A Multilevel,Emergent Journey to Unpack Team Process Dynamics

Point of Departure

To situate the beginning of my scientific journey, it is useful for me to reflect on my graduate school experience as it set me on a unique pathway. Prior to starting grad school, I had held many low-level jobs in manufacturing, retail, sales, and insurance claims. Those work experiences were the source of many salient situations in which leadership, teamwork, and systems thinking were deficient. I entered grad school at The Pennsylvania State University (PSU) as a student in industrial and organizational psychology (IOP) highly motivated to understand organizations as systems. As I encountered IOP as a field of study, my initial reactions were “Where are the teams and systems? It is all just about individual level correlations.” I decided I would pursue the PhD in IOP anyway and would figure out the missing teams and systems aspect along the way.

Early on, I encountered, by happenstance, a book by Roberts et al. (1978) entitled, Developing an Interdisciplinary Science of Organizations. The book did an excellent job of encapsulating the levels conundrum in organizational science. That is, that the study of organizations was/is sliced into distinct disciplinary domains—micro/individual, meso/team, macro/system—with limited attention to the interplay among levels and temporal dynamics. The authors did not offer a solution, but they did articulate the nature of the problem well, which I embraced as my intellectual north star.

Around this time, my advisor, Jim Farr, and an engineering colleague of his were awarded a National Science Foundation (NSF) grant to study factors that would facilitate the technical updating of engineers. I was a research assistant on that grant, which afforded me a great opportunity to interact directly with engineers across multiple organizations as the project team developed a measure of updating climate (Farr, Dubin et al., 1983) and a behavioral anchored rating scale (BARS) of updating performance (Farr, Enscore et al., 1983). Although the focus of the grant was on motivational factors that would facilitate the propensity of engineers to engage in technical updating behavior, I was strongly influenced by my interactions with the engineer informants and the critical incidents they recounted. My focus was squarely on the organizational climate for updating as an explanatory mechanism for engineer updating behavior.

My dissertation was a pioneering (at the time, although primitive from a contemporary perspective) multilevel model of engineer updating behavior. Essentially, it proposed that organization contextual features—technology and structure—would influence individuals’ perceptions regarding updating climate and job characteristics, which would in turn influence their updating behaviors (Kozlowski & Farr, 1988). The NSF grant enabled the project team to collect a rare data set that included 10 organizational-level units, 220 meso-level subunits (supervisors), and 483 individual-level engineers nested in those subunits, thereby allowing me to conduct multilevel research for my dissertation. ¹ I will note that it was very challenging to get the dissertation research published—it did publish—but it was rejected by every top-tier journal to which it was submitted. Although it did find a home, the essential challenges were that multilevel theory (MLT) and methods were in their infancy; largely unknown outside a small circle of scholars; and substantially underappreciated by editors, reviewers, and pretty much everyone in the field. I realized early on that moving MLT from the fringes of organizational science (OS) to the mainstream was going to be a long slog. Nonetheless, I was motivated, persistent, and determined to advance MLT out of the shadows to the center stage of OS.

Formative Development

As I started my professional career at Michigan State University, ² I pursued a variety of research targets—tenure, after all, is a publish or perish endeavor—but was most focused on examining basic theoretical issues in organizational climate theory and on examining the role of leaders in shaping the formation of individuals’ climate perceptions. Although MLT was not mainstream, it gave me a unique theoretical perspective that enabled the development of innovative research ideas.

For example, the origin of the concept of climate can be traced to Lewin et al. (1939). In that stream of research, Lewin and colleagues manipulated the styles of leader behaviors (i.e., authoritarian, democratic, and laissez faire) and observed effects on follower behavior, concluding that leaders have an important role to play in shaping climate. These ideas were later popularized by Lewin and colleague’s students as represented, for example, by McGregor’s famous Theory X and Theory Y styles of leadership (McGregor, 1960). Yet, in the subsequent decades, the conceptual relationship between leadership and climate was forgotten. MLT provided a unique way to link leadership and climate back together. Drawing on leader-member exchange (LMX) theory and applying a MLT lens, I reasoned that those employees with good LMX relationships with their leaders (i.e., in-group members) would have their climate perceptions shaped by their leader (a) such that their perceptions would be similar to the leader’s and (b) similar to the climate perceptions of other in-group members, (c) whereas outgroup members would have climate perceptions that were not similar to the leader’s or to each other (Kozlowski & Doherty, 1989). This research is a good exemplar of the conceptual advantages facilitated by MLT; the paper has been highlighted as one of the 12 most important climate articles published in the Journal of Applied Psychology (Schneider et al., 2017).

My research on climate was very much influenced by leading scholars (i.e., Larry James, Ben Schneider) who were contributing to the development of a cohesive theoretical perspective on climate that was, at the same time, also pioneering formative work on the development of MLT. Schneider and Bowen, (1985; Schneider & Reichers, 1983) was a proponent that climates reflected a unit’s “strategic imperatives” such that they were not generic but were targeted (e.g., climate for service, safety, or updating), influenced collective behavior, and organizational outcomes. James was also examining basic questions and was probing necessary conditions for representing individual perceptions of climate as a higher-level construct (James, 1982; James et al., 1984).

At the time, the level at which climate existed, how it should be measured, and how it should be represented (i.e., conditions for measurement and aggregation) were in ambiguity and controversy. However, an influential review by James and Jones (1974) distinguished between individual-level climate perceptions or psychological climate in contrast to a higher-level representation of organizational- or unit-level climate. Previously, Forehand and von Haller Gilmer (1964) had suggested that collective climate (i.e., aggregating individual perceptions to a higher-level) might reasonably be justified by restricted within unit variance. Early efforts to examine this idea relied on the intraclass correlation (ICC), a form of interrater reliability, to index the extent to which total variance (within unit + between unit) was attributable to between unit differences There are a variety of issues related to the use of ICC as a justification for aggregating data (Bliese, 2000; Kozlowski & Klein, 2000) that make it less than ideal. James et al. (1984) proposed a statistical index, r_wg(j), to assess restricted within group variance directly (i.e., agreement), without reference to total variance.

As a follow on to my dissertation, I was using the NSF data set to examine basic theoretical questions under discussion around climate (Kozlowski & Hults, 1987). With respect to the strategic imperatives for technical updating, (1) do external organizational environments with different pressures for innovation versus maintaining the status quo (James & Jones, 1974; Miles et al., 1978) influence different climates for updating, (2) evidence consensus within organizations showing that an organizational-level mean was meaningful as a collective construct, and (3) predict individual updating responses over time? The answers were yes, yes, and yes. This research employed MLT and contributed to the development of climate theory. Getting that paper published at that point in my career, given the issues I encountered getting my dissertation published, was a small triumph. ³

Shortly after, Schmidt and Hunter (1989) published a critique of r_wg(j), using Kozlowski and Hults (1987) as a vehicle for their critique points, inappropriately so in many instances. As I was just gaining some visibility and hitting my stride as a researcher, I was mortified. Larry James was very supportive and assured me that he was working on a response. Although I was not pleased with the way my research was characterized, the Schmidt and Hunter (1989) critique turned out to be a blessing in disguise. I delved deeply into the research on agreement versus interrater reliability (i.e., consensus vs. consistency, respectively) with respect to MLT—issues that were the crux of the James et al. (1984) derivation of r_wg(j) and applied in Kozlowski and Hults (1987) – and developed a critique (Kozlowski & Hattrup, 1992) of their critique. James et al. (1993) also responded. To put this in perspective, I was basically a just-getting-known scholar, whereas James, Schmidt, and Hunter were all highly influential scholars. I learned quite a bit from the entire experience and developed an even stronger motivation to advance MLT.

By this point, I was teaching a graduate seminar on MLT and methods, delving deep into the basic literature, trying to leverage what I had learned, and working to craft a coherent understanding of MLT from a disparate, contradictory, and confusing literature. The excellent and wide-ranging review by Rousseau (l985) served as a touchstone for many of the readings I used, supplemented by the perspective on MLT I was developing through leading the seminar. That seminar served as a crucible for development of the MLT principles that were advanced by Kozlowski and Klein (2000). Many of the graduate students who took that seminar went onto make their own important contributions to the development and application of MLT (e.g., Brad Bell, Jason Colquitt, David Chan, Stan Gully, Stephen Humphrey, Jeff LePine, Cheri Ostroff, and Rob Ployhart, among many others).

The MLT seminar served as a key energizer for the book that Katherine Klein and Kozlowski (2000) designed and edited, Multilevel Theory, Research, and Methods in Organizations: Foundations, Extensions, and New Directions. I want to highlight that pursuing the book as a means to advance MLT was a deliberate strategy. Given the controversies and debates that surrounded MLT and multilevel research in the literature (e.g., George & James, 1993; Yammarino & Markham, 1992), I believed that designing a book was the best way to present a coherent understanding of MLT. Klein et al. (1994) had published a perspective on MLT that was rooted in within and between analysis (Dansereau et al., 1984). We were friends and often attended multilevel symposia at conferences together, seeking enlightenment, and often being disappointed. Collaborating on an edited book to advance MLT was natural.

From my perspective, there are a few aspects of the book that are worth highlighting. First, the opening chapter was designed to present a set of basic principles that grounded MLT as a meta-theoretical framework to guide theory building, research design, and measurement for MLT-driven research (Kozlowski & Klein, 2000). I describe MLT as a meta-theory because, unlike most theories or models in OS, it is content free; it can be applied to any content domain. Second, we were very deliberate in designing the book’s chapters to cover a range of topics to which MLT could be applied. Notably, we did not have chapters on climate or leadership, which to that point had been the crucibles for developing MLT concepts. We wanted to demonstrate breadth of application to many other topics that, to that time, had not been subjected to multilevel theorizing. Third, methods chapters that covered aggregation issues (Bliese, 2000), multilevel analytics (Dansereau & Yammarino, 2000; Hofmann et al., 2000; James & Williams, 2000), and a head-to-head comparison among the different analytic approaches (Klein et al. (2000), dispelled much of the confusion about analyzing multilevel data that was evident in the extant literature at the time. Publication of the book, with its principles for MLT and measurement, several content exemplars, and guidance on aggregation and analytics coincided with rising interests in OS on team effectiveness. Multilevel research, often centered on teams, took off after the book was published and, more broadly, multilevel research in OS increased exponentially. The MLT book substantially enhanced the visibility and trajectory of my career.

Integration and Transition

At same time that I was pursuing the development of MLT, I was investigating topics that were related to my dissertation. The topics I was studying included implementing new technology systems in organizations (Chao & Kozlowski, 1986; Hattrup & Kozlowski, 1993), as a spin-off of updating climate, and newcomer socialization (Major et al., 1995; Ostroff & Kozlowski, 1992, 1993), as a way to study climate transmission and the process of learning and adaptation. Importantly, I learned early in my career that grant support was very useful for providing resources to advance research. My very first grant was a collaboration with Dan Ilgen (circa 1986), who had then recently joined MSU (replacing Ben Schneider, who had returned to the University of Maryland), on technology implementation. Several other internal grants followed that allowed me to pursue my research agenda focused on socialization, technology implementation, and upskilling (with J. K. Ford). Although the grants were modest in size, I learned the value of resources as an important aspect of advancing research and began to seek extramural support.

In the late 1980s, ongoing conflicts in the Persian Gulf resulted in two tragic events involving the U. S. Navy. In 1987, the USS Stark detected an unidentified Iraqi Mirage and queried it for identification, but did not take a defensive posture. It failed to detect the firing of two anti-ship missiles both of which hit the Stark killing 37 crew members. The following year, the USS Vincennes, engaged in skirmishes with Iranian speed boats, misidentified a civilian Iranian Airbus flight as a military aircraft. In the fog of engagement, the Vincennes fired two anti-air missiles which struck the Airbus, killing all 290 people aboard. The Vincennes was equipped with the advanced Aegis command and control system, and yet a simple misidentification error had cascaded through the system with tragic results (Bell & Kozlowski, 2011). These events resulted in significant interest in ways that team decision making (TDM), especially when it involved teams working through sophisticated technology systems, could be improved. I received two grants from the U. S. Army Research Office to provide guidance for leadership training to improve TDM, and another one (with J. Kevin Ford) to develop TDM training guidelines from the Naval Training Systems Center (NTSC). This latter grant was a small part of a much larger project, Team Decision Making Under Stress, that involved Eduardo Salas and Jan Cannon-Bowers at NTSC. ⁴ Many other grants across a range of funding agencies followed that focused on individual learning, regulation, and adaptation, and on team learning, development, leadership, and adaptation. My research had evolved to encompass individuals and teams, examined through a MLT lens, and with a decided focus on unpacking the underlying processes that were involved. I am making a point of this because the way OS studies processes has not substantially changed. It largely relies on construct approaches, which cannot unpack processes as dynamic phenomena. Herein lies the beginning of my second and complementary meta-theoretical journey (beyond MLT), which is to enhance theorizing and research on process phenomena. This is an enduring legacy of McGrath, as he was tireless in his efforts to build process theories of team functioning (McGrath, 1984, 1990; McGrath, 1991) and to castigate the field for ignoring process dynamics. But I am getting a bit ahead of the story, as my concerted effort to directly unpack process dynamics lies a few years further out.

Individual Learning and Adaptation

I know this essay is supposed to be about teams, but it is challenging to conduct meaningful research on teams if one does not have a solid grounding in the individual-level underpinnings of team-level phenomena. Most team phenomena are not purely collective (i.e., originating at the group-level directly). Rather, they are emergent. “A phenomenon is emergent when it originates in the cognition, affect, behaviors, or other characteristics of individuals, is amplified by their interactions, and manifests as a higher level, collective phenomenon” (Kozlowski & Klein, 2000, p. 55). Emergence is a key feature of MLT thinking. To advance understanding of how to improve team TDM, it was first necessary to figure out how to improve individual TDM, and then extrapolate theory to the team level.

A neat feature of grant support is that it gives one negotiating leverage. Space is always a limited resource, but the grant enabled me to acquire a small research space in which I could house four computer workstations. ⁵ For our initial research, we had acquired a tactical decision making computer task from NTSC called TANDEM (Weaver et al., 1995). It was limited in a number of ways, and we initially worked with NTSC to reprogram it to provide much more experimental control (i.e., ability to manipulate conditions and ability to access keystroke-level data). ⁶ This small computer lab was a precursor for developing an integrated theory-method-measurement paradigm that powered our research on self-regulation and complex skill acquisition for about two decades.

Our initial research using TANDEM assessed relevant individual differences, manipulated two different learning conditions, tracked the effects on a range of multidimensional cognitive, affective/motivational, and behavioral outcomes of learning (Kraiger et al., 1993) over time and their subsequent effects on performance adaptation (Kozlowski et al., 2001). At the time, the research was state-of-the-art. One insight I gained during the publication review process was that one reviewer (and the editor) were completely disinterested in the hypotheses and analyses that unpacked learning over time. They just wanted an overall snapshot of the model and did not care how the phenomena evolved to get to the end state. So, we scrapped those important research points and provided an overall path analysis that is the focus of the published version. This was an important critical incident that demonstrated to me how agnostic the field was to processes and change over time. It was one exemplar of many more to come.

I learned a lot from that initial effort, and those learning points were harnessed to develop an integrated theory-research-measurement paradigm, termed the Adaptive Learning System, to systematically study goal, feedback, and motivational factors that influence the process of self-regulated learning and performance adaptation (Kozlowski et al., 2001). By integrated, I mean that core theoretical concepts are central to the methodology of data collection and to the nature of measurement. The system allowed us to collect individual difference measures online well in advance, manipulate conceptually driven goal, feedback, and/or motivation conditions, compile keystrokes into meaningful indicators of learning behaviors, track learning over time, and assess the effects on performance adaptation. Self-regulation is a process of learning, motivation, and performance that—as a process—is a sequence of events-actions that unfold over time. A learning goal is set, attention and effort are directed to study necessary information, behavior is directed during practice to apply the knowledge, performance results, attention and effort directed during feedback influences subsequent study, and the process cycles until the goal is accomplished or the task terminates. That is a real process, not a retrospective perception of one typified by asking survey items. Most of the learning outcomes were measured by computer software, with some limited use of questionnaires to assess knowledge acquired and, sometimes, motivational perceptions like self-efficacy. A well-developed lab or simulation paradigm provides unique capabilities to systematically study focal phenomena in ways that are otherwise quite difficult to achieve.

An integrated theory-driven research paradigm is a very powerful tool for advancing understanding of a constrained theoretical space. The adaptive learning paradigm energized a stream of productive research on self-regulated learning, complex skill acquisition, and performance adaptation. Perhaps most notable is my collaboration with Brad Bell on goal orientation, active learning, and adaptation (Bell & Kozlowski, 2002a, 2002c, 2008, 2010; Kozlowski & Bell, 2006). The important lessons from this line of inquiry—self-regulatory processes as the underpinnings of learning, motivation, performance, and adaptation and the scientific value of a well-developed paradigm carried over as my work shifted to focus on teams.

Team Leadership, Learning, Development, and Adaptation

As a complement to the work on individual learning, I had started a stream of inquiry focused on team learning and development that was a synthesis of MLT, existing theory on team development, and self-regulated learning. Our initial efforts were conceptual. The grants I had at the time were focused on TDM leadership in team contexts; that is, leadership in volatile, uncertain, complex, and ambiguous (VUCA) situations. Although there was a substantial literature on leadership, there was very little leadership theory focused on teams or the dynamic VUCA situations that characterized TDM.

Fortunately, early on I had an opportunity to observe U.S. Navy teams training in a high-fidelity mission simulator of the combat information center (CIC) of an Aegis cruiser that was housed at the Combat System Engineering Development Site (CSEDS). Aegis, which is a shield in Greek mythology, is a complex command and decision system that integrates multiple sensors (e.g., phased array digital radar, sonar, satellite, etc.) and combat systems to protect a naval battlefleet. The personnel complement of an Aegis CIC included the Captain, Tactical Action Officer (TAO), and three sub-teams of approximately 10 people each—air, surface, subsurface—that worked to detect, monitor, and take action to defend the fleet.

The captain and CIC crew for a newly or soon to be commissioned Aegis cruiser were deployed to CSCDS to train prior to staffing their ship. The facility incorporated a high-fidelity reproduction of the CIC, including all the view screens, equipment, and crew stations. Training consisted of computer simulated missions two times per day. During the simulation runs, each crew member was accompanied by a dedicated coach/trainer. Importantly, although this was a training facility and there was an experienced staff to run it, ship captains had latitude to run the training as they wished, within the constraints of the simulation. Simulation runs were preceded by a pre-brief (i.e., description of the mission, rules of engagement, and current intelligence), the two- to 3-hour simulation run, and then a debrief or after-action review (AAR). Given the teamwork errors inherent in the USS Stark and USS Vincennes incidents, the simulation scenarios invariably included an impossible challenge. If you are a Star Trek fan, think Kobayashi Maru. Although these were simulated missions, they were based on real events, fully immersive, and highly engaging to the crews.

During my observations, I had an opportunity to view two captains with distinctly different leadership styles—one learning oriented and one performance oriented—evidenced by their behavior during the training that had clear implications for team learning. One captain was strongly interested in “winning” the simulation and exerted control of nearly every detail of the action. Although the TAO nominally directs engagements, this captain frequently barked orders during simulation runs directing team members to execute his commands with an emphasis on not “******* up.” Inevitably, by design, the CIC team would make a critical error during the simulation. During the AAR, this Captain would begin by pronouncing, in effect, “who messed up?” There were few takers, and the balance of the AAR consisted of the captain grilling the coaches about what happened and who messed up while the trainers responded as delicately as they could. Little team learning was evidenced over my 2 days of observation of this captain and CIC crew. The other captain monitored and engaged the crew as needed. During the simulation runs, he let the TAO manage the engagement, only stepping in to exercise “command by negation.” In other words, at times the captain would tell the TAO not to take a particular course of action, to delay it, or to consider a different action. In each instance, the captain would explain his intent and reasoning. He was monitoring his crew and, in effect, providing process feedback to prompt deliberation and action in real time. During the AARs, this captain would begin by describing something he had done incorrectly, reflecting on why it was incorrect, and would query the crew on how it could have been better. That simple model offered by the captain prompted a slew of reflective critical incidents across the crew. One could observe active learning taking place as the crew collectively reflected on their performance, what they did incorrectly, and how they could improve it next time. That crew substantially improved over the course of my observations.

My experience at CSEDS provided a range of critical incidents and insights for a stream of theory building that drew on MLT, conceptualized team leadership as a dynamic process of learning and team development, and focused on how teams build and compile adaptive capabilities. This was my initial effort to build theory (vs. targeted theoretical model building for research papers), and there were some points incorporated that I think were quite useful. First, we put leadership in a team context. There was very little material in extant leadership theory relevant to leading teams, my CSEDS observations, and the theoretical insights they provided. “The role of leaders in the development of the coordinated, adaptive, and coherent behavior of effective teams is not well articulated. Although there is a substantial literature on leadership in organizations . . . it is difficult to apply the prescriptions from this research directly to teams” (Kozlowski et al., 1996, p. 255). Leadership theory has generally been agnostic about the work and social context within which it is embedded (Kozlowski et al., 2016), preferring to take a generic approach that implies the theory is applicable across the board. Second, inspired by McGrath (1991), we viewed team leadership as a dynamic process in which the leader focused on coaching team skills during task cycles (i.e., preparation, action, reflection which model a regulation process of learning) and advancing team development to successively more complex team capabilities. Some leadership theories consider leader behavior contingencies across situations, but within situations leader behavior is static. We characterized two types of leader dynamics: (a) task cycles for targeted learning and (b) developmental progression for skill compilation. With respect to task cycles, we situated the team in a task environment that influenced team inputs, team processes that were aligned for resolving the complexities of those task inputs influenced outputs, and outputs feedback to shape the task environment and subsequent inputs to the team task. ⁷ The task cycle was extrapolated from my CSEDS observations as a way for team leaders to build teamwork skills. It consisted of a sequence of (a) preparation (i.e., set learning goals) prior to task engagement (much like a pre-brief), (b) monitor / intervene during task engagement (provide adaptive guidance), and (c) provide process feedback (much like an AAR), and repeat. ⁸ With respect to developmental progression, the leader could successively push the team through a series of phases to build adaptive skills. Third, inspired by Hackman (1988), McGrath’s former graduate student, we situated the theory as prescriptive (i.e., what team leaders should do) versus the more common descriptive (i.e., what leaders seem to do) approach. This necessitated a wide-ranging integration and synthesis of literature to articulate core process issues, extending well beyond the conceptual boundaries of leadership theory in OS.

This was a productive line of theoretical inquiry yielding the original paper (Kozlowski et al., 1996) that articulated the temporal aspects of the team task cycle and developmental progression dynamics; an application-oriented version with examples for implementation (Kozlowski et al., 1996); an articulation of the multilevel emergence—individual, dyad, and team network—involved in team learning, development, and adaptation (Kozlowski et al., 1999); and an updated and more detailed explication of the regulation dynamics inherent in the theory (Kozlowski et al., 2009).

In the late 1990s, I initiated a collaboration with Rick DeShon, funded by the Air Force Office of Scientific Research, that endeavored to apply self- and team-regulation process concepts to team learning and adaptation research. This research involved redesigning and reprograming TANDEM to become the Team Event-based Adaptive Multilevel Simulation (TEAMSim) and setting up a team simulation lab. Conceptually, we integrated concepts from multiple goal regulation (Rick’s area of expertise), dynamic resource allocation (adaptation to shifting task demands), and multilevel theory to develop a multiple goal, multilevel model of individual and team regulation (DeShon et al., 2004). During TDM task accomplishment, individuals have to execute their own role responsibilities, but frequently encounter pressures to shift to engage team responsibilities; that is, they pursue both individual and team goals. How team members manage multiple goal pursuit is often discretionary. Consider an academic researcher. One works to accomplish goals, but is often queried (by students, colleagues, etc.) for assistance. When one switches roles to assist a teammate, one cannot accomplish one’s own goals, although assistance to one’s teammate is good for the collective. How one manages that goal switching is a process of dynamic resource allocation. However, in TDM tasks the tempo is fast paced, time pressure is evident, team members have to coordinate synchronously, and the consequences of error are salient.

There were three learning points from this research. First, by inducing a multiple goal, multilevel action regulation process, we were able to empirically evaluate a multilevel homology of the regulation process that incorporated both individual level and team level tracks. Some conceptual multilevel homologies have been proposed in the literature, but full empirical evaluations with supportive results are very few in number. This allows generalization of some aspects of individual level regulatory processes to the team level. Second, in a reanalysis of this experiment and other data, Chen et al. (2009) demonstrated that team constructs, having emerged bottom-up from the individual level (DeShon et al., 2004), subsequently exerted top-down effects on individual behavior. It is a nice demonstration of the reciprocal influence of bottom-up emergence and top-down effects in teams. Third, the research design and analyses endeavored to get more directly at unpacking team process dynamics—they are there in the data—but the way we analyze data in IOP and OS tends to average over dynamic phenomena in favor of average effects. This was a major learning point for me, to which I will return a few years forward.

Virtual Teams and Leadership

During the 1990s, as team research was moving mainstream in OS, there was a lot of interest in advances in computer technology and Internet connectivity that were enabling teams to collaborate across geographic space and time. Taking a more technology focus, there was a new area of computer supported cooperative work (CSCW; Grudin & Poltrock, 2012). In the team effectiveness space, Townsend et al. (1996) coined the term “virtual teams,” which struck a responsive chord. Unfortunately, much of the thinking at the time treated virtual teams as a “type” of team (i.e., face-to-face vs. virtual teams) rather than as a more complex team that shared many similarities with other teams along with some unique features. ⁹

This is another point on which I took inspiration from McGrath, who created many very useful typologies or frameworks to organize research. ¹⁰ In this instance, I had been invited to participate in a workshop funded by ARI and organized by Stephen Zaccaro and Rich Klimoski and was inspired to think about team leadership in a virtual setting. Collaborating with Brad Bell, we developed a typology to distinguish how VTs ranged across different key dimensions and then extrapolated propositions for how team leaders should adapt to mitigate or leverage those factors (Bell & Kozlowski, 2002b). At the time, the typology was unique in advancing that VTs were not a distinct “type” of team, but rather varied on range of key “virtuality” characteristics relative to the team task structure. Importantly, we treated virtuality not as a categorical type, but as a composite construct based on spatial distance (proximal vs. distributed) and communication (face-to-face vs. technology mediated [rich synchronous vs. poor asynchronous]). That paper was influential in shaping the conceptualization of VTs. Subsequent work has gone on to distinguish how leaders need to employ structural supports to mitigate the challenges of VTs (Hoch & Kozlowski, 2014), craft an integrative review of VT research (Mak & Kozlowski, 2019), and provide principles to VT leaders for building the team, getting work accomplished, and leading for the long haul (Kozlowski et al., 2021).

Team Effectiveness

The theoretical and empirical research on teams that I had conducted was focused on learning, regulation, development, and adaptation. Although our theorizing and research was widely applicable, it was not universal. We contextualized phenomena to a targeted middle range and to particular types of teams and tasks (i.e., high reliability). However, the type of theory we were developing and research we were conducting was getting attention. Around the turn of the century, I was invited to write a handbook chapter on team effectiveness and later, because of the visibility of the handbook chapter, a monograph on team effectiveness for Psychological Science in the Public Intertest (PSPI). ¹¹ These broader review projects proved to be an interesting intellectual challenge. The extant literature touching on team effectiveness was voluminous, going back decades. It had its roots in small group research in social psychology, where it thrived and died, before migrating to organizational psychology and behavior (OPB) in the 1990s. As Levine and Moreland (1990) opined, “Groups are alive and well, but living elsewhere.. . . The torch has been passed to (or, more accurately, picked up by) colleagues in other disciplines, particularly organizational psychology” (p. 620). There were a few extant reviews, but none that did a really good job of organizing the literature in a way that provided insights. This is another point on which I took inspiration from McGrath, who created many very useful typologies or frameworks to organize research. In any case, I wanted to craft a reasonably comprehensive review for the handbook and understood that the value of the review would be substantially enhanced via a useful organizing framework. Brad Bell and I settled on a “lifecycle” organization because it enabled us to cover a wide range of distinct literatures that were relevant to team effectiveness but were not often aligned with it directly. I think we hit the target because that review (Kozlowski & Bell, 2003) and its follow on (Kozlowski & Bell, 2013) have been influential.

Not long after, I was invited to craft a monograph for PSPI. An unusual aspect of PSPI reviews is that they are intended to highlight firm conclusions and to recommend applications. I was encouraged to invite my colleague, Dan Ilgen, to collaborate as my co-author. This review necessitated a very different type of organizing framework—not one driven by topic content per se or time—but one that would highlight meaningful knowledge generation and document an empirical foundation for application. I have described how the organizing framework for that review was developed in detail elsewhere (Kozlowski, 2018), but this is another instance in which I used McGrath’s typology playbook. We started with an Input-Process-Output (IPO) heuristic (McGrath, 1964), albeit a dynamic one that incorporated a feedback loop, and focused first on identifying “processes” ¹² with meta-analytic evidence or multiple rigorous studies supporting a meaningful association with team effectiveness. Second, having identified a set of team cognitive, motivational/affective, and behavioral processes that were meaningfully related to team effectiveness, we then applied the same strategy to identify interventions—team design, team training, and team leadership—that were associated with the key team process constructs. Suffice it to say that the monograph (Kozlowski & Ilgen, 2006) was useful as it has been influential to other team effectiveness scholars.

Overview

There were a number of things going on in my career toward the end of the first decade of the 21st century. After 6 years as one among many Associate Editors, I was selected to become Editor for the Journal of Applied Psychology (incoming 2008, then 5 years as the Editor-in-Chief). That was a big event in my career, prompting reflection on where I was, where I wanted to go, and how I would accomplish my scholarly goals while leading a really large and important journal. My reflections yielded an epiphany about process dynamics. First, by then, I had been theorizing about and researching individual and team processes involved in learning, performance, and adaptation for nearly 15 years, but studying process was difficult and it always seemed that our research was a step or two removed from capturing processes and their dynamics directly (Kozlowski, 2015). Second, MLT had exerted substantial influence on research in OS; it was difficult to pick up a journal issue and not see one or more papers that incorporated MLT in the research. However, virtually all the multilevel research focused on top-down, cross-level effects using cross-sectional data. To paraphrase McGrath, we studied “multilevel statics.” Bottom-up, emergent phenomena and their dynamics were not examined—they were assumed theoretically to justify aggregation—but they were not studied directly (Kozlowski & Chao, 2012b). Third, more generally, virtually all theories in OS have at their core a theoretical rationale that endeavors to explain a process, but theory, hypotheses, and analytics are invariably about construct-to-construct relationships, not processes per se (Kozlowski, 2022; Kuljanin et al., 2024). If IOP and OS more generally were to advance, we needed to use different forms of theorizing, methodologies, and measurement tools to capture processes and their causes directly. Being a senior scholar by then, I decided that making progress toward resolving these issues was a worthy pursuit. Three funded projects resulted from those reflections that guided my work for the next decade and beyond. I describe the projects as translational science, applied discovery science, and theoretical science and innovative methods. They were distinctly different, but each contributed insights into capturing individual and team process dynamics directly. I have summarized these three projects elsewhere (Kozlowski et al., 2015), so here I provide a high-level description along with some insights.

Translational Science

The translational science project involved collaboration with an emergency medicine physician who also directed patient simulation and safety training for a regional trauma center. Emergency medicine involves teamwork at its core, although teamwork is not a deeply embedded aspect of physician training. My colleague’s role was to fill that gap via simulation-based experiences. She was interested in collaborating so as to transform the patient simulator into an experimental platform to conduct research on team and leader training. I consider this project translational because we were not probing the limits of knowledge regarding team effectiveness. Rather, we were applying known science on simulation design, measurement, and team effectiveness principles to improve medical team performance. The work was supported by the Agency for Healthcare Research and Quality. The project involved creating a set of simulation scenarios that would provide opportunities for targeted leader and/or teammate behaviors to be elicited (or not). A major aspect of the project was building a methodology and measurement infrastructure around the simulator to make it a useful research platform (Grand et al., 2013). Simulation runs were video and audio recorded for later coding. Once target behaviors were identified, a coding system had to be developed, coders trained, and coding conducted to translate recorded action into a sequence of behavioral codes relevant to the targeted behavior. Nonetheless, the project trained multiple grad students and generated application insights (Fernandez et al., 2013, 2017; Rosenman et al., 2018, 2021). The capstone of the project was a random controlled trial—the gold standard for medical research—conducted in a regional trauma center with real patients demonstrating that team leader training led to improved trauma doctor leader behavior and to improved patient care outcomes (Fernandez et al., 2020).

Although this approach provides rich process data in that one has a full recording of the behavioral action and communication, only selected snippets of those process data are translated into codes which are the actual data used for analysis. Moreover, this sort of research is exceptionally laborious across the board—designing and building the infrastructure (think years), collecting data for a study and coding it (more years), and analyzing the extracted data (months). Such research is not for the faint of heart. The approach is definitely useful as a way to gain some insights on targeted processes, but some form of automation would go a long way in making behavioral observation more useful.

Applied Discovery Science

Over 50 years ago, in 1969, NASA’s Apollo 11 landed on the moon and Neil Armstrong became legendary as the first human to set foot on the lunar surface. Just 3 years later, after five more missions, Eugene Cernan of Apollo 16 became the last human to walk on the moon. After an incredible decade long effort to reach out to space, the United States of America’s efforts to explore outer space retrenched to human missions in near-Earth orbit focused on the International Space Station. Lunar explorations and interplanetary missions to more distal worlds became the province of probes and robotic platforms. That changed during the early 2000s and NASA began preparations for long-duration human missions back to the moon and, later, interplanetary missions to Mars.

Space exploration is an inherently dangerous enterprise, especially to human physiology, so NASA had been studying how to mitigate risks to space crews related to biomedical and sleep concerns for decades. The prospect of a mission to Mars, however, raised new concerns around risks to team effectiveness. Consider the prospect of being on a mission to Mars with five other astronauts. You would be crammed into a small habitable volume with virtually no privacy, limited communications with anyone outside the mission, and under the persistent pressure of deadly threats. These conditions are characterized as isolated, confined, and extreme (ICE). What is the big deal? Astronauts have endured these conditions previously on lunar missions. Oh, yes, but this mission will take about 3 years, roughly 9 months to get to Mars and another 9 months to return, with the balance for exploration of the Martian surface. Right, 3 years of confinement in a small social world, subjected to persistent stressors, with little to no opportunity to cope (i.e., get social distance, socialize with family and friends, engage in non-work recreation). What could go wrong? Plenty, so NASA initiated a new program of human research focused on team functioning for long-duration ICE missions. I was one of the two initial PI’s, Ed Salas was the other, on NASA funded research to support team effectiveness.

I went into the research thinking that we would conduct a series of lab simulations with the TEAMSim infrastructure at my disposal (which would have allowed intensive longitudinal data collections to get at “processes” related to team cohesion and performance; see Braun et al., 2020), but quickly realized that NASA was not at all interested in lab research with ad hoc teams. They wanted research on teams “in the wild,” and not just any teams, but teams that were composed of astronauts or “analogs” of astronauts (i.e., highly trained professionals) in ICE settings that could serve as analogs for the rigors of space travel. Such people and settings are few and far between and are difficult to access.

The project pivoted around three foci. First, we initiated research on science teams deployed to the ice fields of Antarctica (real ice missions!) for about 6 weeks during the Antarctic summer and on teams housed in Antarctic stations for 9 to 12 months during “winter-over” missions. Using experience sampling methods (ESM), participants provided daily ratings on a range of “team process” constructs and personal reflections in journals. Second, using the same ESM approach, we initiated research on teams in NASA simulations that were analogs for asteroid transit (Human Exploration Research Analog; HERA) and Mars surface exploration (Hawai’i Space Exploration Analog and Simulation; HI-SEAS) missions. Both of these foci examined just one team at a time for short to long durations, but they generated unique and very rich descriptive data on the challenges of ICE team functioning. Third, working with an engineering colleague, we developed a team interaction sensor (TIS) technology that could track interpersonal proximity, motion, heart rate, and heart rate variability. The goal was to develop a technology platform that could—in near real time—assess team functioning, detect anomalies, and deliver feedback or a more active intervention to support team cohesion.

The NASA research foci spanned a decade. Project management was a challenge, as each data collection involved getting approval from multiple institutional research boards; coordinating with multiple investigators; collecting, curating, sharing, and archiving the data; and reporting, presenting, and explaining research findings to multiple stakeholders. The short story is that the TIS was validated as a device that could capture dyadic interactions, predict affective reactions, and predict a breakdown in team cohesion (Kozlowski et al., 2018). The technology was transferred to the NASA Wearable Electronics and Application Research lab in 2018. Unfortunately, we were not afforded an opportunity to develop the feedback system. Program managers move on, project priorities change. My research teams are now looking deeply into data we collected to begin unpacking the team process dynamics of conflict (Somaraju et al., 2022), conflict contagion (Somaraju et al., 2024), and ICE team functioning (Olenick et al., in press; Van Fossen et al., 2021). Nonetheless, the experience of developing, validating, and using the TIS in research was invaluable and has spawned new interdisciplinary collaborations to use technology to capture team process dynamics (Dudzik et al., 2018; Gedik et al., 2023; Zhang et al., 2018a, 2018b).

Theoretical Science and Innovative Methods

The third stream of research, and the one that I find most intellectually challenging, is focused on building (and testing) better theory. That necessitates computational theorizing and modeling (Kozlowski et al., 2013), which is a fusion of theory and methodology As I noted in the overview to this section, virtually every theory in OS is, at its core, interested in explaining a dynamic process. “Theory is about connections among phenomena, a story about why acts, events, structure, and thoughts occur . . .. Strong theory, in our view, delves into underlying processes so as to understand the systematic reasons for a particular occurrence or nonoccurrence” (Sutton & Staw, 1995, p. 378). However, theory almost universally focuses on predicting construct-to-construct relationships, which over time result from process dynamics, but relationships cannot capture processes per se. Processes are event-action sequences that are driven or caused by underlying process or generative mechanisms (Kozlowski, 2022; Kuljanin et al., 2024). Theory typically posits a narrative rationale that endeavors—for better or worse—to explain processes that is then extrapolated into hypotheses that predict construct relationships. Thus, when one finds support for a hypothesized construct relationship, note that those data do not provide direct support for the explanation because the process was not observed; the process rationale is speculative. The explanation is one-step removed from the underlying process, as an event-action sequence, and two steps removed from the process mechanisms; that is, the generative mechanisms that energize the process (Kozlowski, 2022; Kuljanin et al., 2024).

This line of inquiry, funded by the Office of Naval Research, began as part of a project focused on team knowledge building and decision making, which was my original point of entry to research team decision making. The basic problem is widely applicable. A team composed of different experts is confronted with a problem space that they have to resolve, oftentimes under time pressure and with significant consequences for error (i.e., high reliability teams). The experts learn/gather data relevant to their expertise, share it with each other to build common understanding, and then decide on a course of action. This is a very common TDM situation that is widely applicable across organizations. An analog of this decision-making problem known as the hidden profile (HP; Stasser & Titus, 1985) has been studied for decades. Essentially, in an HP task, commonly known (i.e., shared) and unique (i.e., to one person) information is distributed across the team. Meta-analytic evidence (Lu et al., 2012) shows that teams spend far more time discussing and incorporating the common information into their deliberations, often missing the unique diagnostic information that is “hidden,” thereby making suboptimal decisions. Note that in the typical HP paradigm, teams only have one experience, so there is no feedback or potential for learning.

At inception, the project core was a team knowledge typology (TKT) we developed that enabled us to measure with precision how information in a fixed pool of relevant information to the problem (i.e., a pool of common and unique information) was distributed across team members at any point in time (Kozlowski & Chao, 2012a). Validating the TKT using human teams would have taken years to gather sufficient data. Because of my exposure to other disciplines, I was aware of computational modeling (CM) and decided to use it to emulate teams of three experts learning and sharing information so we could generate data to validate the TKT metrics. Meanwhile, we were designing and programing a human task simulation to emulate this TDM problem space.

We implemented the CM, executed virtual experiments (i.e., experiments in silico wherein programed agents emulated human learners), and observed team knowledge building trajectories using the TKT metrics. That validation effort yielded a serendipitous epiphany. Graphs of the TKT metrics provided a clear picture of the process of team knowledge building as it unfolded over time. We were able to gain insights into the process directly, rather than trying to infer process from observing relationships. We learned to think in process terms rather than in relationship terms. Thus, beyond validation of the TKT, CM provided us with compelling insights into how the agents became bottlenecked or where there was friction in the team learning and sharing process. Once we had our TDM task constructed, we extrapolated the insights we gained from the CM effort to build interventions that were embedded in the task (i.e., embedded agents) to aid human team learning and sharing. That research, Grand et al. (2016), published as a Monograph in the Journal of Applied Psychology.

The insights that the project team gained from this research have been profound. I/we finally understood why even really well-designed simulation experiments (e.g., DeShon et al., 2004) could not get at process dynamics directly. Whereas, CM, which implements a set of agent generative mechanisms and then allows the agents to interact, provides compelling insights into the process drivers, the event-action sequences that are generated, and the construct relationships that can be examined by aggregating across process sequences. It covers a full range of explanation—generative mechanisms, processes, and relationships (Kozlowski, 2022)—in contrast to the dominant construct approach that provides prediction regarding relationships, but is speculative regarding the actual underlying processes and their causes. We have been sharing our insights in a series of papers that explain the limited ways in which OS views multilevel emergence and process phenomena (Kozlowski, 2012, 2015; Kozlowski & Chao, 2012a, 2018), how computational theorizing and modeling (CTM) can advance multilevel, process-oriented team research (Kozlowski et al., 2013), how to build a CTM paradigm (Kozlowski et al., 2016), how to use it to build a CT and provide plausible support for the theory using CM and linked human data (Grand et al., 2016), how one can use CTM to build better process-oriented theory and substantiate process explanations (Kozlowski, 2022; Kuljanin et al., 2024), and how CTM and help to build better organizational interventions (Braun et al., 2022).

The original project team remains together with a much more ambitious project goal. Our work, which is now supported by the Army Research Institute for the Behavioral and Social Sciences, is constructing a highly flexible CTM that will be capable of simulating an extensive, network-based, highly flexible team system so we can conduct virtual experiments examining team and system adaptation to shocks.

Multilevel Emergence

There are two primary forces that shape organizational systems, top-down cross-level effects and bottom-up emergent phenomena (Kozlowski & Klein, 2000). Yet, most of the research in OS focuses on top-down effects, most likely because it is tractable with cross-sectional data. Emergence as a process is virtually unexplored (Kozlowski et al., 2013; Kozlowski & Chao, 2012a). Emergence is particularly important in teams as it is essential to the formation of all the interesting team states that bear on team effectiveness. Emergent phenomena can manifest in two distinctly different ways. They can emerge via convergent composition processes in which individuals come to share a similar view of the phenomenon. Or they can emerge via divergent compilation processes in which individuals exhibit differences, but those differences create a pattern or configuration that constitutes a meaningful whole. Yet, when emergence is considered, it is almost always conceptualized as a composition phenomenon, and is treated as a theoretical assumption that is verified by assessing restricted within team variance on the construct in question. Emergence is not directly observed; it is an inference (Kozlowski & Chao, 2012a). We need to study emergence as a process directly and we need to consider it across the full spectrum of emergence types (Kozlowski & Klein, 2000).

Collective “Holistic” Entities Versus Differentiated Network Configurations

The principles of MLT were synthesized from over half a century of scholarship directed toward understanding how to think about multilevel phenomena and measure multilevel data based on construct-to-construct relationship thinking. That was the dominant approach at the turn of the 21st century and it remains so. But MLT and the analytics developed to deal with multilevel data tend to treat levels higher than the individual as undifferentiated. Most typically, a composition form of emergence is assumed, and individual level data are aggregated to the team level (following appropriate measurement and statistical justification, of course) to create collective variables that represent team constructs. This means that research typically treats teams as holistic collective entities.

Certainly, there have been some efforts to think about differentiated phenomena in teams. For example, Harrison and Klein (2007) highlighted different configurations—variety, separation, and disparity—that could characterize particular patterns of differences across members for teams. Although they framed their work under the concept of diversity, such configurations are directly relevant to compilation constructs (Kozlowski & Klein, 2000). Similarly, Gooty and Yammarino (2011) highlighted the scarcity of research on dyads and provided relevant analytic solutions. And, of course, research on group faultlines—the hypothetical fractures caused by group member differences—is an active area of inquiry (Thatcher & Patel, 2012). All these treatments tend toward developing a representation of differences that is holistic and collective. Yet, experience suggests that reality is lumpy.

For example, one of the key findings in the NASA project described previously is that long-duration teams in ICE settings typically started their mission with a high sense of team cohesion and dense social networks. Over time, however, as stressors accumulated, team cohesion tended to disintegrate and team members broke up into smaller dyads and triads. The collective became lumpy. Typical approaches to representing multilevel constructs in teams would not apply (i.e., because perceptions of cohesion became quite variable, there was no “team” cohesion, and ratings could not be aggregated to the team level). However, it was not the case that cohesion perceptions were purely individual level either. Rather, they depended on who team members were connected to in their dyads or triads. Network conceptualizations provide a means to conceptualize team phenomena, and how those phenomena evolve over time, in ways that capture complex forms of differentiation (Griffin et al., 2023; Mohammed et al., 2021). My colleagues and I will have more to say about this in a paper that is under development.

Process Theorizing: Computational Theory and Modeling

I view process theory versus construct theory (Mohr, 1982) as a necessary advance to enhance theorizing on team effectiveness and for better theory building in OS more generally (Kozlowski, 2022). In an insightful book, Explaining Organizational Behavior, Mohr (1982) contrasted what he termed “variance” theory and “process” theory. Variance theorizing, or what I have termed construct or factor theory, is focused on predicting variance in an outcome of interest. Although prediction is most certainly very useful, it is not a theoretical explanation for the why of a phenomenon, which is critical to having a sufficient theoretical explanation. This goes directly to my previous discussion, which indicated that construct theorizing is one step removed from the actual process of a phenomenon as a sequence of events-actions and is two steps removed from the underlying generative mechanisms (i.e., drivers or causes) of process dynamics. Rather than focusing on construct relationships, process theorizing is focused on process dynamics and, importantly, on the generative mechanisms that are the drivers of the process (Mohr, 1982; Pentland, 1999). One can liken the difference between construct and process theory as the difference between focusing on factors vs actors, respectively (Macy & Willer, 2002). CTM allows a theoretical focus on generative mechanisms, links the mechanisms directly to process event-action sequences, and even to construct-to-construct relationships to the extent that they are of interest (i.e., often as a means to verify CTM findings with real world data).

I view CTM as a critical way to advance process theorizing with respect to team effectiveness. In my opinion, CTM is on the fringe of OS much like MLT was on the fringe of OS in my early career. I think CTM is gaining traction and will be mainstream within 5 to 10 years. Construct theorizing will not disappear anytime soon, but CTM will add capability for building better theory and it will force advances in methods to better evaluate process theory.

Methodology and Measurement

At the dawn of small group research in social psychology, researchers were directly interested in group dynamics. Classic research by Bales (1950) and Sherif (1956) observed and coded group members’ interpersonal interactions and behaviors. As I described previously, even with today’s advanced tools (i.e., video and audio recording, coding software) such research is laborious and research on small groups shifted quickly to the use of questionnaires to capture retrospective reports of group members’ perceptions of group processes. Constructs became proxies for group processes, and the situation has remained much the same until quite recently. To put this in perspective, the technology to conceptualize constructs, measure them as latent variables, examine mean differences among them, correlate them, or regress them was in place by the early 1920s, essentially one century ago (Kozlowski, 2022). Although OS measurement and analytic methods are much more sophisticated today, they are still based on techniques that are a century old. It is time to move forward.

I have been fortunate in my career to be exposed to ideas, tools, and techniques that originate from other disciplines that can advance theory and research on the science of team effectiveness. The TIS technology we developed with a computer engineer is a good case in point. Although technology is not a panacea and basic measurement development and validation are still important concerns (Chaffin et al., 2017), existing, emerging, and over-the-horizon technologies will enable team researchers to return to the dawn of small group research and capture team interaction process dynamics directly. These new measurement capabilities and the analytics needed to make sense of them are largely in the province of computer science. There is a substantial and active community of computer scientists working to create automated technologies that capture team interaction processes. Some years ago, when I was presenting the NASA project at (appropriately) an Interdisciplinary Network for Group Research (INGRoup) conference, I met a computer scientist who was interested in our work and we established a collaboration. Our current project is a handbook, Computational Group and Team Dynamics, that is intended to help forge a new discipline that integrates social science theory and research with computer science measurement technologies, techniques, and analytics applicable to capturing group and team dynamics (Kozlowski, Hung, Lehmann-Willenbrock, & Salah, forthcoming). We are on the cusp of new and exciting developments that will advance the science of group and team effectiveness. Stay tuned!