Time to go wild: How to conceptualize and measure process dynamics in real teams with high-resolution

Time and resolution

A key important observation from the review is that studies vary widely with respect to how they operationalize temporal resolution, defined as the number of repeated measurements captured within the phenomenon’s time span. For example, some studies have sampled team process data multiple times per minute throughout a 1-hr performance episode (e.g., Schecter et al., 2017). This reflects a high-resolution approach (albeit using a short time span, as we discuss later). Other studies have collected data multiple times per week over 2 years (e.g., Stuart & Moore, 2017), which is also a high-resolution approach (albeit using a longer time span). Both examples would be considered high-resolution, although the focal phenomenon respectively occurred and unfolded within different time spans (i.e., short vs. long), necessitating measurement with different granularity.

Hence, it is important to acknowledge that dynamic phenomena can theoretically unfold over different time spans. An initial concept of help is to distinguish between two families ¹ of temporal theories: episodic theories and developmental theories (see Table 2, column 1). Episodic theories focus on specific team performance episodes of concrete task work during which the team works towards common goals (i.e., “periods of time over which performance accrues and feedback is available,” Marks et al., 2001, p. 359). In the high-resolution literature, an example for a team performance episode could be an operation carried out by a medical team (Kolbe et al., 2014) or a flight conducted by a crew of pilots (Lei et al., 2016). Developmental theories have a long-term perspective and specify how teams mature over time and/or go through different qualitative stages, specifying how team phenomena change over larger time spans (e.g., Tuckman & Jensen, 1977). An example of developmental theory in the high-resolution literature is research on team performance recovery in ice-hockey teams unfolding over multiple games resulting of adaptive dynamics in team configurations over 2 years (Stuart & Moore, 2017). Both temporal theories are closely connected, such that episodes are usually nested within the long-term developmental life cycle of a team. For example, from an episodic lens, a health-care team may engage in dynamic interactions during a single surgery, which constitutes an area of research focusing on micro-dynamic interactions between team members. In contrast, from a developmental theory lens, the same team may also go through dynamic changes over multiple weeks (and surgeries), thus offering an opportunity to study long-term team dynamics.

OPEN IN VIEWER

Table 2.

Example phenomena of increasing time spans and their alignment with various high-resolution methods.

Temporal theory	Example phenomenon	Phenomenon time span (whole)a	Clock time measurement (when should the “parts” be measured)	Possible high-resolution methods to collect data
	Emotional mimicry	Within 1 secb	Milliseconds	Physiological measures (heart rate, neuroscience)
	Emotional contagion	Seconds to minutesc	Seconds or Minutes Hours	Video data, instant messaging, sociometric badges
	Team negative affect	Weeks to monthsd	Days	Electronic activities in virtual teams or crowd-sourcing competitions
	Team negative climate		Weeks	Written communication (e.g., e-mails)
	Collective burnout	Over multiple yearse	Months	Archival records (e.g., NBA sports records of multiple games)
			Years	Historical team records, yearly performance measures from HR records

^a Phenomenon time span was estimated based on the available literature (see references below) and represent “best guesses” of relevant time spans under which these phenomena most likely unfold.

^b Dimberg and Thunberg (1998).

^c Barsade (2002).

^d Paulsen, Klonek, Schneider, and Kauffeld (2016).

^e González-Morales, Peiró, Rodríguez, and Bliese (2012).

The episodic theory operates on a smaller time span than developmental theory. In other words, developmental theory should focus on team phenomena that unfold over larger time spans, that is, the dynamics and temporal changes of these phenomena unfold over days, weeks, months, or even years. Of note, the focal team phenomenon of interest should always dictate when repeated measurements are taken. That is, considering and carefully conceptualizing the time span of the whole phenomenon is crucial in identifying appropriate high-resolution methods. To give an example (Table 2, column Example phenomenon), emotional mimicry and emotional contagion are both socially dynamic phenomena. The literature suggests that emotional mimicry emerges and changes within seconds (Dimberg & Thunberg, 1998). For emotional contagion, the time span is supposed to be a little bit larger and entail multiple seconds to multiple minutes (e.g., Barsade, 2002; Lehmann-Willenbrock, Meyers, Kauffeld, Neininger, & Henschel, 2011). Hence, to align theory and methods, our recommendation would be to theorize about the dynamics of these phenomena using an episodic lens. In contrast, team negative affect is also a dynamic phenomenon but has a larger time span—most likely displaying weekly or monthly dynamics (Knight, 2013; Paulsen, Klonek, Schneider, & Kauffeld, 2016). Finally, some dynamic team phenomena, such as collective burnout (Gonzalés-Morales, Peiró, Rodríguez, & Bliese, 2012), have an even larger time span, that is, they could unfold and change over months or supposedly years.

Taken together, the concept of “high-resolution” is essentially a relative term that requires specification of the parts (i.e., the repeated measurements) and the whole (i.e., a phenomenon’s time span). That is, the studies that we reviewed have all collected repeated observations (i.e., “parts”), but our review showed that research varies significantly how the researchers operationalized the focal phenomenon time span (i.e., “whole,” ² see Table 2). For example, some studies studied a whole phenomenon with high-resolution across a 1-hr period, while others studied the dynamics of a phenomenon over months. To provide a scientific analogy for the parts-versus-whole problem, think about two phenomena of different scope, for example, “oceanic currents” and “submarine bacteria.” Both phenomena can be studied with high-resolution methods (e.g., satellites and microscopes), but a high-resolution image of oceanic currents would be a low-resolution map of bacterial activities that live in these oceans. Both oceanic dynamic current flows (macro) and bacterial processes (micro) offer ways to understand global climate dynamics. However, empirical findings obtained with one high-resolution method (i.e., satellites capturing dynamic oceanic currents) operate on a different theory level than findings obtained with another method (i.e., microscope about microbiological dynamics in the oceans). While both types of research may focus on related problems (i.e., geo-temporal dynamics that affect our global climate) and both use high-resolution methods, they use methods with fundamentally different scopes for capturing the respective phenomena. Arguably, the use of different scopes also has profound implications for the study of team dynamics.

To illustrate this notion of relativity, we can take into account the time span of a dynamic phenomenon and ask: When should the parts of the phenomenon be measured (see Table 2, column 4)? That is, dynamic phenomena that have a large time span require different sampling intervals than dynamic phenomena that have a short time span. Technically, the level of measurement reflects the actual data source, that is, “the unit to which data are directly attached” (Klein, Dansereau, & Hall, 1994, p. 198). This illustrates the parts-versus-whole issue of which researchers need to be aware. In other words, researchers need to think at what time intervals they should collect “parts” of the “whole” phenomena to properly model the dynamics of the phenomenon (i.e., how is the whole phenomenon unfolding and changing over time?). To illustrate, a phenomenon like emotional mimicry has been argued to have a short lifetime and can emerge and disappear within a second (Dimberg & Thunberg, 1998). Hence, understanding the temporal dynamics of emotional mimicry would require researchers to collect multiple “parts” of the “whole” phenomena with almost a millisecond precision.

Furthermore, we provide suggestions that point out which high-resolution methods best align with phenomena of different time span (Table 2, Column 5). For example, physiological measures captured via ambulatory wearables and tracking devices constitute high-resolution data that offer precision on the second (Bernstein & Turban, 2018; e.g., Wundersitz et al., 2015; Saavedra, Hagerty, & Uzzi, 2011) and would thus be suitable to study phenomena such as emotional mimicry. On a broader time span, high-resolution data from coded video-recorded team activities (Waller & Kaplan, 2018) have shown to provide multiple measures per minute (see Table 1) and hence allow to study micro-dynamic team phenomena that unfold within relatively short team episodes. To illustrate, Lei, Waller, Hagen, and Kaplan (2016) measured changes in team communication multiple times per minute to model how different interaction patterns related to team adaptive crew performance during flights. However, researchers could even consider archival records (which provide repeated measurements of team data over weeks, months, or years) as a high-resolution method. For example, Grijalva, Maynes, Badura, and Whiting (2019) used archival data to capture team coordination multiple times over the course of a basketball season and used the temporal order of games to model changes in the relationship between team composition features and team coordination.

Applying our proposed categorization to the existing literature, Table 1 presents illustrative studies organized according to these concepts. In the Temporal theory column, we classified research into the two categories of episodic (e.g., medical procedures, flight simulations) and developmental theories. It is noteworthy that some phenomena (e.g., team coordination) have been studied from both an episodic and a developmental theoretical perspective. For example, Kolbe et al. (2014) used an episodic lens and measured team coordination repeatedly over minutes (Table 1, column Indicator of temporal resolution). In contrast, Grijalva et al. (2019) employed a developmental lens and measured team coordination repeatedly over weeks.

Multiple approaches

In our review of studies (see examples in Table 1), we also identified three approaches for analyzing team dynamics. First, some researchers use what we refer to as the static approach because it focuses on between-team differences (assuming that team processes are static) to explain differences in team outcomes. Studies within this category present theoretical arguments to focus on fine-grained processes and then—despite using high-resolution methods for data collection—summarize fine-grained process data (over time) to form an aggregate variable. While such a summary reduces the complexity of the data and simplifies the analysis of linkages between team processes and team outcomes, it comes at the expense of truly capturing temporal dynamics. For example, Kauffeld and Lehmann-Willenbrock (2012) argued for the importance of uncovering microlevel interaction processes and used video recordings as a high-resolution method to measure dynamic team communication. However, the authors then aggregated the number of specific communication events over time for each team and used this time-aggregated measure to predict team effectiveness. Using time aggregates of “process” measures to predict some team outcome variable has one problem: This approach discards any form of temporal process variability and, hence, does not contribute toward a better understanding of temporal dynamics. An implicit (or explicit) assumption of this approach is that team activities are more or less stable over time and hence “process” is treated as a static variable. Typically, researchers adopting this approach seek to understand whether and how team processes (i.e., team coordination, planning behavior, etc. see Table 3) explain variance in team effectiveness (e.g., Kauffeld & Lehmann-Willenbrock, 2012; Schmutz et al., 2015). Accordingly, hypotheses are formulated in a way that makes static comparisons between teams: “Effective teams will show more/less of behavior X than ineffective teams.”

OPEN IN VIEWER

Table 3.

Three approaches for using high-resolution methods.

Approach	Treatment of process	Research questions
Static (LePine et al., 2008)Core assumption: “Teams differ with respect to their processes which affect their effectiveness”	Variations of variables between teams	Do differences between teams in the amount of teamwork behavior explain team performance? Do teams with different levels of coordination behavior show differences in team effectiveness? (Schmutz et al., 2015) Do teams that exhibit more planning behavior also show higher team productivity? (Kauffeld & Lehmann-Willenbrock, 2012)
Multiphase or socio-technicala,b (e.g., Gersick, 1988; Marks et al., 2001; Rousseau et al, 2006)Core assumption: “Teams must execute different processes at different times, depending on task demands”	Temporal variations of one-time variable (or one dynamic task feature) and one team process variable over time	How do dynamic changes in task conditions affect the level of teamwork behaviors? Does in-action team reflexivity increase over the phases of a team performance episode? (Schmutz et al., 2018) How do support team leadership behaviors differently stimulate team member voice in action versus transition phases? (Farh & Chen, 2018) How do adaptive crew behaviors (e.g., collecting information and task distribution) vary with changes in dynamic task demands (routine vs. non-routine situations)? Do crisis management teams spend more/less time in different team phases (i.e., structuring, information-sharing, decision-making) (Uitdewilligen & Waller, 2018)? How does teamwork change from routine to non-routine phases? (David & Schragen, 2018) How is team coordination in online teams affected by different levels of task interdependence? (Riedl & Woolley, 2017)
Process dynamicsa (e.g., McGrath & Tschan, 2004; Pilny et al., 2016)Core assumption: “Team processes can be characterized by systematic interaction patterns”	Temporal variations of at least two team variables over time	How does the occurrence of one teamwork behavior (e.g., workload sharing) affect another behavior (e.g., conflict) over time (within a single performance but also as the team develops)? How do effective versus ineffective teams differ in their temporally patterned team interactions (Stachowski et al., 2009)? Do teams during a surgical operation show systematic sequential behavior patterns? (Kolbe et al., 2014) How do temporal variations in team coordination affect temporal variations in team performance? How do team member traits emerge and damage team coordination as the team develops (Grijalva et al., 2019)?

^a Applicable to team episodic and team development model.

^b Predominantly used with episodic models.

The static approach still has research design advantages. That is, high-resolution methods using a static approach may mitigate common method problems if, for example, the team processes are measured with a nontraditional method and other phenomena are measured with a different method (e.g., self-report surveys). However, whether researchers use a time aggregate of a team process variable to predict team performance or if they survey team members about the process (e.g., team coordination) and relate this self-report measure to team performance are just two different methodological approaches to answer the same question (the first one using a multi-method, the latter using a mono-method approach). In other words, research from the static approach does not necessarily yield more time-theoretical insights than research using traditional methods. What is missing from this approach are questions that directed toward understanding at what time team activities were most beneficial for team performance. How did activities within the team change over time? How patterned are these team interactions?

A second approach used by researchers, which we refer to as multiphase and socio-technical, goes towards addressing questions about team dynamics. This approach is grounded mostly in episodic theories (e.g., the temporal multiphase theory from Marks et al., 2001) and socio-technical theory by taking into account that team processes differently affect team outcomes depending on the changing nature of the task itself (e.g., Leenders et al., 2016; Mathieu, Maynard, Rapp, & Gilson, 2008; Roe et al., 2012). An example high-resolution study from the multiphase and socio-technical approach is how effective flight crews adapt their behaviors according to the changing nature of routine versus nonroutine tasks over the course of a performance episode (Lei et al., 2016).

A third approach, which also helps to understand team dynamics, is what we refer to as process dynamics. In this approach, researchers try to understand how two (or more) team phenomena show systematic patterns over time (McGrath & Tschan, 2004; Pilny, Schecter, Poole, & Contractor, 2016). The core assumption is that team processes display systematic temporal patterns. Hence, this perspective fundamentally focuses on co-variations and contingencies of distinct team activities over time. A research hypothesis from this perspective needs to point out temporal variations (e.g., “Over time, team behavior X elicits/is associated with team behavior Y”). An example study following the process dynamics approach is the work by Kolbe et al. (2014) showing how effective health-care teams sequence their activities differently over time when compared to ineffective teams.

Summary of the state of the literature

In summary, the literature examining teams with high-resolution has refined our understanding of team dynamics. While much has been learned from this research, we noted considerable variation in temporal resolution that studies have used, and in the overarching analytical approaches that were applied to generate novel knowledge about team dynamics. Furthermore, studies have not well articulated the time spans for focal team phenomena and we discussed problems that arise when high-resolution approaches and focal phenomena are misaligned. In what follows, we argue there is a need to synthesize these novel concepts and approaches into a coherent and comprehensive framework.

The multiphase and socio-technical approach

As discussed earlier, the multiphase and socio-technical approach is mostly grounded in episodic theories (e.g., Marks et al., 2001) and builds on the core assumption that teams should execute different behaviors at different times during a specific episode in order to be effective. The socio-technical approach incorporates the notion that team processes affect team outcomes depending on the changing nature of the task itself (e.g., Leenders et al., 2016; Marlow, Lacerenza, Paoletti, Burke, & Salas, 2018; Mathieu et al., 2008). Research questions from this approach ask whether and how team processes (e.g., conflict, workload sharing, communication, or coordination) increase or change over distinct team tasks of a performance episode (e.g., a surgery, a flight, or a mission; e.g., how does team reflexivity change over the course of a performance episode?).

This approach advances theory by comparing fluctuations of team processes across different team phases (e.g., Schmutz et al., 2018; Manser, Howard, & Gaba, 2008) or across varying task levels (Hoogeboom & Wilderom, 2019; Lei et al., 2016), thus, challenging assumptions from the static approach by recognizing that activities are contingent on dynamic task characteristics. Teams are part of socio-technical organizational systems and changes of the technical system are tightly related with changes of the social system, that is, the extent to which specific activities play a fundamental role for team performance (Rousseau et al., 2006; Tiferes & Bisantz, 2018; Waller, 1999). This process approach has studied how volatile task features affected team process dynamics in medical emergency teams (Manser et al., 2008) and flight crews (Waller et al., 2004). For example, flight crews change their communication patterns when they encounter unexpected events (David & Schraagen, 2018), cardiac anesthesia teams adapt their coordination processes contingent on the level of task interdependence during an operation (Manser et al., 2008), and health-care teams increase reflection processes over time during medical emergencies (Schmutz et al., 2018).

The process dynamics approach

The process dynamics approach is trying to understand systematic patterns of multiple team phenomena over time (McGrath & Tschan, 2004; Pilny et al., 2016). The core assumption is that team processes display systematic temporal patterns. Hence, this perspective fundamentally focuses on co-variations and contingencies of distinct activities over time. A research hypothesis from this perspective needs to point out within-team variations that occur over time (e.g., “Within team interaction processes, behavior X elicits behavior Y”). Illustrative research includes questions like “How do temporal variations in team coordination affect temporal variations in team performance?” or “Do teams show systematic sequential behavior patterns during a surgical operation?” The process dynamics approach can yield both insights about dynamics regarding phenomena with short time spans (e.g., communication sequences, Bowers, Jentsch, Salas, & Braun, 1998, emotional contagion, Lehmann-Willenbrock, Chiu, Lei, & Kauffeld, 2017) but also about dynamics regarding phenomena with larger time spans (e.g., how do different personalities within a team affect team coordination when team familiarity is increasing over multiple performance episodes, cf. Grijalva et al., 2019).

Future research using this approach can advance our knowledge even further by taking into account both the socio-technical and process dynamics approach. For example, a researcher could use paradox theory to argue how teams require both highly flexible but also highly structured team temporal interaction patterns to be effective (Schad, Lewis, Raisch, & Smith, 2016). On the one hand, adherence to protocols should allow teams to operate most efficiently (Kanki, Folk, & Irwin, 1991), but on the other hand, teams that work together in a predictable manner are less agile and cannot respond to changing task demands (Hollenbeck, Ilgen, Tutle, & Sego, 1995). The paradox of these competing demands (“be flexible” vs. “be structured”) can be resolved by integrating the social-technical approach. More specifically, the researcher can develop hypotheses that denote how teams need different levels of process dynamics (flexible vs. structured interaction patterns) when working on different task types during (or across) performance episodes. That is, routine tasks may require teams to be aligned and structured in their communication, whereas nonroutine tasks may require flexible interaction patterns (Waller, 1999). While this integrated perspective is rather new, initial attempts have found promising results that showed how effective teams switch between temporal patterns when they work on routine tasks and flexible interaction patterns when they work on nonroutine tasks (Hoogeboom & Wilderom, 2019; Lei et al., 2016; Stachowski et al., 2009).

Antecedents of team dynamics

Finally, our model also incorporates input factors on multiple levels that will have an impact on these team dynamics (see left-hand side of Figure 1). That is, we acknowledge that team processes are affected by individual team member characteristics (e.g., personality, knowledge, and abilities), team-level input variables (e.g., team size, gender composition, etc.), and organizational contexts (e.g., team-based HR policies, organizational climate). System variables characterize the team as a whole and do not vary for a given performance episode, but they may change over the course of team development.

Identification of research questions focusing on team dynamics

First, researchers will need to ask a research question that identifies how knowledge about team dynamics is currently underdeveloped or incomplete. As our review of high-resolution methods has shown, empirical studies that unpack the actual dynamics at the core of team phenomena are still very rare, so there are lots of unanswered questions. An exemplary research question could be, “How do coordination patterns affect team effectiveness?” (Table 4, column 3). To answer this question, researchers need to estimate the phenomenon time span. In other words, over what time period does the focal phenomenon occur and fluctuate in meaningful ways? Answering this question will help to determine the temporal granularity at which the team phenomenon should be measured and answer how and at what time intervals measurements need to be taken. Overall, this step taps into the parts-versus-whole issue which is crucial for selecting the appropriate high-resolution measurement approach.

For example, imagine a team of researchers who want to understand how team coordination occurs within a surgical team, and whether team coordination has any impact on adverse patient errors. In terms of time span, based on unstructured observations (from having access to their local hospital) and based on the extant literature (Kolbe & Boos, 2019), the researchers know that team coordination behavior (assisting others, monitoring the patient) can unfold and change during a single operation. In terms of temporal theory, the researchers thus decide to adopt an episodic lens and to focus on coordination during surgical operations as the central performance episode. Based on this decision, the researchers need to decide how they want to measure coordination. That is, they need to decide when (i.e., how often) and how to measure (i.e., selecting the right high-resolution method) coordination behaviors that occur during the operation.

High-resolution data collection and management

The researchers now need to collect and manage high-resolution data to answer their research questions. When they have an idea about the time span over which the phenomenon unfolds as a whole (e.g., within seconds, days, or years?), they have a rough estimate for the observation period that is necessary for data collection. As an initial point of reference for estimation, researchers can use Table 1 which provides time frames for some team phenomena within field contexts and industries. Of note, it is not always easy to estimate the phenomenon time span as there is a dearth of empirical research that gives concrete estimates about such time frames (for exceptions see Delice, Rousseau, & Feitosa, 2019). Furthermore, it is important to select relevant time intervals for collecting repeated measures. In our example, the researchers consult the literature (e.g., Waller & Kaplan, 2018, see also Table 1) which indicates that coordination dynamics in surgical teams have been measured multiple times per minute. Alternatively, the researchers could have determined the temporal resolution by using Table 2: Knowing that the phenomenon time span is operationalized as a 30-min episode (which is the average length of operations in their local hospital), this would mean to use a high-resolution method that repeatedly samples the phenomenon with at least a minute (or even second) precision. The researchers consider that they want to measure coordination behaviors in 10-s intervals by tallying coordination behaviors in these intervals throughout the medical operation.

Following this, it is important to select an appropriate high-resolution method. There are a variety of high-resolution approaches from which the researchers can select. For example, researchers can use video recordings (Waller & Kaplan, 2018), use text data from team transcripts (e.g., Driskell et al., 2017), e-mails or electronic-traces of team interactions (Braun, Kuljanin, & DeShon, 2018), access archival records (cf., Braun et al., 2018; usually used in the context of sports teams, this type of data is easily accessible via websites: e.g., www.basketball-reference.comorwww.nhl.com), or utilize sociometric badges (Kim, McFee, Olguin, Waber, & Pentland, 2012). ³ As we have outlined in Table 2, some of these methods are better suited for phenomena that have a short time span (e.g., video data, instant messaging, or sociometric badges), while other high-resolution methods are better suited for team phenomena that have a larger time span (e.g., archival data). In our hypothetical research study, the researchers have the opportunity to analyze an archive of video-recorded anesthesia inductions that have been collected from the local hospital as part of a training program. The hospital grants the researchers access to this dataset for research purposes. In return, the hospital hopes to receive insights on how patient safety during medical procedures can be improved.

Once the high-resolution method is selected, the researchers need to decide how to manage the complex data set. In essence, this means to decide how the “raw data” (i.e., video records) can be transformed into meaningful (and repeated, that is, time-logged) quantitative measures of the focal team phenomenon. Depending on the selected high-resolution methods (i.e., video observations, chat logs/transcripts, archival data, etc.), data transformation procedures are more or less evolved—both in terms of efforts, that is, manual versus automatic transformations, and in terms of operationalizing different team phenomena. For example, when using a video/observational approach, there is a selection of various manual coding tools (focusing on various team phenomena). Deciding which tools to use should be mainly guided by the research question and which phenomena researchers would like to understand. In this respect, Brauner, Boos, and Kolbe (2018) have organized over 20 coding tools (including an annotated Appendix of an additional 48 coding schemes) into different areas of team phenomena (general group processes, conflict, coordination, cognition, etc.). They also provide decision criteria for selection (Brauner et al., 2018): For example, is the tool available in different languages, are there online versus paper versions, where is the tool typically applied (lab or field research), practical aspects such as resources, training manuals and technical requirements, tool quality in terms of reliability and validity, and existing publications. This information is key in weighing various pros (e.g., reliability and validity data about a specific coding tool, application in many or few studies) and cons (e.g., complexity: Is the tool so complex that it takes 2 hr to code 1 min, is the training manual only available in French?). Furthermore, Waller and Kaplan (2018) provided some guidance for researchers who want to develop their own coding schemes.

While progress is on the horizon to use automatic or machine learning for obtaining meaningful team measures from high-resolution data (e.g., Bonito & Keyton, 2018; Lehmann-Willenbrock, Hung, & Keyton, 2017), the majority of video-based approaches still relies on trained human observers who need to have sufficient context knowledge, cognitive capabilities, and understanding of the focal team process. Moreover, while behavioral and verbal processes are more easily observable, cognitive and affective team processes may be less visible for external observers (Carter, Carter, & DeChurch, 2018). Importantly, in order to understand team dynamics, it is crucial that the selected high-resolution method also repeatedly time stamps the focal phenomenon. For example, when using archival sports data, the researcher needs to know at what year, month, or week the teams played in a game. This is essential to be able to draw an accurate picture of the team’s developments over time. As another example, when coding team interactions from video data, the coding tool should not only code but also time-stamp occurrences of team coordination behaviors. When using chat data, the researchers need to pay attention to get information about the time points that messages have been sent.

To increase the rigor of a research design, we recommend linking data from non-traditional methods with other methods (e.g., self-report team measures are very well suited to assess latent or perceptual team phenomena that are harder to capture with nontraditional methods). That is, researchers are well advised to use multiple methods in their study designs whenever possible. At the same time, reviewers should also understand that field projects can extensively limit a researcher’s ability to use the most rigorous measures (e.g., a 20-item survey measure for a single construct may be difficult to use in the wild). In our example study, the researchers had access to team performance measures (i.e., an experienced staff member rated clinical team effectiveness with an established checklist that covered adverse patient outcomes) but could not implement additional self-report measures after the videotaped operations.

If a researcher decides to collect electronic data, archival records, or text data and aims to transform the data into meaningful team phenomena, we recommend consulting the exemplary studies presented in Table 3. For example, Riedl and Woolley (2017) used raw data (electronic logs) from crowd-working teams and transformed them into indicators of team process measures. Others have used archival data from sports teams to measure team coordination (Halevy, Chou, Galinsky, & Murnighan, 2012; Swaab, Schaerer, Anicich, Ronay, & Galinsky, 2014). If the researchers have some sort of text data (e.g., virtual chat logs, e-mail transcripts), they could rely on various tutorials that explain how to use automatic text analytic approaches (Banks, Woznyj, Wesslen, & Ross, 2018; Bonito & Keyton, 2018; Gonzales, Hancock, & Pennebaker, 2010; Short, McKenny, & Reid, 2018). While text-analytic approaches are more advanced with respect to the automation of data transformation, they may still involve transcription efforts (e.g., if the researchers had decided to transcribe video-recorded communication, e.g., Bonito & Keyton, 2018) and close collaboration with computer scientists (Büngeler, Klonek, Lehmann-Willenbrock, Morency, & Poppe, 2017). If the researchers had decided to use wearables for data collection, we advise them to have a look at Chaffin et al. (2017) who systematically evaluated the validity of different wearable sensor methods (e.g., microphones, bluetooth) for capturing various team phenomena (e.g., boundary-spanning behaviors, emergent leadership) in different study settings (lab vs. field). Scientific advancement will be facilitated when reviewers and editors, in their position as gatekeepers, acknowledge that research “in the wild” often has exploratory elements and a certain degree of messiness. Accordingly, an open mindset toward novel approaches is necessary to allow expanding our understanding of team process dynamics.

In our hypothetical study, the researchers (who decided to use video-recorded data) now consult the aforementioned overview of coding schemes from Brauner et al. (2018) and selected Co-ACT (i.e., a coding scheme to measure coordination in acute care teams). This allows them to capture different types of micro-coordination behaviors. Furthermore, the researchers decided to use software support for coding the videos (after reviewing Klonek, Meinecke, Hay, & Parker, in press; Lehmann-Willenbrock & Allen, 2018; Waller & Kaplan, 2018) which allows them to save time stamps and the sequential order of the micro-coordination behaviors.

Data analysis

In the third step, the researchers of our exemplary study now have a completely coded dataset of time-stamped team coordination behaviors (that were displayed throughout an episode of anesthesia inductions). Going back to their initial questions, the researchers wanted to know how team coordination behaviors occurred in these performance episodes and how these dynamics impacted team outcomes (including adverse patient errors). Statistical approaches should help the researchers in answering these questions about temporal patterns or whether temporal team patterns show a relationship with other variables (e.g., team outcomes). There are a number of statistical guides available that researchers could use as a starting point to detect non-random temporal patterns (see Table 4, demos and tutorials). A helpful distinction in selecting the appropriate analytical approaches is to understand whether they rely on time-stamped categorical measures (e.g., Ballard, Tschan, & Waller, 2008; Herndon & Lewis, 2015; Klonek, Quera, Burba, & Kauffeld, 2016; Waller & Kaplan, 2018) or whether they rely on time-stamped continuous measures (e.g., Collins, Gibson, Quigley, & Parker, 2016). For example, statistical methods like sequential analyses, relational event modeling, or pattern analyses all rely on analyzing the temporal order of different codes that describe how team interactions unfold as discrete categorical events over time (they also have a tradition to be used in research focusing on micro-dynamics). In our hypothetical example, time-stamped codes that reflect temporal coordination would be: [01:20] “talking to the room,” [01:35] “monitoring,” [01:40] “providing assistance.” Other examples (from using electronic data) for time-stamped categorical data would be [00:01] “A sends a message to B,” [00:05] “C sends a message to A” (see Leenders et al., 2016; Schecter et al., 2017). It is also possible to use these methods to study archival data focusing on larger time spans (e.g., [Month1] “project A,” [Month2] “project B,” [Month3] “project D”; for related applications see Biemann & Wolf, 2009).

In contrast, statistical methods such as growth curve modeling (Collins et al., 2016) or multilevel modeling (Hedecker & Gibbons, 2006) mostly build on repeated continuous measures and use statistical approaches that extend regression-type approaches which most management and organizational behavior researchers are more familiar with.

In terms of statistical software solutions, the researchers can rely on detailed tutorials for analyzing categorical data (e.g., for sequential analytic approaches, see Bakeman & Quera, 2011, Biemann & Datta, 2014; Pilny et al., 2016; for pattern analytic approaches, see Magnusson, 2000; for comparisons of analytical approaches and available software options, see Lehmann-Willenbrock & Allen, 2018) as well as continuous team measures (e.g., for using growth curve modeling, see Collins et al., 2016; for using multilevel modeling, see Hedecker & Gibbons, 2006).

Because the researcher in our hypothetical study used time-stamped codes of coordination micro-behaviors, they select one of the various options for analyzing categorical team process measures. Unfortunately, they are unsure whether to select sequential analyses (Bakeman & Quera, 2011), pattern analyses (Ballard et al., 2008) or relational event modeling (e.g., Pilny et al., 2016). If the researchers were interested in the relational communication structures (i.e., the tendencies of team members to reciprocate messages from others), they could use methods like relational event modeling (Pilny et al., 2016). If the researchers were interested to which extent teams synchronize their interdependent activities in a more scripted versus a temporally flexible (or more spontaneous) way, they could use methods like pattern analyses (e.g., Ballard et al., 2008; Magnusson, 2018). Interaction patterns are defined as “a set of molecular actions that […] repeatedly co-occur” (Stachowski et al., 2009, p. 1537). The repeated co-occurrences of activities mean that certain behaviors happen either simultaneously or in close temporal proximity. When teams exhibit strong patterns, they typically adhere more to strict rules and procedures or scripts, much like following a recipe: First team member A does X, then B does Y, and so on. In contrast, teams exhibiting weak patterns are considered to be flexible and agile in their activities.

However, the researchers in our hypothetical example decide to unpack the systematic sequencing of effective team coordination. Sequential associations can be positive (A is likely to be followed by B) or negative (A likely suppresses B). Our researchers wonder whether teams that differently sequence their coordination behaviors (i.e., monitoring and speaking up during the procedure) also show better performance. Hence, they decide to analyze their data set with sequential analyses (using tutorials provided by Bakeman & Quera, 2011; Klonek et al., 2016). Since the researchers had an external rating of team effectiveness, the researchers divide their sample into teams that showed high performance versus teams that showed low performance. In the next step, they test whether those two groups showed different sequential patterns. Using software provided in the tutorials (e.g., Klonek et al., 2016), they create time-lagged matrices and calculate statistical association indices for each team. When comparing the high with the low-performing teams, they find that high-performing teams showed a higher probability for monitoring-speaking up sequences, whereas low performing did not show these behavioral sequences. ⁴

Of note, sequential analyses are not restricted to the phenomenon of small time spans. Researchers who want to focus on team development could use sequential analyses for other types of high-resolution data as well (i.e., archival data). For example, when studying how changes in team membership occur over time, researchers can identify team member exit patterns and relate these patterns to team performance data (for more details, see Biemann & Wolf, 2009; Herndon & Lewis, 2015).

Interpretation of results

In the final step, the researchers need to interpret their results. In some cases, interpretation of results may be straightforward. In particular, when following a deductive hypothesis-driven approach (i.e., validating a hypothesized association) and relying on methodological guidelines that specify benchmarks for measurement fit and statistical significance.

In our example, the researchers have used coded team coordination (from video recordings) and this approach involved transformation of qualitative data into quantitative data (through a coding scheme). Since this approach overlaps with mixed-methods research (Gibson, 2017), it also allows more discovery-oriented research and the use of abductive reasoning (Behfar & Okhuysen, 2018) which may be key in discovering new knowledge about team dynamics.

Mixed methods

As noted above, some high-resolution approaches (i.e., video recordings, the use of text or e-mail transcripts) rely on the transformation of qualitative data into quantitative data (i.e., numbers/categories representing variables). Hence, researchers can apply a mixed-methods approach (Gibson, 2017). The strength of mixed-methods research is that researchers can use both quantitative and qualitative data to triangulate their concepts, to provide richer descriptions of the focal phenomenon, and to extract novel phenomena. By using a mixed-methods approach, researchers could first analyze a subset of data using a qualitative lens (qualitative data is mostly unstructured data such as video recordings, field observations, or written communication) that will help them to extract meaningful patterns (e.g., using a micro-temporal lens, patterns could be re-occurring communication sequences). In a second step, researchers can select from the various manual or (semi-automatic) data transformation methods (Table 4) and use analytical methods to test statistical associations of these observed dynamics.

Researchers can also apply mixed methods in the reverse order (i.e., quantitative followed by qualitative analytical techniques). We have not seen many studies that went back to qualitative data to understand the meaning of certain patterns (Hoogeboom & Wilderom, 2019; Wang, Doucet, Waller, Sanders, & Phillips, 2016). For example, after having identified meaningful patterns (e.g., are there sequences in the performance episode during which team members disagree vs. agree after a solution was proposed?), researchers can investigate what happened immediately before or after these focal sequences. Researchers can also use this mixed-methods to further validate their hypotheses. To provide an example, Wang et al. (2016) developed competing hypotheses about the effect of shared versus unshared laughter for crews during a simulated flight. Using quantitative methods, the authors found that shared laughter was detrimental to team performance. Following this, the authors used qualitative methods to understand what happened during laughter episodes. This in-depth analysis helped them to extend their arguments that shared laughter shifted the task focus of the team towards a “play focus.” That is, shared laughter episodes showed that teams engaged in playful conversations and stopped to focus on important task aspects. In sum, this approach was insightful to understand the nature of counter-intuitive findings.

Abductive reasoning

Abductive research describes a middle-ground approach between theory testing (deductive, typically quantitative approaches) and theory development (inductive, typically qualitative approaches; e.g., Bamberger & Ang, 2016). Abductive approaches apply when researchers are dealing with anomalies and surprising finding (Bamberger & Ang, 2016; Behfar & Okhuysen, 2018). This type of research is more discovery-oriented (and may fit well with novel journals such as the Academy of Management Discoveries). For example, Bowers, Jentsch, Salas, and Braun (1998) used sequential analysis in an exploratory way and concluded that the “analysis provided hypotheses about the nature of effective team process that would not have been revealed by using simple frequency counts” (p. 677). This quote illustrates that researchers shifted their mindset from a static team research lens towards a process dynamics lens. This abductive reasoning can also be found in some reflection reports from researchers who used pattern analyses. These researchers “were disappointed to find virtually no significant difference” when they compared static process measures (i.e., frequencies) between high- and low-performing teams but “found almost all the measures to be significantly different” when using an interaction pattern recognition program (Ballard et al., 2008, p. 328; see also Waller & Kaplan, 2018).

It is noteworthy that some of the reviewed studies that focused on interaction patterns presented their research in a hypothetico-deductive fashion (i.e., using a priori hypotheses) in the published manuscripts (e.g., Zijlstra et al., 2012; Lei et al., 2016)—likely in order to adapt to common expectations of reviewers and editors who are less open to the more discovery-oriented and abductive reasoning approaches described in the above quote from Ballard, Tschan, and Waller (2008). The same likely applies to many other published studies using either sequential or pattern analysis. It is possible that this perspective has already changed as some journals (e.g., Academy of Management Discoveries) are becoming more open toward abductive reasoning approaches and give researchers a chance to describe (unsuccessful) research endeavors (Bamberger & Ang, 2016), report surprising and unexpected results, and provide arguments and possible reasons for unexpected findings in order to tell a coherent story. Nevertheless, it probably still requires a joint effort of authors and reviewers until explorative approaches are truly established in a wider range of top-tier journals. That is, authors need to be more transparent about their explorative analysis trying to “find something in the data,” and reviewers need to be more accepting of the possibility that existing theory can be advanced without a priori assumptions.

In conclusion, we hope that our review and “how to guide” will encourage researchers to “go wild” and use nontraditional high-resolution methods to study dynamic phenomena in real teams. Our conceptual framework can serve as a starting point for theorizing and for aligning theory and methods. Ultimately, we are confident that increasing the variety of methods to study teams in the field will contribute to advancing our knowledge about team process dynamics.