14

GROUPS AND TEAMS IN
ORGANIZATIONS

Studying the Multilevel Dynamics
of Emergence

Steve W. J. Kozlowski

MICHIGAN STATE UNIVERSITY

One of the biggest challenges of conducting insightful, informative, and impactful research on groups is effectively dealing with the multiple levels that comprise group phenomena. That is, group phenomena entail multiple levels of theory and constructs, measurement, and data analysis. What are you talking about, you may say. I want to study groups, that is the level I'm interested in, why should I care about other levels? My answer is, because the world is complex. Even the most “bare-bones” group situations entail a minimum of three levels: group (i.e., between group), individual (i.e., person within group), and time (i.e., within person over time). There are potentially other levels as well. Groups could be embedded in an organizational system with many higher, nested levels. One could be interested in dyadic relations embedded within groups, including families and work teams. You may not be interested in all these levels, but they nonetheless merit consideration in your theorizing, research design, and measurement.

First, groups typically exist in a broader context or setting. I am an organizational psychologist who is interested in team learning, problem solving, and effectiveness, so the groups I study are often in an organizational setting where they are subject to different contextual influences. These influences could include different types of leaders, exposure to different forms of training, or being in work units that use different technologies and administrative structures, among many other potential differences. Community psychologists might be interested in groups that exist in different neighborhoods or who have access to different services. Social psychologists might be interested in how influential group members, leaders, or other factors can shape group identities. Communication researchers might be interested in differences in group behavior and outcomes based on whether members discourse face-to-face or via virtual media. Finally, any laboratory study that manipulates conditions for groups in their research design – as I often do – has manipulated the context with the intent of observing differences across groups (i.e., between group differences) in those manipulated conditions.

Second, groups are comprised of individuals who differ on a variety of personal characteristics (e.g., cognitive ability, personality, interests, demographics, etc.). Collectively, this creates a group composition that can range from very homogeneous to heterogeneous. In addition, group members are exposed to the same contextual factors, share experiences, and interact with each other. Those common experiences and interactions may lead to very similar perceptions, feelings, and reactions, so that there is little within group variation (i.e., the data are dependent; see Kashy & Hagiwara, this volume). Such phenomena are collective. On the other hand, personal characteristics or other factors may lead people to have very different perceptions, feelings, and reactions such that there is a lot of within group variation. Such phenomena are individual level; differences lie between people.

Third, most psychological phenomena that are relevant to individual behavior within groups and to collective group-level behavior emerge over time. This within-person variation can be of interest in its own right. For example, experience sampling studies often examine how individual variation in mood over time can yield between-person differences in affect. Team researchers are interested in a variety of team process constructs that emerge over time (Kozlowski & Bell, 2003). For example, team members may share their reactions to contextual factors and, over time, converge on a shared perception of team climate or atmosphere (James & Jones, 1976), team cohesion, or their collective efficacy. The point is that many phenomena of interest in groups emerge over time from exposure to common events and group member interaction. Time needs to be considered theoretically and it has implications for research design. This is important because cross-sectional designs in the field and laboratory still dominate the groups and teams literature, but very few group phenomena are truly static.

The goal of this chapter is to provide an overview of multilevel theory (MLT) and methods, and to convince you that the principles of MLT have important implications for your theory building, research design, and measurement approach. As the contents of this volume makes clear, it is certainly not the only methodological consideration, but it is a very versatile framework that can help you to design and conduct cutting-edge research on groups. This chapter will highlight key conceptual and methodological challenges in studying groups from a MLT perspective. Although the primary focus of this volume is on methods, effectively using multilevel (ML) methods necessitates that you have a basic understanding of core theoretical principles. Research design and measurement are conditioned on the nature of the theoretical model you are evaluating.

I will begin with a brief overview of the development of MLT in organizational psychology, which is how I trace its application to group research. The primary MLT focus is on individuals, nested in teams, open to higher level contextual influences, with phenomena unfolding over time. I will then discuss key theoretical principles for understanding primary linkages across levels of analysis and the implications of these principles for measurement and research design:

• (a)   the effects of higher-level contextual factors on lower-level entities (i.e., individuals and groups; cross-level effects);
• (b)   the way that dynamic interactions among individuals can coalesce, emerge, and manifest over time as higher level phenomena (i.e., emergent phenomena, as well as within-level process dynamics over time); and
• (c)   how some phenomena (e.g., antecedents, mediating processes, outcomes) can exhibit parallelism across levels of a system (i.e., multilevel phenomena).

One needs to begin with a solid, focused, and well-developed theoretical model. However, if one cannot realize the model with measures that appropriately represent constructs, then the model cannot be effectively evaluated. ML research typically assesses constructs at one level (frequently the individual level), but desires to have those measures represent a different (usually higher) level. Ensuring that the levels of origin (where constructs are fundamentally conceptualized), measurement (where data are assessed), and representation (where constructs reside in a specific model) are properly aligned requires careful research design. In addition, one of the complexities of ML research is that with constructs represented at different levels of analysis, researchers need to devise data collection designs that will yield sufficient variance for each of the constructs at its level in the model.” Business as usual” approaches to data collection are often inadequate to accomplish this and the wise researcher needs to plan this aspect of their project with care.

Finally, I do not address ML analyses because a detailed treatment is beyond the scope of this chapter, other authors in this volume address the topic (e.g., Kashy & Hagiwara, this volume), and there are excellent primers in the literature (e.g., Bliese & Ployhart, 2002; Hofmann, Griffin, & Gavin, 2000). However, it is important to note that problems with theory or methodology cannot be resolved by analyses. That is why theory (first) and methods (second, to mesh with theory) have primacy. I will also note at the onset that the material to be covered is abstract, although I will illustrate key points with examples. Being an organizational psychologist, my examples are about work teams; this is what I know. Hopefully, the reader can map the examples by analogy to your own research interests. An open mind and a spirit of conceptual adventure are essential!

The Nature of Work Groups and Teams: The Need for
a Dynamic, Multilevel Perspective

Work groups and teams are like informal groups in that social interaction is a central aspect of group experience, but they also differ from informal groups in several important ways that have profound effects on team processes and outcomes. These important differences can be summarized as the context, the task, and the interdependencies that arise from each. These differences are captured in the definitions of work teams and in the way teams are conceptualized as embedded in dynamic and interactive systems. Work groups and teams:

The organizational context, team task, and interdependencies necessitates a conceptual approach that can capture the multilevel, dynamic, and emergent aspects of team functioning. The approach advanced by Kozlowski and Ilgen (2006, p. 80) conceptualizes teams

An Abbreviated Overview of Multilevel Theory

Precursors

My own interest in MLT was initially sparked when I was a graduate student and read a book entitled Building a Multidisciplinary Science of Organizations by Roberts, Hulin, and Rousseau (1978). It provided a succinct description of the fragmented nature of the organizational sciences (e.g., human factors psychology, organizational psychology, social psychology, sociology, management, etc.), presented a rudimentary framework for thinking about behavior as multilevel in nature, and discussed the data challenges involved. I was intrigued by this early call to move beyond the systems metaphor. My dissertation research incorporated both individual and higher-level contextual factors that influenced perceptions of the climate (i.e., work environment) in an effort to understand behavioral outcomes (cross-sectionally and over time; Kozlowski & Farr, 1988; Kozlowski & Hults, 1987).

A chapter by Rousseau (1985) provided a thorough scholarly treatment of levels of analysis challenges that, in the “early days,” was the primary reference for ML research. A subsequent paper by Klein, Dansereau, & Hall (1994) helped to move interest in ML research towards mainstream awareness. In addition, the topic of climate in organizations, which distinguished between individual climate perceptions, or psychological climate, and a shared organizational climate (James & Jones, 1976), provided a systematic focus for research that eventually provided a foundation for the development of ML theory and methods (James, 1982; James, Demaree, & Wolf, 1984, 1993; Jones & James, 1979; Kozlowski & Hattrup, 1992; Kozlowski & Hults, 1987; Schneider & Bowen, 1985). By the mid to late 1990s, Katherine Klein and I planned and edited a book on ML theory, research, and methods (Klein & Kozlowski, 2000). Our goal was to extend ML theory and to push the ML perspective solidly into the mainstream of organizational psychology, organizational behavior, and human resources research. In particular, we sought to cut through the (then) confusion in the literature to create principles to guide theory development, measurement, and research design (Kozlowski & Klein, 2000). Much of what I will describe in what follows is based on that treatment. A basic primer on ML research, also based on Kozlowski and Klein (2000), is available in Klein and Kozlowski (2000).

The Influence of General Systems Theory

The roots of MLT are in General Systems Theory (GST; von Bertalanffy, 1968, 1972), albeit MLT is focused on human performance in organizational systems, which is bounded – individual, group, organization, and time – whereas the goals of GST are much broader. GST was intended to establish principles that generalize across phenomena, qualitatively different systems, and even scientific disciplines as a means to promote the unity of science. The orientation of GST is holistic, in that “the whole is more than the sum of the parts,” and it sought to counter-balance the tendency toward reductionism that was viewed as prevalent in “normal science.” There are several key principles from GST that have been important in the development and methods of MLT. Isomorphism is a principle of identity or similarity that has been applied in the conceptualization and representation of parallel constructs at different levels of analysis.Functional equivalence is a principle whereby a construct or a process linkage (i.e., a relationship between constructs) fulfills a similar role in a model or system at more than one level of analysis. Logical homology is a principle that there are phenomena whereby analogous constructs and linking processes hold at different levels or in qualitatively different systems. In essence, homology is a combination of the first two principles to envision constructs and processes that form parallel multilevel models. These principles are critical to establishing generality from one level of a system to another level of that system, or to a different system.

Although GST has had an important influence on theory in the organizational sciences throughout the twentieth century, the influence has not always been positive. Organizations as nested systems of individuals, groups, and subsystems in dynamic interaction over time are incredibly complex and impossible to grasp in their entirety. Thus, holism as a fundamental assumption has been a limiting factor. Indeed, the press for holistic thinking has had the opposite effect. Because the system could not be meaningfully bounded and disentangled, it was instead fragmented into level slices that represented core disciplines of organizational science (Roberts et al., 1978).

Conceptualizations of system behavior that have evolved from GST, such as complexity science, self-organization, and chaos theories, focus more on how simple elemental interactions over time can yield very complex behavior at the collective (system) level. The focus is on the emergence of complex system behavior, but the focus on the basic elements, entities, or agents is not reductionism. Rather, it is an effort to understand how the “wholeness” arises without reifying it. It is an effort “ … to understand the whole, and keep an eye on the parts” (Kozlowski & Klein, 2000, p. 54). The goal of MLT is not to understand the system as a whole, but to decompose it selectively in meaningful ways to capture complex links at multiple levels. As I noted in the introduction, this is directly relevant to group researchers because group research inherently implicates at least three nested levels of analysis.

How Systems Are Coupled Across Levels:
Types of Multilevel Linkages

There are three primary types of ML relationships: contextual, emergent, and multilevel. I have illustrated simple models of these relationships in Figure 14. 1. It is important to recognize that these are basic types of multilevel relations. As a theorist, your model may combine these exemplars in myriad ways to craft a model that captures the group phenomena of interest. Contextual or top-down relationships are important because the setting in which an entity (i.e., a person, group) is embedded, if it is strong, will influence, constrain, and shape the entity. Situational strength trumps personality. Emergent or bottom-up phenomena are important because many team processes have their roots in individual cognition, affect, and behavior but, through social interaction and exchange, can take on collective properties. Constructs such as team efficacy, team cohesion, and team climate are emergent in nature. Indeed, a wide range of team phenomena are emergent. Being able to study such constructs requires methods to represent the constructs at the higher level of analysis appropriately. Multilevel relationships come in many forms. The classic homologous multilevel model captures a phenomenon that consists of parallel constructs and linking processes at more than one level of analysis (e.g., goal effects on individual and team performance). However, complex multilevel models can also involve a mix of contextual, emergent, and parallel processes. I will discuss each type in turn. Because studying emergent constructs also entails some of the thornier conceptualization and measurement problems, I will address those issues in that section.

FIGURE 14. 1 An illustration of contextual (cross-level), emergent, and multilevel models. (Note subscript i signifies the individual level; subscript t the team level.)

Contextual effects

Contextual effects are top-down, where factors at a higher level influence or constrain phenomena at the lower level. That is, they exert an effect that is cross-level. The hierarchical nature of social organizational systems is characterized by a nested structure. Individuals are nested in teams, teams are nested in subsystems, and subsystems are nested in the broader organization. Each level above is an embedding context for the level below that can exhibit potent effects on the lower levels. For example, the extent to which a control structure is more centralized vs. more decentralized will influence group and individual perceptions and behavior.

There are two kinds of possible cross-level linkages. One type of linkage is a direct effect in which a contextual construct at the higher level crosses down to account for variance in a construct at a lower level of analysis. For example, the nature of work unit technology (i.e., whether it is more complex or simple) can influence individual perceptions about the nature of their job (whether it is enriched or constraining, respectively; Kozlowski & Farr, 1988; Rousseau, 1978). Different types of leaders can influence group member feelings or how satisfied they are. In a now classic study, Lewin, Lippit, and White (1939) showed that different types of leadership climates influenced individual attitudes, so that “democratic” climates were associated with positive attitudes and “authoritarian” climates were associated with negative attitudes. These examples illustrate a simple cross-level direct effect.

The other kind of cross-level linkage is a moderating one in which a construct at the higher level changes the nature of a bivariate relationship at the lower level. For example, the well-established relationship between general cognitive ability and job performance at the individual level necessitates complexity and the discretion to utilize one's ability. Thus, a less centralized, formalized, and standardized structure is necessary to enable the relationship between ability and performance to hold. In highly structured contexts, because the context limits the discretion and the utility of ability, the relation is likely to decline or disappear (Hunter & Hunter, 1984); unit structure moderates the lower level relationship. Similarly, other higher level factors that might influence one's motivation to apply ability, such as leadership or team cohesion, could also evidence cross-level moderation.

Cross-level relationships – direct and moderating – tend to be most heavily represented in the literature, because there are well-developed exemplars, consensus on techniques for assessing contextual factors, and well-tailored analytical systems (e.g., multilevel random coefficient modeling; MRCM) that specifically examine these relationships (i.e., intercept differences for direct effects and slope differences for moderating effects). Because work groups and organizations are inherently hierarchical, research on contextual effects is important and useful. One key limitation is that much cross-level research relies on cross-sectional data, so that the (implied) causality inherent in most models cannot be examined. Moreover, cross-level models often assume a process of emergence of constructs from the lower level (e.g., team cohesion, team efficacy, team mental models), but rarely actually examine emergence as a process. It is the result of an assumed process of emergence (i.e., the manifestation of a collective construct) that is examined. Thus, one scientific challenge is that the attention devoted to cross-level relationships means that emergent processes that are also important get much less research attention.

Emergence

The nature of emergence

Emergence is the result of bottom-up processes whereby phenomena and constructs that originate at a lower level of analysis, through social interaction and exchange, combine, coalesce, and manifest at a higher collective level of analysis. “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). There are many exemplars of emergent constructs, what Marks, Mathieu, and Zaccaro (2001) describe as “emergent states,” that are used as indicators of team processes in the literature such as team learning, team mental models, and team effectiveness (Kozlowski & Ilgen, 2006).

This view of emergence is rooted in complexity science, where the idea is to understand how complex, system-level phenomena can be produced by lower level entities operating according to a few simple rules or principles. It is not reductionism. Rather, it is an effort to understand the system not as a collective “whole,” but as a phenomenon that emerges from the dynamic interactions of its behaving entities.

Craig Reynold's (1987) simulation of flock behavior in birds is a very elegant illustration of this perspective for modeling and understanding the dynamic interactions among individual entities that undergird complex collective behavior. In his simulation, “boids” are computer agents that simulate the motion of birds in a flock. Each boid is programmed to optimize a few simple rules: (a) separation – move away from other boids that are close to avoid collision; (b) alignment – fly in the average direction of the flock; and (c) cohesion – approach the center of the flock and avoid exterior exposure. Collective flock behavior is simulated by each agent maximizing the rule set in dynamic interaction with the other agents. As the simulation runs, the set of boids move essentially at random and, as they encounter other agents, the rules are applied dynamically. These three simple rules produce clumps of boids that flock together (see Figure 16. 7 in Flake, 1998). The addition of one more rule, (d) view – move to the side of boids blocking the view (Flake, 1998), yields the classic V-formation of a migrating flock of birds (see Figure 16. 9 in Flake, 1998). There is a compendium of useful information on Craig Reynold's website (http://www.red3d.com/cwr/boids/) and there are a variety of implementations of boids on the internet. A simple search will bring them up for your viewing pleasure and fascination. What is important is that the simulation provides a very palpable example of the emergence of a group-level phenomenon from dynamic individual interaction. Many analogous phenomena emerge in teams such as learning, coordination, and performance.

Computational modeling and simulation (the boids) are just beginning to make small inroads in organizational behavior research (Ilgen & Hulin, 2000; Larson, this volume; Tschan & Semmer, this volume). Meanwhile, we model most team phenomena by using measures that are designed to capture cognitive, affective– motivational, and behavioral constructs representing important aspects of team functioning (Kozlowski & Bell, 2003) that have emerged from the individual to the team level. How team constructs are represented is determined by the way or form in which they emerge. Different forms of emergence have implications for measurement and representation.

Forms of emergence

For the sake of keeping the discussion straightforward from here on out, there are a few caveats to note. First, I will focus on individuals as the lower level and teams as the higher level entities. However, it is important to note that the theoretical principles I describe apply to any other coupled lower and higher level relations. Second, I am describing phenomena that have emerged across levels. Thus, they originate in individual cognition, motivation–affect, and/or behavior, but through interaction emerge to manifest as team-level constructs. Representation at the team level has to be consistent with the form of emergence. Third, and this is important, it is the manifestation that is being represented, not the actual process of emergence. I shall touch on this issue more later. Fourth, there are obviously characteristics (e.g., group size, team function) or constructs (e.g., team diversity) that only have meaning and representation at the team level. Kozlowski and Klein (2000) describe these as global properties and I will not dwell on them further.

There are two “ideal” forms of emergence – composition and compilation – that represent distinctly different ways that a team construct can emerge from the individual level. One can think of these types as anchoring the ends of a continuum, with different emergent forms distributed between them (for more elaboration, see Kozlowski & Klein, 2000, pp. 52–77). As illustrated in Figure 14. 2, each form of emergence has a different underlying theoretical model. Composition is based on the principle of isomorphism, where essentially the same construct exists at the individual and the team level (i.e., it has the same structure and function across levels). Composition captures emergent phenomena that are convergent. Compilation is based on the principle of discontinuity, where the construct has the same meaning and function across levels, but it is not measured or represented using essentially the same content (i.e., it is not structurally similar). Compilation captures emergent phenomena that are configural.

FIGURE 14. 2 Characteristics of composition and compilation forms of emergence.

At one end of the continuum, composition emergence describes phenomena that combine, converge, and coalesce into a shared team property. A team mental model (Cannon-Bowers, Salas, & Converse, 1993), for example, is a shared property. Individual team members develop cognitive representations of their task, equipment, and how it is connected to other members’ tasks. However, since individual team members are exposed to the same situations and task, and are prone to sharing their views, over time members come to agree on their team mental model and variation among individuals is reduced as team members share the same cognitive representation. This shared mental model allows them to coordinate implicitly, since they have a common understanding of how to accomplish the collective task.

The underlying theoretical principle is isomorphism in that the construct exhibits structural and functional similarity across levels (Morgeson & Hofmann, 1999). That is, individuals perceive the same elemental content (e.g., team mental model perceptions), it is shared in common, and it carries essentially the same meaning and function in a model at both levels of analysis (Kozlowski & Klein, 2000).

At the other end of the continuum, compilation emergence describes phenomena that are variable, patterned, or configural across members. Transactive memory (Wegner, 1995), for example, is a configural property. Because individual team members possess different types of expertise or hold different roles, they specialize in tracking information relevant to different aspects of the overall team task. Unlike the mental model example described previously, team members do not share all the same knowledge. Rather, they each possess distinct knowledge that can be accessed by members knowing “who knows what.” In that sense, transactive memory is distributed across individual members in a pattern that is analogous to a network of memory nodes or pieces of a puzzle (Kozlowski & Bell, 2003).

The underlying theoretical principle is discontinuity in that the construct exhibits functional, but not structural, similarity across levels. That is, elemental content at the individual level is different (i.e., different knowledge or expertise), but the configuration and access to that knowledge at the team level is functionally equivalent to individual memory (i.e., it performs the same function at both levels of analysis).

Aligning the levels of construct origin, measurement,
and representation

Levels of conceptualization and measurement

The level of origin for a construct is where it fundamentally exists. The level of measurement is the level at which data are collected to assess the construct.

The level of representation is the level within a theoretical model where a construct resides. The world is simple when one is interested in, say, cognitive ability. It originates at the individual level, it is assessed by having individuals take an intelligence test, and it is generally examined in models where individual ability predicts individual performance on intellective or cognitively loaded tasks. Origin, measurement, and representation are aligned.

Things become more complicated if one is interested in, say, team mental models (TMM) and their influence on team performance. TMMs originate at the individual level, but they need to be represented at the team level (predicting team performance) in a model. How does one make this leap? At what level should the construct be assessed?

Aggregation

Many (most even) team constructs are fundamentally rooted in individual cognition, affect, or behavior. Based on the theoretical principles of emergence, higher level constructs that originate at the individual level should be assessed at the level of origin. So, in the TMM example, individuals within teams should also be the level of measurement. Okay, fine, but how does one then bridge the levels gap for representation? In the previous section I described how TMMs emerge via a composition process. Since the process of convergence typically takes some time, groups should be mature enough for emergence to have occurred (note that we are not examining the hypothetical process of emergence, but rather the manifestation of the process). The researcher then has to establish that team members exhibit sharing on TMMs. This is typically accomplished by examining consensus (i.e., interrater agreement) or consistency (i.e., interrater reliability; see Kozlowski & Hattrup, 1992), or both. The basic idea is to justify aggregation of the individual level perceptions to the team level by establishing empirically that variance within teams is restricted. Evidence of restricted within-team variance is an indicator of sharing, thus it provides empirical support for the underlying model of composition and evidence for the construct validity of the aggregated measure. With that support in place, we can now aggregate the individual-level TMM measures to the team level. Typically, investigators use the group mean as the team level representation.

Consensus (i.e., interrater agreement) and consistency (i.e., interrater reliability) are conceptually related, but they are not interchangeable and investigators are advised to be aware of the differences when selecting an approach: Bliese (2000) and Kozlowski and Hattrup (1992) address this issue in detail. Consensus is generally indexed using r_wg (James, Demaree, & Wolfe, 1984, 1993; or one of its later-day variants), which indicates the degree of within-team agreement. High agreement provides evidence that the construct is shared.

Consistency is indexed using the intra-class correlation coefficient (ICC). There are two types of ICC and they tell us different things about consistency in the group. ICC(1) provides an indication of data clustering or nonindependence (i.e., the extent to which group membership exerts an effect on individual member responses). A significant and ‘reasonably large’ ICC(1) indicates that team members are essentially interchangeable and provides evidence supporting sharing. ICC(1) provides evidence to support aggregation. ICC(2) provides an indication of the reliability or stability of the group mean. A high ICC(2) can give us information regarding how confident we can be in the results of analyses. Prevailing practice is to accept r_wg and ICC(2) values using “rule of thumb” thresholds (i.e., values in excess of 0. 70). ICC(1) values are more difficult to gauge by magnitude, although there is advice in the literature (e.g., Bliese, 2000). Some of the advice suggests that ICC(1) values in the neighborhood of 0. 10 might be acceptable. Although the details are too complex to address in this brief essay, I would be wary of ICC(1) values lower than 0. 30 and would prefer to see them well above 0. 50. Remember, ICC(1) is an indication of the effect of group membership, so values below 0. 50 are indicating that most of the variation is within groups, not between them. That is not strong support for aggregating data.

There is some inkling that item referents may play a role in agreement (Rousseau, 1985). Chan (1998) distinguished between aggregated measures based on items with a self-referent (“I perceive”), what he labeled as direct consensus, and aggregated measures based on items with a team referent (“My team perceives”), which he labeled referent shift. There has not been a great deal of research on this issue, but Klein, Conn, Smith, and Sorra (2001) reported that scales composed of team-level referents generally yielded higher agreement compared to scales with self-referents. I would add that my experience is consistent with this finding. So, if you are using individuals to assess team- or group-level constructs, you probably want to use the referent shift approach; I do.

Finally, what do you do if you do not find enough agreement to aggregate your data to the team level? In the mid-1990s Ken Brown and I were expanding some of the data simulations used by Kozlowski and Hattrup (1992) to see what the effects on agreement might be. One thing I was interested in was how different types of within-team social processes (e.g., leader–member exchange) could fragment a group and create really unusual patterns of within-team variance. We discovered that group polarization yielded observed variances that were much larger than would be expected from individuals responding in a purely random fashion. On that basis, we proposed that rather than using agreement as a criterion for aggregation that it instead should be treated as a construct (or more correctly, a phenomenon) and examined in its own right (Kozlowski, Brown, & Hattrup, 1996) since it could moderate the effect of the aggregated measure. In other words, when the variance is restricted, hence meaningful, the aggregated measure would have effects whereas when it was not restricted by group membership, the effect would be attenuated or reduced in magnitude. We later formalized this conceptualization as a dispersion theory (Brown & Kozlowski, 1999). Subsequent applications of this idea in the area of group climate have been efficacious (e.g., Gonsalez-Roma, Piero, & Tordera, 2002; Schneider, Salvaggio, & Subirats, 2002).

Multilevel phenomena

Multilevel homology

Early on in this chapter I noted that an important goal of GST was to identify homologies – essentially analogous phenomena – across different levels of a system or even across qualitatively different systems. If one can identify homologies, then one can generalize a knowledge base that exists at one level to another without having to replicate all the basic research. It is an appealing prospect for leveraging knowledge. Perhaps one of the best examples of an effort to apply this logic is that of J. G. Miller (1978) who sought to construct a system of logical homologies transcending the subatomic to universal levels. It is an impressive exercise in scholarship, but by necessity a bit too abstract and general to be widely applicable as a knowledge generating framework.

In any case, the goal of MLT in organization science is more modest, but has the potential to be very powerful (Kozlowski & Klein, 2000). The focus is on identifying parallel phenomena in organizational systems. Until very recently, this exercise in multilevel thinking was confined primarily to conceptual exercises. One very nice and oft-cited example is the threat-rigidity model proposed by Staw, Sandelands, and Dutton (1981). In this model, the authors drawn on theory and research addressing how individuals become rigid in response to stressful inputs and generalize those principles to postulate how groups and organizations react. In other words, they extrapolate research at the individual level to build parallel propositions for how collectivities would react. Although analogy is a useful way to build multilevel theory, the real value of multilevel modeling is empirical evaluation so one can be confident in the generalization. Phenomena do not always replicate across levels. How can these sort of relationships be modeled with data?

With the turn of the new millennium, the advent of a more coherent approach to MLT and research (Kozlowski & Klein, 2000) allowed multilevel theorizing to develop an empirical foundation to begin actually to apply the logic beyond conceptual exercises. The basic problem that needed to be solved was how to capture a parallel phenomenon simultaneously at more than one level of analysis. There were methods in place, such as within-and-between-analysis (WABA), which allowed relations to be examined at multiple levels, but not simultaneously. Rather, WABA determined at what level a set of relations resided and then examined that level (Dansereau & Yammarino, 2000; Klein et al., 2000).

A key methodological development that enabled homology to be addressed was the conceptual distinction between measures referenced to the self and those that referenced a higher level entity. By conceptualizing the full model and parallel relations in advance, and then developing measures for each specific construct at its intended level of theoretical representation, researchers were able to begin probing homologous models empirically.

To my knowledge, the first successful example of validating a multilevel homology is represented by DeShon, Kozlowski, Schmidt, Milner, and Weichmann (2004). Note that there are many other examples of efforts to examine multilevel relations (cf. Chen & Bliese, 2002; Chen, Thomas, & Wallace, 2005; Chen et al., 2002; Gibson, 2001); I just think DeShon et al. (2004) was the first effort to validate homology that showed good evidence for parallelism across levels. The study provides a good illustration of application MLT to develop a model and method capable of examining contextual, emergent, and multilevel phenomena.

DeShon et al. (2004) were interested in the process of self-regulation (SR), and the potential for a homologous process of team regulation (TR) to emerge over time in a team context. SR is a useful conceptual heuristic for describing the process by which individuals initiate action, invest effort, and adjust their behavior to accomplish valued goals. In brief, SR involves an iterative process whereby individuals set goals, monitor progress, react to performance feedback, and adjust their effort and/or strategy to move toward goal accomplishment. SR is a dominant heuristic for explaining learning, motivation, and performance in psychology (Karoly, 1993), but it has largely been applied to individuals pursuing single goals. In most team task settings, individuals have their own goals and responsibilities to accomplish, but they also have requirements to back up team mates and to help accomplish the team objective. Thus, they actually have to regulate attention and effort to accomplish not just their own goal, but the team goal as well. At the individual level, this is a multiple goal model of regulation (DeShon et al., 2004).

Figure 14. 3 illustrates the multiple goal model, with dual goal-feedback loops referencing both individual and team goals. As noted above, the basic idea is that individuals monitor discrepancies between their respective goals and current performance states. When a gap exists between the desired goal and the current state, the individual is expected to invest more effort or to modify their strategy to close the gap. In a team context, where the individual has discretion to work on their own goal or to contribute to the team, there is a second goal loop that has to be monitored. All things being equal (i.e., that both goals are important, valued, etc.), one would expect that the goal loop with the largest discrepancy would garner regulatory resources. As individuals work to accomplish both goals, one would anticipate that the discrepancies would be in flux and that team members would have to allocate their regulatory resources dynamically across both goal loops to manage the dual discrepancies.

If one imagines a set of team members simultaneously trying to accomplish their individual goals, while also contributing to team goal accomplishment, and one plays out the dynamic resource allocation process described above over time, one can extrapolate that dynamic process yielding a process of individual SR and a parallel process of regulation that emerges at the team level. Figure 14. 4 illustrates the resulting homologous multilevel of regulation.

FIGURE 14. 3 A multiple goal model of individual and team regulation. From: DeShon, R. P., Kozlowski, S. W. J., Schmidt, A. M., Milner, K. R., & Wiechmann, D. (2004). A multiple goal, multilevel model of feedback effects on the regulation of individual and team performance. Journal of Applied Psychology, 89, 1035–1056. Published by the American Psychological Association. Reprinted with permission

How was the model evaluated? DeShon et al. (2004) had three-person teams perform a complex computer-based radar simulation over a series of multiple trials. All team members could view the same “synthetic world, ” and were responsible for monitoring a specific sector. The task was to monitor contacts, identify them, and take action to clear them. This contributed to individual performance. However, the task was designed to overload each team member systematically at different points such that team mates needed to help or the team would fail. Resource allocation to help a team mate constituted a contribution to team performance. Over the series of trials, teams performed the task iteratively, received feedback, and provided measures of the constructs. In addition to these task processes, the investigators manipulated feedback. In one condition, consistent with the multiple goal model shown in Figure 14. 3, teams received individual and team performance feedback. In two other conditions, team members received only individual or team feedback only. The purpose of this design was to maximize variation across teams. Remember, if you want to be able to detect between team differences, you have to have variation and this was accomplished by manipulating sensitivity to the multiple goal loops. In other words, the context was manipulated to create variance between teams.

NOTE: Constructs above dashed line represent team-level constructs. Constructs below line represent individual-level constructs.

FIGURE 14. 4 A multilevel homology of individual and team regulation. From: DeShon, R. P., Kozlowski, S. W. J., Schmidt, A. M., Milner, K. R., and Wiechmann, D. (2004). A multiple goal, multilevel model of feedback effects on the regulation of individual and team performance. Journal of Applied Psychology, 89, 1035–1056. Published by the American Psychological Association. Reprinted with permission.

What levels are inherent in these data? Well, there is a temporal or within person element. Individuals provided repeated measures over time. Level 1 is within person. Next, individuals are nested within teams. Level 2 is the person within team. Level 3 is the between team level. How were the team construct measures modeled? To keep the discussion focused, I will simply describe the constructs within the team regulation and self-regulation “intentions” sections of Figure 14. 4. At the individual level, the intentions of goals, commitment to goals, and one's self-efficacy to accomplish the goals (i.e., a self-perception of capability) were assessed by items that referenced the self. For the team level constructs, individual team members responded to parallel items that referenced the team as a whole (i.e., reference-shift; Chan, 1998). Because the team intentions were conceptualized as composition constructs, an indication of restricted within-team variance was used to justify aggregation of the data within teams to the team level. That is, we conceptualized the team-level regulation constructs as collective phenomena that would converge across team members. Significant ICC(1) values and indices of within team agreement (r_wg) provided the supporting evidence. Since there were multiple assessments of the SR and TR constructs over trials, this evaluation was conducted for each wave of data.

How was homology established? Principles based in GST indicate that to support a multilevel homology, one must establish that (a) constructs at both levels of analysis are parallel and (b) that the process linkages connecting parallel constructs are functionally equivalent (Kozlowski & Klein, 2000). As described previously, restricted within-team variance on core constructs provides evidence to support the construct validity, meaning, and aggregation of team constructs assessed via individuals’ perceptions. This provided evidence for parallel constructs.

The second piece of support, to demonstrate that the linkages connecting constructs within each level are also parallel, was established by examining configural invariance or the pattern of significant and nonsignificant linkages across levels. This is essentially a demonstration of functional equivalence across levels. I should note that scholars have also identified more restrictive criteria for functional equivalence – metric and scalar invariance – that may be useful as our theories develop more precision (Chen, Bliese, & Mathieu, 2005; Chen et al., 2002). Since researchers are just beginning to examine multilevel models empirically, I think we have a ways to go before we get to the necessary level of precision.

Beyond homology

As I noted previously, homologous models have the potential to be a potent way for us to generalize knowledge about phenomena from one level where they may have been well explored to another level where they have not. In that sense, establishing homologies around meaningful phenomena can advance our science. Certainly, in the middle range focus on team processes, we can potentially generalize much of what we know about individual learning, motivation, and performance. However, I think it would be a mistake to think that homology per se is the only important focus for multilevel modeling.

It is important to remember that each level in a hierarchically structured social system serves as a context for the level below it. For example, for the DeShon et al. (2004) study, the individual is the context for within-individual variation over time, the team is the context for within-team variation, and the feedback manipulation (analogous to an organizational subsystem feature) is the context for between-team variation. Thus, one could argue that the team-level motivational processes in the homology act as contextual constraints (i.e., top-down effects) on their lower level counterparts. Thus, the phenomenon is not merely parallel, but it is also inexorably entwined. Indeed, Chen & Kanfer (2006) have proposed just such a model to account for team motivation, and individual motivation in the team context, that goes beyond homology per se. Here the focus is not just on the parallel processes, but rather on the interplay across levels over time as constructs emerge from the lower level to the team level and exert influence on subsequent motivational processes at the individual level. From an organizational systems theory perspective, this is a nice example of the duality of process and structure (Katz & Kahn, 1966). Perhaps more importantly, it begins to make salient the dynamic aspects of multilevel phenomena that are inherent but latent in homologous multilevel models (DeShon et al., 2004).

Indeed, Chen and colleagues (Chen, Kanfer, DeShon, Mathieu, & Kozlowski, 2009) evaluated the Chen and Kanfer (2006) model using two datasets that had previously been used to examine individual and team motivational homologies; a study by Chen et al. (2005) and DeShon et al. (2004). They found good support for the Chen and Kanfer model across both datasets. Given that the data had previously examined homology, that cross-level interconnections were established, and that the model was supported in both datasets (which each used different tasks and construct operationalizations), this is pretty solid support for broadening our view of inter-level linkages.

Dynamics, the next frontier

Although MLT has had a substantial influence on team and group research in terms of theoretical principles, research methods and measurement, and analysis, there is one important area that is changing only ever so slowly. The growing number of multilevel articles (see Figure 14. 1) is still largely based on cross-sectional data. Although researchers apply MLT and hypothesize about process dynamics and emergent phenomena, for the most part the research is still focused on static models evaluated with static data. For MLT and team research to advance, this must change. We need to be serious about incorporating dynamic processes into our theory and then we need to be serious about modeling dynamic processes directly in our data. Our understanding of many phenomena is limited by the use of static models and between (without within) comparisons.

Dynamic within-person relationships

There are some good examples of modeling dynamic motivational processes in the literature that help illustrate my point. These examples are not team or group level, but they are multilevel because they examine within person variation over time. So, there are two levels, the within person level and between person level. One line of research that has led to an interesting debate is the work by Vancouver and colleagues, who have examined self-efficacy (SE) and performance relationships within persons over time (Vancouver, 2005; Vancouver, Thompson, Tischner, & Putka, 2002 ;Vancouver, Thompson,& Williams, 2001). Without trying to get into all the complexities, the research shows that, at the within-person level, SE tends to exhibit small negative correlations with subsequent performance. Essentially, when individuals perform well, they feel more confident. They then tend to reduce effort, and subsequent performance declines. This dynamic process playing out within persons over time yields, on average, a small negative relationship between SE and performance at the within-person level.

What is interesting about Vancouver's findings is that they are in stark contrast to the well-demonstrated (and widely accepted) positive relationship between SE and performance at the between-person level of analysis. As I noted, these differences have energized a vigorous debate (Bandura & Locke, 2003; Vancouver, 2005). However, although some frame this as a debate about “who is right,” the differences in perspective are likely due to difficulty in understanding that between-person differences and within-person dynamics are very different types of relationships in the data. It is entirely reasonable for the relations to be quite different in form! They are different views of the phenomenon. In any event, there is small cadre of researchers doing some exciting work examining the within person dynamics of motivation (e.g., Schmidt & Dolis, 2009; Schmidt, Dolis, & Tolli, 2009 ;Yeo & Neal, 2006, 2008).

Modeling the dynamics of emergence

One of the caveats I noted at the beginning of the section on emergence is that representation of composition and compilation constructs generally treats the process of emergence as theoretical (i.e., it is not directly assessed) and then tries to verify the assumed theory checking assumptions in the measurement. That is, investigators check for composition measures for evidence of within-team sharing and compilation measures for evidence of within-team variance or configuration. This is, admittedly, the application of a somewhat circular logic. The tendency among researchers to use cross-sectional designs (let's face it, they are easier) is the primary culprit. Really to begin to extend a multilevel modeling perspective, researchers are going to have to move toward truly longitudinal designs and to model the dynamic process of emergence directly. I wish I could describe some exemplary research, but unfortunately, I'm not aware of any as yet.

As a point of departure, it would not be that difficult to model composition emergence or compilation emergence. What would you need to get started? Well, first you would need to identify phenomena that emerge over time. Socialization of new team members, team development, and team mental models vs. team transactive memory would be my top picks, but you can examine anything that fits the models and your interests. Next, you would need at least a rough idea of the relevant time frame for the phenomena of interest to emerge. Our theories tend to lack precision regarding time, if they treat it at all, so do not expect theory to be very helpful. A good research-guided guess, coupled with as many measurement periods as possible would be a good way to generate descriptive data that could then be used to build theory. With such data in hand, one could model how individual mental model perceptions or climate perceptions begin to converge over time, how antecedents such as team leadership or team training facilitate or inhibit convergence, and how convergence on some key constructs may have ripple effects on the convergence of others. Similarly, one could also model how transactive memory linkages develop over time, how antecedents can facilitate links, and how errors could potentially have catastrophic effects on team performance. This is an incredibly rich and important area for research inquiry and, at the moment, it is not garnering much attention from researchers. Time will tell.

Conclusion

Groups and teams in organizations are collective entities. Although one can study them holistically, such approaches tend to miss both the effects of the broader organizational (or other) system within which teams are embedded and the dynamic emergent contributions from team members that create collective properties. The purpose of this chapter was to describe basic principles from MLT and methods that enable researchers to study teams in ways that take account of (a) higher-level contextual factors that shape team phenomena and (b) lower level interaction and exchange that yield emergent collective phenomena. Although MLT is grounded in organizational systems, the principles are applicable to groups in a wide range of settings including education, communities, families, and friendship networks, among many others. I hope the chapter has conveyed the value of applying MLT to the study of group and team phenomena and that the astute reader will be sufficiently intrigued to delve into the source material.

References

Arrow, H., McGrath, J. E., & Berdahl, J. L. (2000). Small groups as complex systems: Formation coordination, development, and adaptation. Thousand Oaks, CA: Sage.

Bandura, A., & Locke, E. A. (2003). Negative self-efficacy and goal effects revisited. Journal of Applied Psychology, 88, 87–99.

Bliese, P. D. (2000). Within-group agreement, non-independence, and reliability: Implications for data aggregation and analysis. In K. J. Klein & S. W. J. Kozlowski (Eds.), Multilevel theory, research, and methods in organizations (pp. 349–381). San Francisco: Jossey Bass.

Bliese, P. D., & Ployhart, R. E. (2002). Growth modeling using random coefficient models: Model building, testings, and illustrations. Organizational Research Methods, 5(4), 362–387.

Brown, K. G., & Kozlowski, S. W. J. (1999). Toward an expanded conceptualization of emergent organizational phenomena: Dispersion theory. In F. P. Morgeson & D. A. Hofmann (Chairs), New perspectives on higher-level phenomena in industrial/organizational psychology. Symposium conducted at the 14th Annual Conference of the Society for Industrial and Organizational Psychology, Atlanta, GA.

Brown, K. G., Kozlowski, S. W. J., & Hattrup, K. (1996). Theory, issues, and recommendations in conceptualizing agreement as a construct in organizational research: The search for consensus regarding consensus. In S. Kozlowski & K. Klein (Chairs), The meaning and measurement of within-group agreement in multi-level research. Symposium presented at the 56th Annual Convention of the Academy of Management Association, Cincinnati, OH.

Cannon-Bowers, J. A., Salas, E., & Converse, S. A. (1993). Shared mental models in expert team decision-making. In N. J. Castellan, Jr (Ed.), Current issues in individual and group decision-making (pp. 221–246). Hillsdale, NJ: Lawrence Erlbaum.

Chan, D. (1998). Functional relations among constructs in the same content domain at different levels of analysis: A typology of composition models. Journal of Applied Psychology, 83, 234–246.

Chen, G., & Bliese, P. D. (2002). The role of different levels of leadership in predicting self and collective efficacy: Evidence for discontinuity. Journal of Applied Psychology, 87, 549–556.

Chen, G., Bliese, P. D., & Mathieu, J. E. (2005). Conceptual framework and statistical procedures for delineating and testing multilevel theories of homology. Organizational Research Methods, 8, 375–409.

Chen, G., & Kanfer, R. (2006). Toward a systems theory of motivated behavior in work teams. Research in Organizational Behavior, 27, 223–267.

Chen, G., Kanfer, R., DeShon, R. D., Mathieu, J. E., & Kozlowski, S. W. J. (in press). The motivating potential of teams: Test and extension of Chen & Kanfer's (2006) cross-level model of motivation in teams. Organizational Behavior and Human Decision Processes.

Chen, G., Thomas, B. A., & Wallace, J. C. (2005). A multilevel examination of the relationships among training outcomes, mediating regulatory processes, and adaptive performance. Journal of Applied Psychology, 90, 827–841.

Chen, G., Webber, S. S., Bliese, P. D., Mathieu, J. E., Payne, S. C., Born, D. H., & Zaccaro, S. J. (2002). Simultaneous examination of the antecedents and consequences of efficacy beliefs at multiple levels of analysis. Human Performance, 15, 381–409.

Dansereau, F., & Yammarino, F. J. (2000). Within and between analysis: The variant paradigm as an underlying approach to theory building and testing. In K. J. Klein & S. W. J. Kozlowski (Eds.), Multilevel theory, research, and methods in organizations (pp. 425–466). San Francisco: Jossey Bass.

DeShon, R. P., Kozlowski, S. W. J., Schmidt, A. M., Milner, K. R., & Wiechmann, D. (2004). A multiple goal, multilevel model of feedback effects on the regulation of individual and team performance. Journal of Applied Psychology, 89, 1035–1056.

Flake, G. W. (1998). The computational beauty of nature: Computer explorations of fractals, chaos, complex systems, and adaptation. Cambridge, MA: MIT Press.

Gibson, C. B. (2001). Me and us: Differential relationships among goal setting, training, efficacy, and effectiveness at the individual and team level. Journal of Organizational Behavior, 22, 789–808.

Homans, G. (1950). The human group. New York: Harcourt, Brace, and Company.

Hunter, J. E., & Hunter, R. F. (1984). Validity and utility of alternative predictors of job performance. Psychological Bulletin, 96, 72–98.

Ilgen, D. R., Hollenbeck, J. R., Johnson, M., & Jundt, D. (2005). Teams in organizations: From input–process–output models to IMOI models. Annual Review of Psychology, 56, 517–543.

Ilgen, D. R., & Hulin, C. L. (Eds.) (2000). Computational modeling of behavior in organizations: The third scientific discipline. Washington, DC: American Psychological Association.

James, L. R. (1982). Aggregation bias in estimates of perceptual agreement. Journal of Applied Psychology, 67, 219–229.

James, L. R., Demaree, R. G., &Wolf, G. (1984). Estimating within group interrater reliability with and without response bias. Journal of Applied Psychology, 69, 85–98.

James, L. R., Demaree, R. G., & Wolf, G. (1993). r_wg: An assessment of within group interrater agreement. Journal of Applied Psychology, 78, 306–309.

James, L. R., & Jones, A. P. (1976). Organizational structure: A review of structural dimensions and their conceptual relationships with individual attitudes and behavior. Organizational Behavior and Human Performance, 16, 74–113.

Jones, A. P., & James, L. R. (1979). Psychological climate: Dimensions and relationships of individual and aggregated work environment perceptions. Organizational Behavior and Human Performance, 23, 201–250.

Karoly, P. (1993). Mechanisms of self-regulation: A systems view. Annual Review of Psychology, 44, 23–52.

Katz, D., & Kahn, R. L. (1966). The social psychology of organizations. New York: Wiley.

Klein, K. J., Bliese, P. D., et al. (2000). An integrative perspective on multi-level analyses. In K. J. Klein & S. W. J. Kozlowski (Eds.), Multilevel theory, research, and methods in organizations (pp. 512–553). San Francisco: Jossey Bass.

Klein, K. J., Conn, A. B., Smith, D. B., & Sorra, J. S. (2001). Is everyone in agreement? An exploration of within-group agreement in employee perceptions of the work environment. Journal of Applied Psychology, 86, 3–16.

Klein, K. J., Dansereau, F., & Hall, R. J. (1994). Levels issues in theory development, data collection, and analysis. Academy of Management Review, 19, 195–229.

Klein, K. J., & Kozlowski, S. W. J. (Eds.). (2000). Multilevel theory, research, and methods in organizations. San Francisco: Jossey Bass.

Kozlowski, S. W. J., & Bell, B. S. (2003). Work groups and teams in organizations. In W. C. Borman, D. R. Ilgen, & R. J. Klimoski (Eds.), Handbook of psychology: Industrial and organizational psychology (Vol. 12, pp. 333–375). London: Wiley.

Kozlowski, S. W. J., & Farr, J. L. (1988). An integrative model of updating and performance. Human Performance, 1, 5–29.

Kozlowski, S. W. J., Gully, S. M., McHugh, P. P., Salas, E., & Cannon-Bowers, J. A. (1996). A dynamic theory of leadership and team effectiveness: Developmental and task contingent leader roles. In G. R. Ferris (Ed.), Research in personnel and human resource management (Vol. 14, pp. 253–305). Greenwich, CT: JAI Press.

Kozlowski, S. W. J., Gully, S. M., Nason, E. R., & Smith, E. M. (1999). Developing adaptive teams: A theory of compilation and performance across levels and time. In D. R. Ilgen & E. D. Pulakos (Eds.), The changing nature of work performance: Implications for staffing, personnel actions, and development (pp. 240–292). San Francisco: Jossey-Bass.

Kozlowski, S. W. J., & Hattrup, K. (1992). A disagreement about within-group agreement: Disentangling issues of consistency versus consensus. Journal of Applied Psychology, 77, 161–167.

Kozlowski, S. W. J., & Hults, B. M. (l987). An exploration of climates for technical updating and performance. Personnel Psychology, 40, 539–563.

Kozlowski, S. W. J., & Ilgen, D. R. (2006). Enhancing the effectiveness of work groups and teams (Monograph). Psychological Science in the Public Interest, 7, 77–124.

Kozlowski, S. W. J., & Klein, K. J. (2000). A multilevel approach to theory and research in organizations: Contextual, temporal, and emergent processes. In K. J. Klein & S. W. J. Kozlowski (Eds.), Multilevel theory, research and methods in organizations: Foundations, extensions, and new directions (pp. 3–90). San Francisco, CA: Jossey-Bass.

Levine, J. M., & Moreland, R. L. (1990). Progress in small-group research. Annual Review of Psychology, 41, 585–634.

Lewin, K. (1951). Field theory in the social sciences. New York: Harper.

Lewin, K., Lippitt, R., & White, R. K. (1939). Patterns of aggressive behavior in experimentally created social climates. Journal of Social Psychology, 10, 271–301.

Likert, R. (1961). The human organization: Its management and value. New York: McGraw-Hill.

Marks, M. A., Mathieu, J. E., & Zaccaro, S. J. (2001). A temporally based framework and taxonomy of team processes. Academy of Management Review, 26(3), 356–376.

Miller, J. G. (1978). Living systems. New York: McGraw-Hill.

Morgeson, F. P., & Hofmann, D. A. (1999). The structure and function of collective constructs: Implications for multilevel research and theory development. Academy of Management Review, 24, 249–265.

Roberts, K. H., Hulin, C. L., & Rousseau, D. M. (1978). Developing an interdisciplinary science of organizations. San Francisco: Jossey-Bass.

Rousseau, D. M. (1978). Measures of technology as predictors of employee attitude. Journal of Applied Psychology, 63, 213–218.

Rousseau, D. M. (1985). Issues of level in organizational research: Multi-level and cross-level perspectives. In L. L. Cummings & B. M. Staw (Eds.), Research in organizational behavior (Vol. 7, pp. 1–37). Greenwich, CT: JAI Press.

Schmidt, A. M., & Dolis, C. M. (2009). Something's got to give: The effects of dual-goal difficulty, goal progress, and expectancies on resource allocation. Journal of Applied Psychology, 94(3), 678–691.

Schmidt, A. M., Dolis, C. M., & Tolli, A. P. (2009). A matter of time: Individual differences, contextual dynamics and goal progress effects on multiple-goal self-regulation. Journal of Applied Psychology, 94(3), 692–709.

Schneider, B., & Bowen, D. E. (1985). Employee and customer perceptions of service in banks: Replication and extension. Journal of Applied Psychology, 70, 423–433.

Schneider, B., Salvaggio, A. N., & Subirats, M. (2002). Climate strength: A new direction for climate research. Journal of Applied Psychology, 87, 220–229.

Staw, B. M., Sandelands, L. E., & Dutton, J. E. (1981). Threat-rigidity effects on organizational behavior. Administrative Science Quarterly, 26, 501–524.

Vancouver, J. B. (2005). The depth of history and explanation as benefit and bane for psychological control theories. Journal of Applied Psychology, 90, 38–52.

Vancouver, J. B., Thompson, C. M., Tischner, E. C., & Putka, D. J. (2002). Two studies examining the negative effect of self-efficacy on performance. Journal of Applied Psychology, 87, 506–516.

Vancouver, J. B., Thompson, C. M., & Williams, A. A. (2001). The changing signs in the relationships between self-efficacy, personal goals and performance. Journal of Applied Psychology, 86, 605–620.

von Bertalanffy, L. (1968). General system theory. New York: Braziller.

von Bertalanffy, L. (1972). The history and status of general systems theory. In G. J. Klir (Ed.), Trends in general systems theory (pp. 21–41). New York: Wiley.

Wegner, D. M. (1995). A computer network model of human transactive memory. Social Cognition, 13, 319–339.

Yeo, G., & Neal, A. (2006). An examination of the dynamic relationship between self-efficacy and performance across levels of analysis and levels of specificity. Journal of Applied Psychology, 91, 1088–1101.

Yeo, G., & Neal, A. (2008). Subjective cognitive effort: A model of states, traits and time. Journal of Applied Psychology, 93, 617–631.