Team conversations are central to collaborative science, influencing both the quality of discourse and the formation of productive working relationships. Understanding the underlying dynamics that govern who speaks, when, and for how long is critical to fostering equitable participation and effective team functioning. In this work, we propose and validate a simple, mechanistic mathematical model for predicting speaking time and turn-taking patterns in group discussions. Our model is based on the idea that social cues play a major role in group conversations and that turn-taking is primarily driven by participants’ intrinsic desire to speak rather than by the content of the conversation.

We apply this model to a rich dataset of recorded conversations from the Scialog conference series, which brings together early-career scientists to form interdisciplinary collaborations. Using audio recordings from over 100 breakout discussions across multiple conferences, we extract turn-taking data and analyze patterns of participation. Our model employs minimal assumptions, treating conversation as a dynamic competition for airtime among individuals driven by their intrinsic desire to be heard. We show that this model, despite its simplicity and independence from conversational content, can accurately predict individual speaking times and the overall structure of interactions.

We further tested the predictive power of our model through preregistered experiments at the 2024 Scialog Molecular Basis of Cognition (MBC3) conference. Using data from the prior year’s conference to infer model parameters, we forecast speaking times, speaker rankings, and turn-switch structures for the subsequent meeting. Our model demonstrated strong predictive power, outperforming baselines in rank prediction and overall speaking time estimation. The model also provides new insight into the importance of social cues and stable personality-like traits in guiding conversational dynamics.

Our findings suggest that group conversation, even in scientific settings, is governed as much by social signaling as by information exchange. This insight has implications for designing team environments that encourage balanced participation and for developing tools to facilitate equitable dialogue. Moreover, understanding these patterns can support team formation processes, as individuals’ conversational behavior may influence collaboration choices.

This work advances the science of team science by providing a quantitative framework for predicting conversational behavior in group settings and offering evidence that simple models can capture complex human interaction dynamics. Future work will explore extensions of this model to incorporate factors such as facilitator/moderator influence and interest in the discussion topic.