There are many widely used emotion and stress measures that have been well- validated at the individual level. However, to what extent are these applicable to team-level research? On one hand, emotion and stress are individual responses to internal and external stimuli. A team has no faculty to experience either one, so it could be argued that there is no such thing as team emotion or team stress. Rather, team-level expressions of emotion or stress can be discovered through patterns of interpersonal behavior, which may be observed directly or derived from artifacts such as activity logs or communications records. Often researchers looking at either factor in a team setting compute a team measure by averaging each team member’s score on a mood or stress scale and using this as a variable in the analysis of the team’s performance. For example, Pfaff (2012) averaged self-reported mood and stress scores from pairs of participants working on a shared task as a manipulation check. This was necessary because the task had a single shared outcome that provided only a team-level measure of performance, making correlation with individual mood or stress inappropriate. This could be described as a bottom-up approach to team emotion and stress.
On the other hand, the top-down approach asserts that when many people are gathered together, individual emotions are governed more by the emotional dynamics of the group. Much of this perspective comes from early research on emotion and crowd dynamics (McDougal, 1923). It argues that individuals imitate the emotional characteristics of those around them as a result of interpersonal relationships and social norms. Barsade and Gibson (1998) argue that a blend of top-down and bottom-up approaches can be valuable when studying group emotion, by considering both the mean and the variance of emotion measures within the team, as well as the influence of the most emotionally extreme members of the team. By extension, this approach equally applies to measuring team stress.
Once emotion and stress are measured, measuring their impact on team cognition and performance is a more complex problem. Aggregation of individual performance measures is a less valid measure of team performance, given that true team tasks involve interdependent actions and reactions among team members. Therefore, an individual’s score that does not take into account the dependency of that individual’s performance on another individual will misrepresent its contribution to team performance. On the other hand, when team tasks have a single measurable outcome, such as three individuals in different roles collaborating on a targeting task, the outcome of that task could be considered an appropriate measure of team performance. Unfortunately, a single performance score only reveals the what in terms of task outcome, but obscures the more important why of the specific individual and team cognitive activities that produced that outcome. An actionable understanding of team cognition must address the processes and behaviors that allow teams to become more adaptive and resilient under challenging conditions (Cooke et al., 2013).
Teams perform complex tasks via distributed cognition, in which cognitive functions are distributed among team members, and direct interaction between team functions occurs socially via communication channels. Indirect interaction happens through systems that allow one team member to monitor another, without requiring explicit communication between the two, such as with over-the-shoulder displays that allow one user to remotely observe a teammate’s interaction with an interface (Gutwin & Greenberg, 2001). As communications can be captured in both colocated and geographically distributed work, it becomes a valuable data source that has been applied to performance, coordination, and more (Cooke & Gorman, 2009). However, though the data are relatively easy to gather (via recordings or chat logs), the researcher faces difficult choices about the most appropriate ways to analyze the data. Methods to analyze communication data range from fast and automatable quantitative methods, which are efficient but risk oversimplifying the data, to complex and effortful qualitative methods, which are labor-intensive but can produce richer and more detailed results.
Many researchers choose quantitative methods to analyze communication data, especially when there are large volumes of data to process that would be impractical to analyze manually. Measures that can be drawn from recordings or communications logs include frequency and duration of messages as well as turn-taking behaviors, from which a researcher could derive patterns of dominance or other conversational dynamics that could predict team outcomes.
Another popular quantitative approach employs dictionary-based word count software like the widely used Linguistic Inquiry and Word Count (LIWC; Tausczik & Pennebaker, 2010) that analyzes text for characteristics such as attentional focus, social relationships, cognitive styles, individual differences, and even emotions. This analysis can provide broad characterizations of the relative proportions of different types of verbal expressions, such as social words (indicating interactions or relationships between team members) or cognitive words (references to causation or insight). However, such analysis pays little attention to context that could change the implications of words under circumstances, and also can require substantial preprocessing of data to remove extraneous artifacts (e.g., time stamps) or correct misspellings. Boonthanom (2004) developed a taxonomy of verbal and nonverbal cues of emotion expression unique to text-based communication that go beyond literal emotion words (e.g., sad, happy) to other cues that are analogs to nonverbal physical behaviors, like gesture or tone of voice. These include things like vocal spelling (mimicking vocal inflections by distorting the spelling of a word, like weeeellll or soooooo), lexical surrogates (verbal expressions that normally would not be written out, like uh huh or haha), emoticons, and others. Unfortunately, these cues must be detected and counted manually by the researcher as they do not match the standard dictionary of terms. Using these methods, the researcher can analyze whether the effects of different emotional or stress conditions are reflected in counts or proportions of many different types of communications cues collected during a team task. For example, Pirzadeh and Pfaff (2012) analyzed chat logs from a team emergency management exercise to find that stressed participants used significantly more negative emotion words, but fewer vocal spellings than nonstressed participants. Such a finding demonstrates stress influencing mood, as well as influencing how individuals choose to express that mood to other team members. In this case, if people are less likely to express emotion using vocal spellings when under stress than during routine communications, other members of the team are less likely to know how their teammates are feeling, and consequently how that might affect their ability to perform their tasks.
Qualitative communications analysis methods such as content analysis and conversation analysis require manual inspection of the series of utterances between individuals to identify themes, categories, and patterns that reflect the cognitive activities of the team members. Sequences of utterances are the means by which individuals coordinate and accomplish team activities (Mazeland, 2006). These could take the form of a request followed by a decision, a question followed by an answer, or comment followed by an acknowledgment. Taken in context with the unfolding team activity, these sequences are examples of ways the researcher can uncover the complex cognitive activities underlying the team’s success or failure at the current task.
A third option is to take a hybrid approach blending the strengths of both qualitative and quantitative approaches. When the task domain is well known, and there are boundaries to the types of things the team will discuss, a well-defined coding scheme can capture both quantity and meaning of utterances between team members. Adaptive Architectures for Command and Control (A2C2; Entin & Entin, 2001) is a coding scheme that classifies utterances into three categories: information (the status of events), action (ground-level activities), coordination (delegating or accepting responsibility to act), and acknowledgments. Each of the first three categories is further divided into requests and transfers. For example, an information request could be a question asking about how long before a resource will arrive at its destination, and an information transfer would be the response to that question. The ratio of transfers to requests produces an anticipation ratio, where high numbers reveal teams that anticipate the need to transfer information, action, or coordination without waiting for a corresponding request. This can be used as a measure of communication efficiency.