The association between synchrony and intellectual productivity in a group discussion: a study using the SenseChair

The study addresses the challenge of quantifying intellectual productivity in group settings such as meetings, brainstorming sessions, and hackathons. While such formats are intended to enhance creativity through the exchange of ideas, their performance is inconsistent, partly because intellectual productivity is difficult to measure numerically. Prior work has sought indices correlated with group intellectual productivity. Nonverbal synchrony (NVS)—an unconscious coordination of bodily movements during interpersonal communication—varies with conversational context and has been linked to creativity in dyadic interactions and rapport. In seated, time-limited discussions where overt gestures may not always occur, seated swaying can integrate subtle nonverbal cues, including behaviors tied to affect such as laughter or exclamations. Using SenseChair, a non-intrusive sensing chair measuring center of gravity and weight shifts, the study investigates whether group NVS, measured either as a global group index or as aggregated local (pairwise) indices, is associated with intellectual productivity in a general population (non-experts). The hypothesis is that higher NVS will be positively associated with intellectual productivity, and that pairwise NVS may better capture relevant dynamics than a strict global measure requiring all members to move in synchrony.

Related work on supporting group intellectual production includes interfaces that visualize meeting artifacts and real-time thematic overviews of conversation using speech and NLP (e.g., Doganata & Topkara, 2011; Chandrasegaran et al., 2019). While such systems aid access and summarization, they do not directly quantify group intellectual productivity, for which clear quantitative indicators are lacking. Research on group performance suggests a latent group intelligence factor linked to social sensitivity and conversational turn-taking (Woolley et al., 2010), but it requires prior psychometric measures and reflects member traits more than in-situ interaction dynamics. Nonverbal sensing approaches using microphones and cameras can extract cues such as pitch emphasis, conversational excitement, gaze, and gestures, but face limitations with silence, occlusion, installation burden, and user discomfort. NVS—an unconscious alignment of movement—has been associated with conversation content, smooth communication, and creativity in dyads (Condon & Ogston, 1966; Bernieri & Rosenthal, 1991; Bernieri et al., 1996; Won et al., 2014; Fujiwara et al., 2021). These motivate an unobtrusive, bodily movement-based indicator for group intellectual productivity. SenseChair offers a chair-based sensing approach to capture seated swaying as a proxy for nonverbal dynamics without making users aware of sensors, addressing limitations of audio-visual sensing.

Sensing device and signals: SenseChair is equipped with four pressure sensors (strain gauge-type, 50 kg capacity) mounted under the seat at the four corners, enabling acquisition of weight (W) and center-of-gravity (CoG) position (X: lateral; Y: anterior-posterior) at up to 100 Hz, transmitted via Bluetooth to a PC. CoG coordinates are computed from the four sensor readings (FL, FR, BL, BR) and seat geometry (sensor separations Lx, Ly). Weight is W = FL + FR + BL + BR. Sensors are mounted to distribute load and avoid saturating a single sensor; signals are amplified (~300x), filtered via bridge circuit, A/D converted, and transmitted. Data acquisition and preprocessing: During 15-minute discussions, CoG and weight fluctuation data were sampled at 20 Hz. Short-time Fourier transforms (STFT) were computed using a Hanning window of 256 samples (~10 s), sliding by one sample (overlap 255), yielding time-series amplitude spectra (0–10 Hz) and phase spectra per participant for X, Y, and W. The analysis focused on amplitude spectra; for each participant and movement dimension (MD in {X,Y,W}), the time-varying amplitude spectrum PU,MD(f,t) was normalized by the maximum within the first 180 s to account for individual movement magnitude differences and maintain sensitivity to smaller movers: PU,MD(f,t) = PU,MD(f,t) / max(PU,MD(f,t)) over the initial 180 s. A single per-participant spectral envelope Pu(f,t) was formed by taking the maximum across dimensions at each (f,t): Pu(f,t) = max(PU,X(f,t), PU,Y(f,t), PU,W(f,t)). NVS computation: Global NVS for an N=6 person team was computed as the product across participants: PGlobal(f,t) = Πk=1..6 Pk(f,t), which is high only when all members move concurrently, providing a conservative measure robust against long immobility. Pairwise NVS for any pair A,B was PpairAB(f,t) = PA(f,t) × PB(f,t). For each team, all 15 pairwise NVS time series were computed; summary statistics included the mean and variance of pairwise NVS over time. To leverage decomposition, pairwise NVS values were also ranked (highest to lowest) within teams to assess which ranks relate more to performance. Participants and task: Thirteen groups of six adult male participants (N=90; ages 20s–40s; mean 25.5, SD 7.1), strangers to each other, were recruited via a staffing agency. To reduce potential confounds from cross-sex dynamics and reported gender differences in synchrony, only men participated. Participants sat in SenseChairs, were informed of CoG/weight measurements and recording, and provided consent. Seating placed three participants facing the other three across a table. After self-introductions, teams performed a 15-minute alternative-use task: generate as many uses for a brick as possible. Guidelines: select one facilitator to initiate discourse; no standing; note-taking optional. Recordings used a 360-degree camera and two IC recorders. Outcomes (proxy for intellectual productivity): For each team: (1) number of ideas; (2) number of idea categories (posterior classification by two experimenters); (3) originality score computed as uniqueness across teams (idea seen in 1/13 teams=3 pts; 2/13=2 pts; 3/13=1 pt; ≥4/13=0 pts), summed within team. Analyses computed Spearman’s rank correlations between performance metrics and (a) mean global NVS, (b) mean and variance of pairwise NVS (averaged over the task). A shuffle reanalysis assessed robustness: participants were randomly shuffled to create 13 pseudo-teams; for each, average pairwise NVS was computed and correlated with original teams’ idea counts; repeated 1000 times to estimate probability of observing the initial correlation by chance.

- Pairwise vs global NVS dynamics: Despite small fluctuations in global NVS, pairwise NVS showed wider variation and peaks aligned in trend with global peaks. Correlation between global NVS and aggregated pairwise NVS was high (r = 0.945). - Global NVS correlations with performance (Spearman): - Average global NVS vs number of ideas: ρ = 0.377, p = 0.204 (ns) - Average global NVS vs number of idea categories: ρ = 0.154, p = 0.614 (ns) - Average global NVS vs idea uniqueness: ρ = 0.363, p = 0.223 (ns) - Pairwise NVS correlations with performance (Spearman): - Mean pairwise NVS vs number of ideas: ρ = 0.543, p = 0.055 (marginal, positive) - Mean pairwise NVS vs number of idea categories: ρ = 0.284, p = 0.347 (ns) - Mean pairwise NVS vs idea uniqueness: ρ = 0.374, p = 0.208 (ns) - Variance of pairwise NVS vs number of ideas: ρ = 0.408, p = 0.167 (ns) - Variance of pairwise NVS vs number of idea categories: ρ = 0.452, p = 0.121 (ns) - Variance of pairwise NVS vs idea uniqueness: ρ = 0.290, p = 0.336 (ns) - Shuffle reanalysis (1000 iterations): Probability of observing correlation ≥ 0.543 between average pairwise NVS and idea count by chance ≤ 0.044, supporting that the observed association reflects actual member interactions. - Within-team rank analysis of pairwise NVS: Higher-ranked (maximum) pairwise NVS values showed a tendency for negative correlation with idea originality, suggesting excessively high pair synchrony may reduce originality (e.g., large simultaneous actions or off-task movement). Lower-ranked (minimum) pairwise NVS values showed a weak positive correlation with number of ideas, suggesting that higher minimum synchrony within a team may relate to producing more ideas.

Findings indicate that pairwise NVS—reflecting local coordination between team members—is more informative for predicting idea generation than a global NVS measure that requires all members to synchronize. The marginal positive correlation between mean pairwise NVS and idea count, supported by shuffle analysis, suggests that frequent, distributed dyadic coordination aligns with more active idea exchange. Global NVS’s low variability may mask meaningful dynamics given its conservative construction and attenuation as group size increases. Within-team analyses suggest nuanced effects: very high synchrony in specific pairs may reflect behaviors not conducive to originality, whereas maintaining a higher lower bound of synchrony across pairs may facilitate broader participation and idea proliferation. These results support focusing on dyadic (and potentially small-cluster) synchrony within groups to understand and potentially enhance intellectual productivity. The work complements prior findings reporting null associations for whole-group NVS and opens avenues to explore synchronization beyond pairs (e.g., triads or clusters) and its temporal emergence in relation to task phases and outcomes.

The study presents a method to evaluate group intellectual productivity via nonverbal synchrony derived from seated body sway using SenseChair. Analyzing 13 six-person groups performing a divergent ideation task, the mean pairwise NVS showed a marginally significant positive association with the number of ideas generated, whereas global NVS did not significantly correlate with performance metrics. Results suggest that aggregations of local (pairwise) synchrony better capture discussion dynamics relevant to ideation output than a stringent global synchrony measure. A shuffle-based reanalysis supported that the observed association is unlikely due to chance. Future work will examine temporal dynamics of pairwise NVS across task phases and refine measures of idea quality, aiming toward a robust NVS-based evaluation of group intellectual productivity.

- Sample size and power: Only 13 teams (N=90 participants) led to mostly non-significant results; the key correlation (ρ=0.543) was marginal (p=0.055). - Participant demographics: All-male, non-expert participants; limits generalizability across genders, professions, and mixed-gender dynamics. - Measurement modality: NVS measured solely via seated oscillations (CoG and weight shifts); excludes other nonverbal channels (e.g., nodding, hand gestures, gaze) that may capture additional dynamics. - Global vs local measure sensitivity: Global NVS is conservative and low-variance, potentially obscuring meaningful interaction patterns; pairwise emphasis may not capture larger synchronization clusters. - Outcome assessment: Idea originality defined by cross-team overlap counts; may not reflect broader or domain-specific quality; limited number of evaluators and criteria. - Temporal aggregation: Analyses averaged NVS over entire 15-minute tasks, not differentiating phases (early/middle/late) or idea-generation moments. - Potential confounds: Individual traits (e.g., leadership, personality) and roles not explicitly modeled; seating configuration and facilitation differences may affect synchrony. - Experimental constraints: Prohibition on standing and lab setting may differ from natural workplace meetings; note-taking optionality not analyzed in relation to NVS.