, Cohen, & Klein, 2012).
Most PS research focuses on intact-team PS (i.e., the overarching PS climate of the entire assembled team). Intact-team PS has implications for team functioning, including greater goal achievement and financial performance in midsized German companies (Baer & Frese, 2003), learning in interorganizational product development teams (Bstieler & Hemmert, 2010), greater employee voice behaviors in a major U.S. financial institution (Walumbwa & Schaubroeck, 2009), improved team learning and performance in many settings including health care (Brueller & Carmeli, 2011; Edmondson, 1999, 2002, 2003; Schaubroeck et al., 2011), better team adaptation to change (Edmondson et al., 2001; Schein, 1993; Tucker et al., 2007), and reduced escalation of commitment to unpromising decisions (O’Neill, 2009).
As currently defined, team PS is a “shared belief” among team members (Edmondson, 1999: 350), raising the question of how PS is studied in teams where members do not share PS beliefs. The predominant approach to measuring team PS employs a referent-shift consensus model (Chan, 1998; Cole, Bedeian, Hirschfeld, & Vogel, 2011), employing Edmondson’s (1999) seven-item PS scale (see Table 1). Researchers collect individual perceptions concerning the generalized PS team climate from all or most team members and then aggregate the data into a team average (e.g., Bstieler & Hemmert, 2010; Edmondson, 1999, 2002, 2003). In consensus models, as in the current definition of team PS, evidence of within-team agreement is required before data may be aggregated. Intraclass correlation coefficient (ICC) estimates indicate how much of the variance in individual PS ratings is attributable to team membership (Bliese, 1998; Kenny & LaVoie, 1985). If the ICC value exceeds a threshold level, aggregation is considered justified (Bliese, 1998; LeBreton & Senter, 2008). Otherwise, aggregation is abandoned, relegating that aspect of the study to the “file drawer.” An unfortunate side effect of these aggregation practices is that only studies of high agreement groups exist in published research (Cole et al., 2011: 720), despite evidence that team members often possess divergent perceptions of team states and processes (Mathieu, Maynard, Rapp, & Gilson, 2008). The implication is that we know nothing about teams in which members do not share beliefs about team PS.
Intact-Team Psychological Safety (PS) Scale Items (Edmondson, 1999)
Recently methodologists have criticized the use of consensus models (Cole et al., 2011: 719-720) for their strict reliance on mean scores and lack of sensitivity to variance in individual responses. High variance indicates PS instability, and low variance indicates a stable, homogenous team PS climate. High PS variance also signals the possible presence of PS subteams. According to Cole et al. (2011: 720), “Prior results suggest that using only mean-based variables derived from direct consensus and referent-shift models may oversimplify group-level phenomena and result in biased (i.e., understated) estimates and equivocal findings (e.g., Colquitt, Noe, & Jackson, 2002; Dineen, Noe, Shaw, Duffy, & Wiethoff, 2007; Naumann & Bennett, 2002).” Consequently, mean-based team PS measures are insufficient to represent team environments with diversity in PS perceptions or with subteam PS microclimates.
We have demonstrated how intact-team PS conceptualization and measurement overlooks subteam dynamics within intact teams. Next we discuss the foundations of subteam PS microclimates in preparation for introducing a new multilevel theory of team PS.
Foundations of Subteam Psychological Safety Microclimates
Subgroups have long been recognized as an important factor in group dynamics (e.g., Homans, 1951, 1961). Much organizational work and decision making occurs at the subteam level, with implications for team learning and performance (Carton & Cummings, 2012; Gibson & Vermeulen, 2003; O’Leary & Mortensen, 2008). Subteams can produce outputs directly and trust between subteams influences team performance (Aubert & Kelsey, 2003). Whether intentional or accidental, subteams represent distinct microclimates within the team, characterized by relationships, commonalities, and contrasts among subteam members (e.g., Carton & Cummings, 2012; Deaux, Reid, Mizrahi, & Ethier, 1995; Lau & Murnighan, 1998). A positive (e.g., high PS) subteam environment can free members to feel comfortable and openly share ideas and information, while a negative subteam environment can lead to avoidance and generally uncooperative behaviors, with clear implications for team learning and performance.
Carton and Cummings (2012) offer a typology featuring three types of subteams in the interest of better explaining and delineating subteam formation, process and outcomes: identity-based subteams, resource-based subteams, and knowledge-based subteams. Next we briefly review this typology (Carton & Cummings, 2012) and describe how distinct PS dynamics likely develop within each subteam type.
Psychological Safety Within Identity-Based Subteams
Identity-based subteams form when “individuals perceive others as belonging to subgroups that represent shared values and social characteristics” (Carton & Cummings, 2012: 444). The social identity literature discusses subteam microclimates in terms of in-groups and out-groups (Tajfel & Turner, 1979), where individuals perceive themselves as similar to, and close with, other in-group members (e.g., Frey & Tropp, 2006). This in-group “closeness” evokes the likelihood of a unique PS microclimate. Out-group members, who may form bonds around issues of exclusion, shared values, or behaviors, could also develop into a high PS subteam. A low level of PS in an intact team may initiate the formation of PS subteams as members who value authenticity seek out high PS interactions.
The faultline literature also indicates identity-based microclimates that could result in differential subteam PS. Faultlines, “hypothetical dividing lines that may split a group into subgroups based on one or more attributes” (Lau & Murnighan, 1998: 328), can create identity-based subteams based on geographic colocation (Polzer, Crisp, Jarvenpaa, & Kim, 2006) or demographic factors such as gender, race/ethnicity, personality, age, background, or affiliation (e.g., Lau & Murnighan, 1998; Pelled, Eisenhardt, & Xin, 1999; Rico, Molleman, Sanchez-Manzanares, & Vegt, 2007; Thatcher, Jehn, & Zanutto, 2003). When activated by events, faultlines trigger team members to more closely identify with their subteam rather than with the larger team (Deaux et al., 1995; Polzer et al., 2006), catalyzing conflict and mistrust and negatively influencing team outcomes (Lau & Murnighan, 2005).
Psychological Safety Within Resource-Based Subteams
Resource-based subteams exist when “a clear hierarchy of subgroups is already in place and also when members attempt to create a hierarchy of subgroups that does not already exist” (Carton & Cummings, 2012: 445). The leader–member exchange (LMX) literature provides an excellent illustration of resource-based subgroups and origins of subteam PS. Team members and leaders can vary greatly in their perceptions of one another (e.g., Boies & Howell, 2006), with teams naturally splitting into in- and out-groups based on the quality of leader–member relationships, and with in-group members receiving higher quality and more positive information, emotional content, assessments, and opportunities from team leadership over time (Dansereau, Graen, & Haga, 1975; Graen, Novak, & Sommerkamp, 1982; Liden et al., 1997).
Psychological Safety Within Knowledge-Based Subteams
Members of knowledge-based, specialized subteams benefit from sharing common languages, technical jargon, perspectives, paradigms, and social cues (Carton & Cummings, 2012; Dougherty, 1992; Galbraith, 1974). Within a cross-functional marketing/engineering team, for example, discipline-based subteams are likely to develop. When subteams form based on functional expertise, each subteam is likely to develop its own microclimate and work process derived from differing cultural norms. Members of these subteams will likely experience a degree of PS within the subteam that is different than that in the intact team.
Having established that intact-team PS measures do not accurately represent teams with subteam PS structures, and that PS subteams can emerge from a variety of influences and are likely a common feature of organizational teams, we next introduce a multilevel theory of PS in work teams using social network theory and measures. In the process of introducing each new multilevel PS measure, we propose the influence of the measure on team and subteam learning and performance outcomes.
Psychological Safety Subteams Through the Multilevel Lens of Social Network Theory and Methods
We apply social network theory to illustrate our multilevel model of PS. Social network theory posits that individual behavior and relationships are embedded in an interdependent, complex, and often competitive system of social relations (e.g., Brass & Krackhardt, 2012; Burt, 1992; Granovetter, 1985). Social network researchers collect data concerning dyadic perceptions, relationships, or behaviors, combining all data into a social network diagram that visually depicts the network of such qualities. There is a growing literature depicting trust, friendship, advice, and LMX relationships (or “ties”) as social networks within organizations (e.g., Krackhardt, 1992; Krackhardt & Porter, 1985; Shah, Dirks, & Chervany, 2006; Sparrowe & Liden, 2005; Sparrowe, Liden, Wayne, & Kraimer, 2001), with each network revealing dyadic, subgroup and whole-network structures that were previously difficult to access or understand.
Beyond providing visuospatial information, social networks can be analyzed using metrics describing the density and shape in the network, the presence of cliques and “structural holes” (Balkundi, Kilduff, Barsness, & Michael, 2007; Burt, 1992), and the network centrality (e.g., Bonacich, 1987) of individual members. Thus, social network theory and metrics are uniquely suited to demonstrating, comparing, and contrasting the multilevel structure and dynamics of subteam and intact-team PS climates.
Social network methods make subteam PS microclimates visible, allowing for a more granular analysis that broadens the scope of team PS research. For example, Figure 1 depicts a social network illustration of a single work team including a team leader and six subordinate team members, with arrows (“ties”) between members representing directed feelings of PS (a detailed description of network features and metrics is found in the next section). Immediately visible in the social network picture are two high PS subteams, connected only by a single “brokerage” relationship between Members 1 and 4. It is immediately apparent that each of the subteams has potential for performance and learning outcomes, as does the overall team, particularly if work process is appropriately matched to the team’s PS structure.
Multilevel Team Psychological Safety (PS) Climates and Performance Outcomes
This approach allows explicit mapping of the relationships between multilevel PS characteristics and team learning and performance outcomes in complex work teams. These include teams where work dynamics make subteam existence/emergence more likely: where team tasks are purposefully designed for subteams (e.g., by expertise or function); where team leaders have partial (or fluid) participation in team workflow or leader–member relationships create favored in-groups and out-groups in the team; or where demographic or functional (or other) faultlines exist in the team. In these types of settings, the likelihood of PS microclimates existing is high, and these microclimates may contrast in important ways with intact-team PS measurements, calling for a multilevel approach.
Next we describe in detail the features and implications of the multilevel team PS network, explaining first the network in general then elucidating each multilevel feature and proposing its specific implications for team and subteam learning and performance.
The Multilevel Team Psychological Safety Network
The team PS network represents the collection of individual perceptions (“ties”) of dyadic PS (Tynan, 2005). Because PS ties indicate feelings of safety with risk taking and honest communication, they also represent likely patterns of authentic behavioral interaction. Tynan (2005: 228) notes that those who perceive a coworker as unsafe “are more likely to abandon clear and honest communication and favor silence, vagueness, misrepresentation, or understatement in the interest of avoiding negative implications.” PS ties are similar to trust ties (Chung & Jackson, 2013) but specific to evaluations of risk involved in communicating critical ideas or contradictions to held assumptions. PS ties are distinct from advice ties (e.g., Lazega, Mounier, Snijders, & Tubaro, 2012) in that a team member could solicit advice (e.g., expert opinion), even in low PS relationships, while avoiding contradiction or other risk taking that could improve the quality of learning. PS ties are distinct from “friendship” ties (e.g., Ellwardt, Steglich, & Wittek, 2012) in that PS evaluations can exist absent of friendship. Likewise, friendship feelings can exist absent of high PS feelings, particularly in friendships marked by inauthentic behavior (e.g., Theran, 2010).
In Figure 2 we introduce four hypothetical examples of multilevel PS networks within seven-member work teams. Although there are many permutations of team PS structures, we chose these four for their illustration of contrasting subteam and intact-team PS dynamics. Arrows indicate the direction of perceived PS between individuals. In three of the teams (Irreverent Cliques, Potential Mutiny, Controlled Conflict) there is significant PS “hidden” within an unsafe intact team. In the fourth team (Competitive Advisory) the team leader shares high PS perceptions with all team members, likely elevating intact-PS levels, while there is little PS among members. Next to each multilevel network picture are several representative network metrics produced using the Ucinet software package (Borgatti, Everett, & Freeman, 2002), along with an estimated intact-team PS measure (or likely range) for comparative purposes.
Multilevel Team Psychological Safety (PS) Structures, Metrics, and Implications
Team PS density is the total number of existing PS ties in the team network divided by the total number of possible ties in the network, expressed as a decimal proportion between zero and one. Team network density is commonly employed by organizational researchers. Balkundi and Harrison (2006: 49), in a meta-analysis, found that teams with a greater density of “interpersonal ties” were more likely to achieve team goals and more committed to team membership. Density can be imagined as a measure of fullness. A team entirely filled with PS ties (i.e., where all team members and the team leader find one another safe for authentic interaction) has a density of 1, with no room for addition of PS ties. Any subteam within this team will also be full (i.e., having the maximum number of PS ties) and therefore rich with authentic expression and productive risk taking. By contrast, a team with 50% of possible PS ties (half full; or with a density of .5) would be characterized by far less authentic interaction. Likewise, such a team would have fewer possible configurations of high PS subteams.
Proposition 1a: The greater the PS density in the team, the greater the multilevel learning and performance outcomes in the team.
Likewise, subteam PS density is the total number of existing PS ties in the subteam network divided by the total number of possible ties. Subteams with greater density of PS ties will be characterized by authentic interactions among subteam members.
Proposition 1b: The greater the PS density in a subteam, the greater the multilevel learning and performance outcomes in the subteam.
Member-only PS density is the number of existing PS ties in the member-only network divided by the total number of possible ties. This metric is important to understanding team learning and performance potential in the temporary absence of the team’s leader, where nonleader subteams complete much of the team’s work, and in those teams and environments where the leader’s direct participation in the team’s work is fluid or inconsistent. These conditions are all common, in particular where team leaders have both internal and external responsibilities (e.g., Gibson & Vermeulen, 2003; Manz & Sims, 1987).
Proposition 2: The greater the member-only PS density, the greater the learning and performance outcomes in the absence of the team leader.
The number of high safety PS subteams within the team is calculated using the Ucinet “cliques” procedure (where reciprocal PS perceptions connect multiple persons; minimum clique size = 2). Where subteams are apparent in a given team PS network, the presence of a single line between subteams represents “brokerage” between the subteams (Marrone, Tesluk, & Carson, 2007). The connected individuals are socially valuable liaisons filling “structural holes” and increasing coordination within the team (Balkundi et al., 2007; Burt, 1992). Where no line exists between subgroups, there is a PS-relational faultline in the team preventing the flow of authentic behavior.
Proposition 3a: The greater the number of PS subteams, the lower the intact-team learning and performance outcomes.
Proposition 3b: The greater the number of structural holes, or faultlines, creating unbrokered subteams in the team PS network, the lower the intact-team learning and performance outcomes.
The fourth metric is the team leader’s normalized in-degree PS centrality (“leader centrality”), or the proportion (a 0-1 value) of existing incoming PS ties that the leader has out of all possible incoming ties (e.g., Kilduff & Krackhardt, 1994; Sparrowe & Liden, 2005). Leaders are commonly cited as being most influential on the intact-team PS climate (e.g., Edmondson, 2004; Edmondson & Woolley, 2003). Positions of network centrality have been repeatedly associated with power and social influence in organizations (e.g., Hinings, Hickson, Pennings, & Schneck, 1974; Krackhardt, 1990; Sparrowe & Liden, 2005). In the team PS network, leader centrality is an indication of the proportion of high-quality exchanges that the leader will have with team members. In the meta-analysis mentioned earlier, Balkundi and Harrison (2006: 49) found that teams with a highly central leader had greater performance than other teams.
Proposition 4: The greater the team leader’s normalized in-degree PS centrality, the greater the subteam and intact-team learning and performance outcomes in the leader’s presence.
The Multilevel Team Psychological Safety Index (mPSi)
The multilevel team PS index (mPSi), newly introduced here, is designed to predict learning and performance outcomes in teams with multilevel work and PS structures, including teams characterized by in- and out-group dynamics and those that organize fluidly into subteams to meet various work demands. The latter are common and include top management teams and those dedicated to product development, management consulting, sales, engineering, and many other functions. The mPSi calculation is the average of member-only PS density and in-degree leader centrality (each may vary from 0 to 1), and as such it weighs the two quantities equally.
In a single measure, mPSi captures the inordinate importance of team leader influences on PS (e.g., Edmondson et al., 2001; Edmondson & Woolley, 2003) while encompassing key aspects of both intact-team and subteam PS climates. In a team where PS ties are common with the exception of the team leader, the leader’s low centrality, representing the leader’s strong negative influence on PS climate (Edmondson & Woolley, 2003), will reduce the index value, as any subteam configuration (or the intact team) that includes the team leader will be characterized by low PS leader–member interactions. In a team where the leader has many positive PS ties, mPSi is elevated, capturing the leader’s disproportionately positive influence on both subteam and intact-team work climates. However, the leader’s influence cannot raise or lower mPSi entirely (as it likely would using a mean-based, intact-team measure), as the density of the member-only PS network is an important feature of the team climate and likely characterizes much of the team’s work, particularly the part of it performed in subteams or in the absence of the team’s leader.
Proposition 5: Greater mPSi predicts greater learning and performance outcomes in teams with multilevel work structures.
Beyond predicting performance outcomes, mPSi indicates the degree of developmental intervention required to transform the team into one capable of high levels of multilevel learning and performance. In the lowest mPSi teams, both leaders and multiple team members contribute to an inefficient multilevel PS climate (or system of climates and microclimates) within the team. As such, multiple team members and the team leader will need to be replaced, trained, mentored, incentivized, coached, or otherwise altered. In teams with a midrange mPSi, less intervention is required, involving fewer causes of PS-related inefficiency (e.g., only replacement or development of team members). In teams with the highest mPSi levels still fewer interventions, if any, are necessary to transform the team into an efficient learning and performance state.
Proposition 6: Lower mPSi predicts a greater degree of developmental intervention required to make a team capable of high levels of learning and performance.
The Multilevel Team PS Index as Alternative to the Intact-Team PS Measure
Nearly all teams encompass aspects of both subteam and intact-team work process, making mPSi an excellent choice as a single team PS measure, with the added benefit that there is no requirement for researchers to establish within-team agreement before aggregation. All team environments are eligible to be measured, analyzed, and interpreted, reducing the number of team PS studies relegated to the file drawer due to low in-team agreement (Cole et al., 2011). It is not necessary, however, to employ only mPSi in a study of PS, learning, and performance in work teams. Taken together, a team’s PS network picture and metrics tell an expansive story, allowing simultaneous and separate consideration of PS dynamics at a variety of levels: individual (e.g., Will a particular employee thrive in a specific subgroup? Is a member/leader helping or limiting performance?), subteam (e.g., Which members/leaders should be assigned to a subtask? What subteam outcomes can be expected?), and overall team (e.g., What learning and performance outcomes are expected? Does the team have flexibility to configure into multiple, productive, subteam structures?).
To illustrate applications and interpretations of mPSi and the other multilevel PS measures, we next explore the PS structure of four organizational teams (Figure 2), explicating the meanings and implications of the network diagrams and metrics, and contrasting them with hypothetical estimates of intact-team PS. Figure 2 is organized with teams having the lowest PS density at the top of the figure, and we examine the teams in this order.
Irreverent Cliques Team
Based on leader influences alone the members of this team would likely rate the intact team as having very low PS. However, there are two very high PS subteams with potential for immediate cooperation and productivity if work process is matched to the PS subteam structure. The value for mPSi is low (.20), due to the combination of the leader’s very low centrality (0) and the member-only PS density value (.40), indicating that the team’s multilevel learning and performance outputs will suffer from the lack of authentic risk-taking behavior in the team. This team’s mPSi predicts learning and performance outcomes less than half as positive as the Potential Mutiny team (mPSi = .42). Clearly, the team leader plays no positive role in the PS dynamics of the team, suggesting that a more appropriate leader could potentially bridge the “structural hole” (Burt, 1992) between the two high PS subteams. Creating this structural PS bridge would allow risk-taking behavior, and associated learning, to travel between the subteams without members fearing negative implications.
Without multilevel PS information, a researcher knows only that the Irreverent Cliques intact team is psychologically unsafe. With multilevel PS information, the subteam PS structure is explicitly clear, as is the leader’s specific negative impact on PS dynamics and team outcomes. Although there are several “safer” members in the team, none create a safety bridge between the subteams. A lower mPSi team requires a greater degree of developmental intervention, in this case indicating that both team leadership and membership require developmental action. Specifically, a researcher could hypothesize that a new leader should come from outside the team, as no existing team members could conceivably build short-term positive relationships with members of both subteams.
Of the three team types introduced by Carton and Cummings (2012), this multilevel PS shape is most likely to occur for identity- or knowledge-based reasons, where value, knowledge, or responsibility differences create “us versus them” attitudes within the team (e.g., Jehn & Bezrukova, 2010; Polzer et al., 2006). In such cases subteam members have reasons to view themselves as fundamentally different from other subteams. Differentiated subteams are more likely to experience identity threat, decreasing the likelihood of PS development, particularly if there are two subteams of equal size as in the Irreverent Cliques team (Carton & Cummings, 2012). Identity threat is closely related to issues of authentic expression, as threatened subteams feel that the existence and expression of other subteams (and in this case the team’s leader) threatens their capability to easily and honestly voice their own uniqueness (Hornsey & Hogg, 2000). Carton and Cummings (2012: 459) point out that “a great amount of identity threat will likely impair a team’s ability to attain [an intact] climate of PS (Edmondson, 1999). As a result, members will not feel comfortable taking risks and performing trial-and-error experimentation with members from other subgroups.”
Competitive Advisory Team
The team leader represents a PS hub and is the sole PS boundary spanner (Marrone et al., 2007). Without the leader there are very few PS ties (density = .07). This PS structure is undesirable for teams that frequently work in the absence of the leader, or that require high levels of cooperation directly among team members. However, if the team’s work matches the PS structure of the team (as in an advisory team to a powerful politician or executive that meets in dyads), the PS structure may have some merit. As constructed, this team appears ill suited to creative problem solving in numbers of three or greater. However, as we discuss later, there is the possibility that the team leader, if very powerful and encouraging of safe participation in intact-team settings, could inspire unexpectedly high intact-team PS levels.
Cohen’s study of a health care management team exemplifies this team structure, where low PS in the intact-team due to scarce member–member PS leads to silence in team meetings, while small subteams (each including the team leader) have higher PS and greater communication.
The team relies upon Jean’s decisiveness to resolve conflicts among members. In meetings, managers more frequently respond to Jean than to each other. While this hub-and-spoke style of meeting leadership does keep overt conflict low, it also further contributes to member passivity. (Cohen, 1990b: 71)
Concerning team development, Jean seems talented in engendering safe dyadic relationships but not with enacting a team environment rich in PS. This may be a statement on Jean’s flaws (i.e., Does she play a role in member conflicts?) or on the character and history of the team members, who may have been purposefully recruited for their separate identities and knowledge bases. Another explanation is that Jean’s team, and the Competitive Advisory team, could be displaying traits of a resource-based subteam (Carton & Cummings, 2012), where the leader creates a series of high PS alliances (Levine & Moreland, 1998) rather than building a collaborative unit. Such a leader would control power and influence (Finkelstein, 1992) by failing to delegate meaningful, goal-oriented tasks to independent subteams, and could actively dissuade goal-oriented communication among team members. Each of these actions would result in a “hub and spoke” PS shape.
Potential Mutiny Team
This network PS structure could emerge from LMX relationships in the team where the leader favors a single team member. The outstanding feature of this team is a large (five-member) high PS subteam. The PS structure of this team illustrates relatively high PS density among members (.67) and very-low leader PS centrality (.17). Only one team member (M5) shares high PS perceptions with the leader, and no other member finds either individual safe for authentic interaction. If measured using traditional PS scale items (Edmondson, 1999) or a leader-focused scale (Edmondson & Woolley, 2003) this team likely has a very low intact-team PS level (due to the leader’s negative influence). However, it is important to note that if both the team leader and member M5 are absent from the team, the remaining subteam has perfect PS (density = 1). This subteam has corresponding potential for high levels of performance and learning. This team PS structure could derive from dynamics surrounding a new leader that has only “won over” a single member of the team. Alternatively, this could be an established leader who has “lost” a significant group of team members over time due to a lack of technical competence in knowledge-based subteams, displays of low behavioral integrity (Simons, 2002) in identity-based subteams, or unfair resource allocation in resource-based subteams (Carton & Cummings, 2012).
Cohen’s (1990a) qualitative study of a corporate restructuring team illustrates a large subteam with a seemingly high PS climate that becomes dysfunctional and apprehensive in the presence of its oft-absent and disruptive leader (who prefers communication with some members over others). This team would likely have a low intact-team PS (caused by the leader’s influence), obscuring the high PS subteam and the likelihood that a new high PS leader could radically transform the team’s performance.
Gersick’s (1990) study of “the students,” a temporary team of novice graduate management students engaged in a racially sensitive consulting project, illustrates another Potential Mutiny example. Composed of three White (Will, Grace, and Alexandra) and two Black members (Ken and Louanne), the team experienced reduced PS after Ken passionately described the team’s project client company as a racist organization during an early team discussion. Ken acted as the team’s initial leader, being knowledgeable, demonstrative, and eager to lead. However, only Louanne sympathized with Ken’s framing of this otherwise “respected” organization, although she favored a less controversial approach. Divided by racial and ideological faultlines (features of identity-based subteams), the overall team was unable to communicate or make progress toward its task, leading the course instructors to transfer Ken to another group (a mutiny, of sorts) and directly influencing a “major change in the way the group had been operating—and a dramatic leap forward in progress” (Gersick, 1990: 94).
In the Potential Mutiny figure, as in the case of “the students,” only one team member shares PS perceptions with the leader, while all others share high PS. If presented with an intact-team measurement scale, “the students” would likely measure as a very low PS team, obscuring the PS in the large identity-based subteam that would eventually produce the team’s output. The safety that may exist in “unsafe teams” can be highly relevant to team outcomes, whether those outcomes are existing or contingent on developmental action.
Controlled Conflict Team
The third type of subteam presented by Carton and Cummings (2012: 452-453) is knowledge based, illustrated by the fitting example of a team divided into engineering and marketing subteams. This scenario could easily result in the Controlled Conflict PS structure, where knowledge-based subteams develop into PS subteams, however with a leader who shares knowledge bases, and high PS perceptions, with members of both groups. In the absence of the team leader, intact-team PS density drops from a relatively high level (.62) to a much lower level (.47). The two PS subteams have different sizes (3 vs. 5), each with perfect PS (density = 1) and each including the team leader, who serves a brokerage role (Marrone et al., 2007). If the smaller subteam is the “marketing” subteam, then the ideas of engineering are likely to dominate the team (Carton & Cummings, 2012: 456) unless this tendency is mitigated by the team leader through balancing actions. Task-relevant ideas will flow freely within each subteam, giving each of the subteams high potential for learning and performance. The leader’s PS connections into both subteams give her or him access to, and the ability to disseminate, information throughout the team. This ability gives the overall team an opportunity, if not a likelihood (due to lack of direct exchange among divided subteam members), to balance engineering and marketing considerations in the pursuit of team goals.
Besides knowledge-based divisions, this team PS structure could result from identity-based faultlines, such as differences in values, age, income, or ethnicity. It is possible that this team could rate anywhere from low to medium on the intact-team PS scale, dependent on the division and process of work within the team (e.g., Are members of the two subteams required to mutually develop ideas despite their feelings of unsafety?). There is also the strong chance that this team would have a relatively high disagreement concerning the level of intact-team safety, as members of the larger subteam could possibly rate the intact team as high PS while the marginalized small subteam members could rate the intact-team PS as very low. As noted earlier in this article, studies involving such teams would likely be discarded to the file drawer using current mean-based, consensus models (Cole et al., 2011).
If intact-team PS were measured alone (as low or medium), we would lose the image of a team with two pockets of high PS. Instead we would imagine a single low PS and homogeneous team climate. If the intact team somehow rated as high (likely requiring careful planning and extremely skilled leadership), then we would similarly lose visibility of the two subgroups, leading to the incorrect conclusion that all members could cooperate.
In summary, conceptualizing PS as a construct existing at both the subteam and intact-team levels lends a richer, more accurate understanding of PS dynamics. The intact-team PS concept and measurement remains meaningful as it captures the pure quality of the intact team (if there is agreement among team members). In teams that perform most work together, where all team members are present throughout the work process, where in-group/out-group subteams have not inadvertently formed, and where subteams are not responsible for autonomous and significant discussion, innovation, or productivity, the intact-team method is an appropriate sole approach. However, we believe such teams to be relatively rare, signaling the need for a multilevel approach to team PS conceptualization.
Implications of Intact and Subteam Psychological Safety Incongruence
As illustrated, modeling multilevel PS reveals potential incongruence between subteam and intact-team dynamics. In this section we delineate the learning, performance, and developmental implications of such incongruence. Learning and performance implications concern how multilevel PS incongruence influences likely progress toward team goals and the acquisition and incorporation of new productive behaviors within a team (e.g., Edmondson, 2002; Edmondson, Dillon, & Roloff, 2007). Developmental implications concern how multilevel PS incongruence influences team intervention requirements to first improve PS dynamics and then learning and performance outcomes (e.g., Hackman, 1987). We discuss these implications relative to both incongruence possibilities: (a) where there are high PS subteams in low PS intact teams and (b) where there is high intact-team PS despite a scarcity of subteam PS. Based on our review of the subteam literature, we consider the latter case to be far rarer than the former but still worthy of brief discussion.
Subteam Psychological Safety in an Unsafe Intact Team
Team learning and performance implications
Team-based learning studies have typically pointed to the negative effect that subteam formation has on overall team learning outcomes (e.g., Lau & Murnighan, 2005; Van der Vegt & Bunderson, 2005). However, researchers have largely ignored the positive learning and performance potential of highly functional (e.g., high PS) subteams within dysfunctional (e.g., low PS) intact teams. Teams will perform better in subteams if the subteam PS structure is matched to the team’s work demands and process. High PS subteams have both immediate and future potential for learning and productivity, which is best understood by reviewing the fundamentals of team (and subteam) learning processes.
Gibson and Vermeulen (2003: 222) refer to team learning as a “cycle of experimentation” involving mutual reflection, a process that relies on PS in the team climate. Teams with stable membership more readily acquire tacit knowledge (e.g., nuanced understandings of how other team members behave) related to improving efficiency and coordinating behavior (Edmondson et al., 2007) in the pursuit of team goals. This learning through experience relies on exposure to other team members’ unadulterated behaviors, opinions, and attitudes (Reagans, Argote, & Brooks, 2005), an exposure that is inhibited in low PS intact-team environments. However, within the microclimate of a high PS subteam members can learn through the experience of subteam interactions, freely exchanging information, revealing novel ideas, and developing shared mental models concerning the most efficient methods of completing subteam work (Edmondson et al., 2007).
Alignment of work structure and “hidden” PS structure
Team leaders typically design the task structure of the team in addition to coaching the team and communicating its progress to interested parties (e.g., Hackman & Wageman, 2005). As discussed concerning the Irreverent Cliques and Potential Mutiny teams, where leader PS centrality is very low (or zero) and intact-team PS is likely very low, these learning and performance implications will benefit the greater team only if the high PS subteam is an important, functioning subunit of the team (i.e., assigned to a subtask), thus representing an alignment between the PS structure and the working structure of the team. High PS subteams that include the team leader (who coordinates work process) or that are connected to the leader through brokerage are more likely to produce learning or performance outcomes that benefit the greater team. There are many examples of teams in which a subteam yields all of a team’s meaningful output, as any teacher of students organized into project teams is well aware. However, in the absence of alignment with work tasks or brokering to connect positive team outcomes with the team leader (and therefore the reporting structure outside the team) the PS of the subteam likely represents unrealized learning and performance potential. To unlock the potential of “hidden” PS subteams, a team most likely requires developmental intervention(s) to connect the learning and performance potential of PS subteams with the team leader and other team members.
Team development implications
A “hidden safety” PS structure has both descriptive implications for team researchers and prescriptive implications for practitioners. Developmental implications of “hidden subteam safety” are intertwined with performance implications, in that the performance of the high PS subteam can be theoretically unlocked through development actions. Researchers have found evidence that teams develop in positive ways through changes in membership, goals, and incentives, or through identification and implementation of training or mentoring programs among other options, with varying degrees of success (e.g., Hackman, 1987; Hackman & Wageman, 2005; Stevens & Campion, 1994).
Researchers adopting a longitudinal design using multilevel PS methods, optimally combined with a field experiment with some control over team intervention choices, could deeply explore the question of how low PS intact teams with high PS subteams develop over time. These teams are typically characterized by low leader PS centrality as well as by structural holes (i.e., unbrokered subteams), with the behavior of the team leader being a primary cause of both conditions. As such, we theorize that hidden safety teams that experience leader-focused interventions, involving replacement, goal/incentive adjustment, or training of the team leader with an emphasis on safety building, will develop greater mPSi, intact-team PS (Edmondson, 1999), and leader centrality than other teams and will also have fewer structural holes (due to the leader’s improved PS brokering ability). We further theorize that these same interventions will improve team learning and performance outcomes, moderated by the degree of alignment between subteam structure and team work structure. Finally, we believe that leader-focused interventions will be more effective than member-focused interventions in improving both mPSi, intact-team PS, and team learning and performance outcomes in hidden safety teams.
Psychologically Unsafe Subteam(s) in a Safe Intact Team
Team learning and performance implications
As discussed concerning the Controlled Conflict and Competitive Advisory teams, positive, highly skilled, and influential team leaders and members can occasionally create a high PS intact-team climate from low PS pieces. Influential team leaders or members can encourage safe behaviors in the intact team by example, through active influence tactics, or through a system of well-designed incentives (e.g., Edmondson, 2004). The team’s leader, being the only network “broker” between subteams, potentially symbolizes safe behavior within the team, and has potential to elevate the PS climate of the intact-team to unexpected levels. Leaders who are both powerful and considered particularly trustworthy or virtuous may inspire cooperative behavior from employees (e.g., Dirks & Ferrin, 2002), particularly when employees are under the leader’s direct and immediate supervision. In the leader’s absence, however, employees are more likely to revert to natural behavioral patterns. In rarer cases, exceptional team members can have the same unifying effects. For example, Abramis’s (1990) study of a semiconductor manufacturing team shows how the efforts of two “gregarious” and respected team members successfully promoted norms of patience, tolerance, and helpfulness out of a collection of feuding team members.
We theorize that if a medium to high PS intact team divides into low PS subteams, the team is best suited to both learning and performing as an intact team, and will experience better intact-team outcomes than subteam outcomes. Reducing this type of team, for example the Competitive Advisory team, into subteams to meet work demands is a sort of “subteam roulette,” as dyadic safety is so sparse among members. We further theorize that such teams, when engaged in subteam work, will experience lower performance than medium to high PS teams with greater member–member PS density. It is important to note that without a multilevel PS perspective, incongruence between subteam and intact-team PS would be invisible to a researcher, and the intact-team would resemble any other high PS team, including those with several configurations of high PS subteams. However, multilevel PS modeling reveals the precise perils associated with specific subteam work structures.
Team development implications
Again, researchers adopting a longitudinal design using multilevel PS methods, combined with a field experiment with some control over team intervention choices, could effectively explore the question of how high PS intact teams with a scarcity of member-only PS develop over time. In contrast to hidden safety teams, these teams are typically characterized by low member-only PS density and high leader centrality, with the team leader serving as the primary, or only, broker between team members or small subteams. The emergence of this PS structure may be due to the leader’s behavior (e.g., sabotaging member–member perceptions in an effort to maintain control) or that of the team members. Regardless, we theorize that teams experiencing member-focused interventions, involving replacement, goal/incentive adjustment, or training of the team members with an emphasis on safety building and self-management, will develop greater mPSi and member-only PS, will maintain or build intact-team PS (Edmondson, 1999), and will also have greater incidence of nonleader brokerage in the PS network. We further theorize that these same interventions will improve learning and performance outcomes, again moderated by the degree of alignment between subteam structure and team work structure. We propose that leader-focused interventions with a similar focus on building member–member PS will have both direct and moderated influence on mPSi, member-only PS, and team learning and performance outcomes in these teams.
Approaches to Measuring Multilevel Psychological Safety
A multilevel conceptualization of PS requires rethinking how the concept is measured. In this section we make recommendations for measurement, contingent on team characteristics, to best enable effective research on teams with subteam PS dynamics.
Is Multilevel PS Always Necessary? Matching Team Structure to PS Measurement
Whether a multilevel approach to PS measurement is necessary depends on the nature and likelihood of subteam dynamics. As described earlier, subteams may be intentionally created to match the work requirements of a team, or may emerge as a result of in-group/out-group dynamics. The two-by-two matrix in Figure 3 illustrates a contingency approach to measuring PS within a team based on the extent to which (a) subteams are a formal component of the team’s work and (b) PS subteam dynamics are likely to emerge within the team.
A Contingency Approach to Multilevel Psychological Safety (PS) Measurement
The x-axis in the 2 × 2 concerns the intentional design of work in the team (i.e., is subteam work planned?). The y-axis represents the likelihood of the emergence of PS subteams based on the activation of faultlines or other in- and out-group dynamics. The cells recommend the level(s) of analyses at which PS should be measured in each circumstance. Below we describe contingent usage of intact team measures and subteam measures including a description of each cell in Figure 3.
Intact-team measurement
In all cases, Edmondson’s (1999) team-level PS measure likely adds descriptive value either as a sole measure (for intact-team work process and low likelihood of PS subteam emergence) or complementary measure (in all other cases), although studies with low agreement within teams will prohibit the use of intact-team PS. A researcher could also consider using a dispersion composition model (Cole et al., 2011), which adds the variability of within-team PS perceptions, as an independent variable (e.g., greater disagreement concerning intact-team PS may hinder learning). Dispersion composition models do not offer the levels or types (subteam PS structure and metrics) of information that our recommended multilevel approaches offer, but these models do draw clear distinctions between intact-team climates that previously appeared identical without additional data collection.
Social network PS measurement
The multilevel approach is recommended for any team in which subteams are planned, likely to emerge, or in uncertain team environments. The team PS network is an aggregation of dyadic PS (e.g., Tynan, 2005) perceptions. To collect these perceptions, we recommend that each team member answer one to three questions, using the popular roster method (e.g., Carpenter, Li, & Jiang, 2012; Wasserman & Faust, 1994), concerning the specific PS-related behaviors of every other team member (including the formal team leader if one exists). The roster method of network data collection is well suited to small work teams and allows for rapid consideration and completion, as questions concerning coworkers are read once and then coworkers are rapidly rated in succession. Single items are a viable option (particularly in larger teams), as they are convenient and remain popular in many published network-based studies, but can raise validity concerns with some reviewers.
Our suggestions for these items are dyad-specific rewordings of three of Edmondson’s (1999) team PS items: (a) “If I make a mistake working with [coworker], he/she is likely to hold it against me.” (b) “[Coworker] would never deliberately act in a way that undermines my efforts.” (c) “I feel that [coworker] values my unique skills and my contributions to our mutual work.” Following the example of many others (e.g., Goodwin, Bowler, & Whittington, 2009; Shah et al., 2006), we recommend using a 5- or 7-point Likert-type scale (1 = strongly disagree, 5 or 7 = strongly agree). After reliability and dimensionality checks, the items may be averaged to obtain a dyadic PS level. If a particular dyadic PS level exceeds a cutoff value, logically the value assigned to agree in the Likert scale, then the data are dichotomized and a directed “tie” (a line with an arrow) exists in the team PS network (e.g., Goodwin et al., 2009; Shah et al., 2006). Once data are collected, the researcher may calculate network PS metrics, including team PS density, member-only PS density, leader centrality, subteam densities, and mPSi. Contingent on the researcher’s underlying research questions, each metric may be employed as an independent or dependent variable.
Targeted subteam PS (multilevel, non-network)
This is a reduced effort (in data collection) multilevel alternative to network PS, recommended where network data collection is impractical, or for situations where a team is purposefully divided into process-based, goal-directed subteams and where the likelihood of further, informal subteams developing is considered low. This deductive approach involves an adaptation of intact-team PS methods to measure the PS of the predesignated subteams in addition to the PS of the intact team. Essentially the subteams are treated as nested intact teams during data collection, allowing comparison of team and subteam PS levels. As discussed, PS levels within stable subteams may differ from or even contradict the PS level of the intact team. What is “lost” using this method is information concerning brokerage between subteams, leader centrality, and specific team member influences on PS dynamics and outcomes.
Conclusions
Further Implications for Both Theory and Practice
Although existing approaches to PS measurement and theory are valuable as indicated by the amount of research showing significant relationships among team PS, learning, and performance, there is obvious room to expand approaches to better reflect the complexity of team behavior. In our field we do not typically see studies published without significant main effects—a distinct possibility where subteam dynamics influence or contradict theorized intact-team effects. As explained, there are likely a large number of potentially important PS studies residing in file drawers (electronic or real). These include studies containing low agreement among team members (Cole et al., 2011: 720) concerning the PS climate and studies featuring negative or ambiguous results that could, with the multilevel theory and approaches that we propose, advance the PS and team learning and performance literatures significantly.
We have been careful to explain how multilevel PS provides different information than intact-team methods and serves as either an improvement or a complement dependent on the research context. A multilevel approach to PS measurement reveals team properties that are invaluable to both team researchers and practitioners looking to develop their organizations. Within-team dynamics such as the “poisonous team member” are made obvious in a multilevel model. Team leaders who were assumed to be nurturing by senior management may be revealed as unsafe and prohibitive of team development. Team members, and subteams, that are perceived as being safe within unsafe teams can be recognized for their positive qualities. Using longitudinal models, team leaders who transform disparate parts into a safe and cohesive whole can be differentiated from those who simply manage a group of high PS team members.
Researchers note that leaders are most influential on intact-team PS outcomes (Edmondson & Woolley, 2003), and this seems an obvious idea. However, it is possible that leaders have been over- or underattributed as the cause of intact-team PS outcomes (or are contingently so dependent on context). The multilevel concepts and methods described in this article can provide visibility into those dynamics. In teams where leaders are not consistent contributors to the team, it is possible that important nonleaders are as influential, or more influential, than the formal team leader in shaping PS.
Limitations of Network Measures and Approaches
Network measures, although providing many advantages, do have two important limitations that deserve mention. The networks described here are representations of binary information. Ultimately, each PS tie exists dependent on a cutoff value, without a representation of the degree of PS intrinsic to the tie. PS values can be represented in social network diagrams (e.g., with varying tie thicknesses and labels) by programs such as Ucinet (Borgatti et al., 2002), but these values do not factor into the calculation of the various multilevel PS metrics. Instead, the researcher chooses a threshold value for the existence of a tie, dichotomizing the valued data into binary data. Something very significant is gained in the process (the PS team network and metrics) and something is also temporarily lost (valued PS-level data). However, once high PS subteams are identified the researcher may easily calculate the average dyadic PS in the subteam. Many network studies avoid this loss of precision by simply collecting yes/no data (e.g., “Is person X your friend?”). As a second limitation, collection of network data is fundamentally more sensitive (e.g., highlighting evaluation of specific individuals and relationships), and therefore difficult, than collection of referent-shift consensus data. Both institutional review boards and participating organizations will be cautious in granting approval or access to teams and individuals, and researchers will need to exercise caution in protecting the confidentiality of participants (e.g., Borgatti & Molina, 2003).
Future Directions
Kilduff, Tsai, and Hanke (2006: 1031) called for progressive multilevel network theory development “from a set of core concepts (Lakatos, 1970) comprising primacy of relations, ubiquity of embeddedness, social utility of connections, and structural patterning of social life.” This multilevel modeling of PS in work teams, revealing rich multilevel dynamics, is an example of such work. The “safety in unsafe teams” featured in the title of this writing, although socially complex, is rendered clearly and simply with the aid of a multilevel, social network lens. Through this rendering one can easily grasp both the health of the “safe” subteam and its embedded peril within the larger, unsafe team, along with the myriad of other dynamics and predictions introduced in this article. The most obvious future direction for these ideas is their application in team performance and development scholarship and management decision-making analysis. We expect benefit in both applications from the greater visibility into psychological safety and team behavior that this multilevel approach offers.
Multilevel Modeling of Other Well-Known Team Constructs
Beyond the present in-depth exploration of multilevel PS in work teams, there is similar benefit to be gained from applying multilevel principles to other concepts that are currently imagined exclusively at the team level. These concepts include prevalent “independent variables” such as (sub)team cohesion (e.g., Harrison, Price, & Bell, 1998), (sub)team efficacy (e.g., Durham, Knight, & Locke, 1997), (sub)team empowerment (e.g., Kirkman & Rosen, 1999), and (sub)team leadership (e.g., Hackman, 2002)—and also the “dependent variables” that they are often employed to predict, such as (sub)team learning (e.g., Edmondson et al., 2007) and (sub)team effectiveness (e.g., Hackman, 2002). As each of these traditionally team-level constructs is unique in its mechanisms and influence on behavior, we believe that each deserves its own in-depth multilevel theory development.
Footnotes
Acknowledgements
This article was accepted under the editorship of Deborah E. Rupp. Christopher J. Roussin wishes to thank Carla Ryder and Bill Stevenson for their exceptional support at important moments during the development of these ideas.Chris Roussin wishes to thank Carla Ryder and Bill Stevenson for their exceptional support at important moments during the development of these ideas. We all wish to thank two anonymous reviewers and a special group of Boston College OS PhD alumni (“Fulton 214”) for their valuable feedback.