Deep Structures of Collaboration: Physiological and Behavioral Underpinnings of Team Performance and Cohesion

Keywords: Human-Computer Interaction, CSCW, Machine Learning, Experimental Design and Statistics

Measures of Team Performance and Cohesion

Collective Intelligence (CI) is an objective measure of team performance or “smartness” (analogous to IQ) of work teams. It is defined as a group’s capacity to perform a wide variety of tasks, and is consistently predictive of their future performance. Previous research has found the team members’ average sensitivity to social cues to be a good predictor of CI. CI is also associated with teams’ ability to engage in tacit coordination, or coordination without communication. However, we lack the understanding of the deep structures of CI, that is how it develops, and how details of physiological responses and behavior are related to CI. Hence, in this project, we will sense and measure physiological and behavioral mechanisms, and examine their relation with CI. Along with CI, we will also explore whether members’ satisfaction with the team, as a measure of team cohesion or how members “feel” about the interaction, is associated with similar mechanisms.

Previous Work

Here are some findings about CI from previous work (see “References”):

  • The following results have been replicated in face-to-face and computer-mediated interactions, as well as across multiple cultures:
    • CI predicts future team performance on a wide variety of tasks.
    • Team members’ average social perceptiveness is a good predictor of CI.
  • Proportion of women in the team has been found to positively correlate with CI.
  • Average number of speaking turns has been found to positively correlate with CI.
  • While average social perceptiveness, proportion of women, and average number of speaking turns are predictive of CI, only the predictive power of average social perceptiveness was found to be statistically significant.
  • No correlation has been found between CI and group satisfaction.
Experimental Design and Data Collection

In the experiment, teams of two completed the Test of Collective Intelligence (TCI), which includes 6 tasks: typing, matrix reasoning, brainstorming, unscrambling words, solving sudoku, and recalling information from visual memory. Members of each dyad were seated in different rooms, but were able to see and hear each other via video conferencing during the TCI. We collected individual measures of social perceptiveness and demographics before the TCI and group satisfaction after its completion. Throughout the TCI, we recorded audiovisual data from Skype using a software called Evaer, electrodermal activity and heart rate using the Empatica E4 wristband, body movements using Microsoft Kinect v2, and EEG using the Muse headband.  Each session lasted approximately 30 minutes. None of the participant pairs knew each other before the experiment.

We recruited 116 (60 male, 56 female) participants from the participation pool of a large Northeastern university in the United States with the age range of 18 to 61 years old (M = 26.4, SD = 8.45). All participants were compensated 15 US dollars. We ran the study using both same- and mixed- gender teams (18 male-only dyads, 20 female only dyads and 20 mixed gender). We failed to capture physiological signals and video for six dyads due to technical equipment issues.

Synchrony in Facial Expressions, Electrodermal Activity (EDA), and Heart Rate (HR)

cscw_fig

I conducted a study in which we measured physiological synchrony in facial expressions, electrodermal activity and heart rate using Dynamic Time Warping as an indicator of coordination, and examined its relation with CI and group satisfaction.

Our findings are as follows:

  • Synchrony in facial expressions positively correlated with CI, while synchrony in electrodermal activity (EDA) positively correlated with group satisfaction. This indicates that different processes drive performance (CI) and cohesion (satisfaction) in groups, emphasizing the need to develop different models for them.
  • Synchrony in facial expressions mediated or partly explained the positive correlation between CI and social perceptiveness.
  • Age diversity, measured by distance, had a negative indirect effect on CI through reduced synchrony in facial expressions. However, it did not have a significant direct effect on CI, controlling for synchrony in facial expressions. So, partners with larger age difference typically synchronized less in their facial expressions, and thus had lower CI.
  • Ethnic diversity had a positive indirect effect on CI through increased synchrony in facial expressions. It also had a significant direct effect on CI, controlling for synchrony in facial expressions. So, ethnically diverse groups typically had high CI, regardless of their synchrony in facial expressions.
  • Ethnic diversity had a positive indirect effect on group satisfaction via synchrony in pEDA. However, had a negative direct effect, controlling for synchrony in pEDA. This suggests that ethnically dissimilar dyads are less likely to report satisfaction with their partner; however, ethnically dissimilar dyads who synchronized in electrodermal activity reported higher levels of group satisfaction.

These results will be published in CSCW 2017:

Chikersal, P., Tomprou, M., Kim, Y. J., Woolley, A. W., & Dabbish, L. (forthcoming). Deep Structures of Collaboration: Physiological Correlates of Collective Intelligence and Group Satisfaction. In Proceedings of the 20th ACM Conference on Computer-Supported Cooperative Work and Social Computing (CSCW 2017). [link to paper]

Contributions of our paper include:

  • New insights into the physiological mechanisms underlying CI (see results above).
  • Method that can be used to represent and measure synchrony in different types of physiological and behavioral signals.
  • Creates potential for technological interventions that improve CI and satisfaction by augmenting synchrony in facial expressions or electrodermal activity.
  • Since social cues and physiological responses are mechanisms that underlie CI, our work suggests potential for technological interventions that improve CI and satisfaction by augmenting other physiological and behavioral patterns of participants (to be explored in future work).
Amount of Speech and Synchrony in Acoustic Features (preliminary)

Preliminary speech analysis results revealed that:

  • Amount of speech has a negative correlation with CI. Based on previous work, this wasn’t expected and is thus a surprising finding! However, based on subjective analysis of limited samples, we hypothesize that this might be because dyads with low CI typically tend to speak a lot about topics irrelevant to the task at hand. I’m currently exploring how we can quantify relevance of speech content by manual annotations. In future, I’m interested in quantifying relevance of speech content by modeling topics in speech transcriptions.
  • Synchrony in standard deviation of loudness has a positive correlation with CI. I’m currently trying to interpret this finding better. Based on previous work on speech analysis, synchrony in standard deviation of loudness could indicate mutual laughter or excitement.
Multimodal Machine Learning: Ongoing Work

I’m currently also working on fusing the modalities (data from different sources) together to build models that predict CI and group satisfaction. I’ll update this page once I obtain some preliminary results.

References

The findings mentioned under “Previous Work” were reported in the following papers, which are excellent sources to learn more about CI:

  • Anita Williams Woolley, Christopher F. Chabris, Alex Pentland, Nada Hashmi, and Thomas W. Malone. 2010. Evidence for a collective intelligence factor in the performance of human groups. Science 330, 6004: 686– 688.
  • D. Engel, Anita Williams Woolley, Ishani Aggarwal, et al. 2015. Collective intelligence in online collaboration emerges in different contexts and cultures. CHI ’15 Proceedings of the SIGCHI Conference on Human Factors in Computing Systems.
  • D. Engel, Anita Williams Woolley, Lisa X. Jing, Christopher F. Chabris, and Thomas W. Malone. 2014. Reading the mind in the eyes or reading between the lines? Theory of mind predicts collective intelligence equally well online and face-to-face. PLoS ONE 9, 12: e115212.
  • Young Ji Kim, D. Engel, Anita Williams Woolley, Jeffrey Lin, Naomi McArthur, and Thomas W. Malone. 2015. Work together, play smart: collective intelligence in League of Legends teams. In Proceedings of Collective Intelligence 2015.
  • Anita Williams Woolley, Ishani Aggarwal, and Thomas W. Malone. 2015. Collective intelligence and group performance. Current Directions in Psychological Science 24, 6: 420–424.
  • Ishani Aggarwal, Anita Williams Woolley, C.F. Chabris, and T.W. Malone. 2015. Cognitive diversity, collective intelligence, and learning. In Proceedings of Collective Intelligence 2015.
  • Nicoleta Meslec, Ishani Aggarwal, and P. L Curşeu. 2016. The insensitive ruins it all: Compositional and compilational influences of social sensitivity on collective intelligence in groups. Frontiers in Psychology.

Our Publications:

Prerna Chikersal, Maria Tomprou, Young Ji Kim, Anita Williams Woolley, & Laura Dabbish. (forthcoming). Deep Structures of Collaboration: Physiological Correlates of Collective Intelligence and Group Satisfaction. In Proceedings of the 20th ACM Conference on Computer-Supported Cooperative Work and Social Computing (CSCW 2017). [link to paper]