Complexity matching as a measure of team dynamics

Abstract

Introduction: Interpersonal coordination and reciprocity in team members’ actions can be very complex but are essential for a team’s success during multi-agent tasks1. Although various measures have been used to quantify team dynamics, most of them are limited using very simple, rhythmic tasks. This talk will present the use of complexity matching (CM) to quantify interpersonal dynamics in a complex multi-agent search and retrieve task. The CM hypothesis states that the exchange of information between two complex systems is maximized when their complexities are similar2 and potentially can be used as a predictor of team success. Methods: We used a desert-herding task wherein three players are required to corral target agents (TAs), randomly spread across a field, by controlling their avatars with standard first-person-shooter game controls. Ten teams completed four sessions of 16 trials each where the task manipulations were- Target Number (9 or 18 targets), Visibility (fog or no fog), Heads Up Display [HUD] (HUD or no HUD). Task performance was measured by using each trial’s duration in seconds. The structure of participants’ movement and search dynamics during the trials was quantified using detrended fluctuation analysis [DFA]3 for each player during each trial resulting in a measure DFAα. A CM value was calculated for each trial as a measure of the average difference in the DFAα values of all players on a team. Results: Multi-level linear models were fitted where the fixed effects were the task manipulations and random intercepts were fitted for each team member and team and the dependent measure was CM. Significant Wald tests were found for all manipulations (χ2 > 7.51, p = .006) except Visibility (χ2(1) = 1.53, p = .22). Critically, the CM values were lower with experience (1st ,2nd vs. 3rd ,4th session; p < .001) and when HUD was available. When controlling for the task manipulations, CM was not predictive of task performance (b = 38.10, t = 0.82, p = .41), but teams completed the trials significantly faster with experience (χ2(3) = 94.62, p < .001), such that fourth session trials were 44.2 s faster than first session (p < .001). Discussion/Conclusions: Although CM was not a predictor of task performance of a team, the structure of their movement and search behaviours became similar with experience. Future research will investigate other CM measures including conversational data and apply similar methodologies to a competitive task. References 1Schmidt R, Richardson M. Dynamics of interpersonal coordination. Coordination: Neural, Behavioral and Social Dynamics, pp. 281–308, 2008. 2Rigoli LM, Lorenz T, Coey C, Kallen R, Jordan S, Richardson MJ. Co-actors Exhibit Similarity in Their Structure of Behavioural Variation That Remains Stable Across Range of Naturalistic Activities. Sci Rep 2020 10:1, vol. 10, no. 1, pp. 1–11, Apr. 2020, https://doi.org/10.1038/s41598-020-63056-x. 3Hardstone R, et al. Detrended fluctuation analysis: A scale-free view on neuronal oscillations. Front Physiol, vol. 3 NOV, p. 450, 2012, https://doi.org/10.3389/FPHYS.2012.00450/BIBTEX.