Three presenters from LUNDSTEP presented their work on communication dynamics in the context of medical student training (van Klaveren) and to newly formed teams in an escape room board game (Schelin, Nordbeck) at University of Montpellier this summer. Find the abstracts below. (Link to workshop.)
RQA of Dynamic Team Learning Behaviors and Their Relation to Performance
Michelle Schelin∗ , Lisa-Maria Van Klaveren , and Patric C. Nordbeck∗
In complex work environments teams must continuously learn and adapt to perform effectively. While theoretical advancements conceptualize teams as complex, adaptive systems shaped by non-linear interactions and emergent processes, empirical team learning research continues to rely on static methodologies that do not capture these dynamics (Mulder, 2022). To address this gap we apply Recurrence Quantification Analysis (RQA) and chromatic (cRQA) to explore the temporal dynamics of team learning behaviors: knowledge sharing, co-construction, constructive conflict and procedural coordination (Decuyper et al., 2010). Using transcribed conversational data from teams completing escape room simulations in a controlled laboratory experiment, this study explores (1) intra-team patterns of learning behaviors, (2) inter-team differences, (3) the relationship between behavioral patterns and team performance, and (4) the role of constructive conflict as necessary for higher performance. Key RQA metrics (including recurrence rate, determinism, entropy, and laminarity) will quantify behavioral stability, adaptability, and complexity over time. We expect high-performing teams to exhibit mixed structured-variable recurrence patterns, while lower performing teams display either excessive rigidity or erratic behavioral shifts. Exploring the ability of RQA to quantify team learning behavior patterns for performance prediction, the findings have practical implications for designing organizational training and guidelines to develop adaptive and resilient teams.
Keywords: team learning, dynamic systems, recurrence quantification analysis, team performance
Opening the black box of team-based learning: Exploring dynamics of team conversations in online application sessions
Lisa-Maria Van Klaveren∗, Patric C. Nordbeck , Tobias Boerboom , Rien De Vos, and Linda Roossien
Team-Based Learning (TBL) promotes in-depth discussions and deep learning. To better understand how conversational dynamics emerge from task and team constraints during online TBL-application sessions and their relation to learning outcomes, we conducted an observational study involving twelve TBL-teams. Each team participated in two sessions, completing five different tasks. We collected preparation and outcome scores, and audio- recorded the conversations. Transcripts were coded for team learning behaviors, including sharing, co-construction, and constructive conflict. We applied chromatic recurrence plots and recurrence quantification analysis at team and task levels. Wilcoxon Signed-Rank tests assessed differences in conversational dynamics, while General Linear Mixed Models examined relationships between preparation, dynamics, and performance. Findings indicate that entropy varied across teams, suggesting that teams exhibited distinct dynamics. Additionally, differences in recurrence rate and entropy across tasks highlight how they shape team learning. Relationships between preparation, conversational dynamics and performances provided mixed results that may hint at the complexity of team learning. These insights contribute to how to intentionally vary task design to enhance team learning in TBL. Understanding how conversational dynamics emerge from task and team constraints can inform evidence-based approaches to stimulate in-depth discussions and deep learning.
Keywords: Conversational dynamics, team learning, constraints, recurrence quantification analysis
Direktorat Pusat Teknologi Informasi
What is the purpose of using Recurrence Quantification Analysis (RQA) to study team learning behaviors?
lundstep
Hi there and thank you for your question!
In general, different RQAs (CRQA, cRQA, mdRQA, etc.) quantifies both simple and more complicated patterns that repeat over time. So it has an advantage over linear statistical analyses (including Machine Learning techniques that rely on linear equations/algorithms) in that it can quantify a, or several, patterns within the same (or between multiple) dataseries. It can thus also capture non-linear shifts in patterns between one part of a data series and another. A dynamic pattern, or a particular value, within the dataseries (depending on the theorizing of the concept/phenomenon at hand) can be considered as a stable state: perhaps high performance (measured repeatedly, discretely or continuously) is defined as above 0.9 on a 0-1 scale, then an RQA can be performed to understand if high performance is embedded in a larger/more complicated pattern. Using a CRQA, we could then also see if there is a lead-lag relationship between some behavioral variable and the performance variable (in effect, a repeating pattern between the two dataseries). We could also define “state” a bit differently, so take communication within a team as an example, are there patterns of topics that a team discusses? Do certain patterns represent certain qualitative phenomena when teams are performing a task? For example, a team that continuously circulates between a particular set of discussion topics might be exactly what is needed for high performance or learning in a particular context. In another, it might signal that the team is getting stuck in certain states and does not move on, hampering learning (but then one should also ask, learning of what?). So, in these ways, we can understand the process by which teams are or are not learning, working, or developing, and it can be tied to performance, or task solution, or learning, or whatever else is of importance for the researcher.
I also visited your webpage by the way, I found it interesting and with good explanations on Machine Learning and Deep Learning! I am sure that different RQAs could be interesting to use in a Machine Learning approach, the difficult thing I find with RQAs is rather the interpretation of the metrics that an RQA produces. What does it mean for a particular team and for a certain learning outcome that ‘percent determination’, ‘entropy’, or ‘laminarity’ is high or low in certain contexts? RQAs do allow for these to be compared across conditions though and can be subject to regular linear statistical analysis (and so fit in more structurally simple experimental paradigms). I like that RQA also produces a visual representation of the pattern repetetitions of a dataseries under analysis, as well as being computationally light and a mathematically (relatively) simple approach.
Thank you again for your question, if I have not given a great explanation here or you have further questions, feel free to reply back or send me an email (which you can find here).
Regards,
Patric
lundstep
Hi there and thank you for your question!
In general, different RQAs (CRQA, cRQA, mdRQA, etc.) quantifies both simple and more complicated patterns that repeat over time. So it has an advantage over linear statistical analyses (including Machine Learning techniques that rely on linear equations/algorithms) in that it can quantify a, or several, patterns within the same (or between multiple) dataseries. It can thus also capture non-linear shifts in patterns between one part of a data series and another. A dynamic pattern, or a particular value, within the dataseries (depending on the theorizing of the concept/phenomenon at hand) can be considered as a stable state: perhaps high performance (measured repeatedly, discretely or continuously) is defined as above 0.9 on a 0-1 scale, then an RQA can be performed to understand if high performance is embedded in a larger/more complicated pattern. Using a CRQA, we could then also see if there is a lead-lag relationship between some behavioral variable and the performance variable (in effect, a repeating pattern between the two dataseries). We could also define “state” a bit differently, so take communication within a team as an example, are there patterns of topics that a team discusses? Do certain patterns represent certain qualitative phenomena when teams are performing a task? For example, a team that continuously circulates between a particular set of discussion topics might be exactly what is needed for high performance or learning in a particular context. In another, it might signal that the team is getting stuck in certain states and does not move on, hampering learning (but then one should also ask, learning of what?). So, in these ways, we can understand the process by which teams are or are not learning, working, or developing, and it can be tied to performance, or task solution, or learning, or whatever else is of importance for the researcher.
I also visited your webpage by the way, I found it interesting and with good explanations on Machine Learning and Deep Learning! I am sure that different RQAs could be interesting to use in a Machine Learning approach, the difficult thing I find with RQAs is rather the interpretation of the metrics that an RQA produces. What does it mean for a particular team and for a certain learning outcome that ‘percent determination’, ‘entropy’, or ‘laminarity’ is high or low in certain contexts? RQAs do allow for these to be compared across conditions though and can be subject to regular linear statistical analysis (and so fit in more structurally simple experimental paradigms). I like that RQA also produces a visual representation of the pattern repetetitions of a dataseries under analysis, as well as being computationally light and a mathematically (relatively) simple approach.
Thank you again for your question, if I have not given a great explanation here or you have further questions, feel free to reply back or send me an email (which you can find here).
Regards,
Patric