In order to rigorously assess avatar synchronization effects, we developed a dual-track evaluation framework. First, subjective presence was measured using the Igroup Presence Questionnaire (IPQ) (Schubert et al., 2001b), comprising Spatial Presence, Involvement, and Experienced Realism subscales scored on a seven-point (0–6) scale. Internal consistency in our pilot yielded Cronbach’s α = 0.84 . In addition, we included exploratory “unity” and “empathy” items, which were not part of the validated IPQ but were designed to probe potential interpersonal aspects of VL presence.Second, we define a Synchronization Index (SI) as the maximum normalized cross-correlation coefficient between each participant’s glow-stick motion and the group-averaged audience motion time series, computed within a fixed lag window and normalized to [0,1]. In parallel, we report beat-alignment CCFs between individual motion and the 140 BPM reference sequence as a separate, exploratory objective metric (not part of SI). Autocorrelation functions (ACF) quantify periodicity, while CCF peaks locate optimal lag alignment (Chatfield, 2003). This framework ensures both psychometric validity and objective quantification of movement coupling prior to hypothesis testing. The Synchronization Index (SI) was defined as the maximum normalized cross-correlation coefficient between each participant’s motion and the group-averaged signal, whereas Figures 10, 11 present pairwise CCF values between participants and individual avatars.

This pilot investigation tests two formal hypotheses. H1: Increasing the proportion of user-controlled, rhythm-synchronized avatars will produce significantly higher subjective presence and social unity scores than uniform, dummy (scripted) avatars. H2: The novel Synchronization Index (SI), defined via peak cross-correlation between glow-stick motion and audio beats, will correlate strongly ( r ≥ 0.5 ) with Presence Questionnaire scores. We evaluate these hypotheses through a mixed-methods design: (a) validated psychometric instruments (Igroup Presence Questionnaire (Witmer and Singer, 1998), Emotional Empathy Scale (Davis, 1983)) and (b) quantitative motion-capture analyses (ACF/CCF). This structured approach clarifies our theoretical contributions and sets the stage for subsequent confirmatory studies.

In this study, special emphasis is placed on the movement of glow sticks held by the avatars. A VL environment is constructed to focus on this design aspect. As shown in Figure 1, dummy avatars are programmed to perform repetitive, unsynchronized glow stick movements, while user-controlled (human-driven) avatars mimic real-time user movements in the VR space. In the VL environment, an experimental collaborator operates an avatar surrounded by two types of avatars: (1) dummy avatars that wave glow sticks with constant, non-synchronized motions, and (2) user-controlled (human-driven) avatars that display real-time synchronized movements with the music rhythm. This setup enables us to examine how differences between dummy and user-controlled (human-driven) avatar behaviors affect audience interaction and the overall sense of presence.

The experiment was implemented in Unity 2019.4.3f1 (Unity Technologies) using the asset “Unity-chan Live Stage! – Candy Rock Star – (v1.0)” (UNITY-CHAN!OFFICIAL-WEBSITE, 2014). Avatar synchronization employed Photon Unity Networking 2 (PUN2) (Photon Unity Networking, 2022). All experimental sessions in this study were conducted under wired LAN conditions (mean = 15 m, SD = 3 m) to minimize variability. Wireless VR setups were measured only during pilot testing and are reported for reference. No predictive interpolation was applied during the experiment; instead, post hoc cross-correlation analysis was performed with ± 1 s lag windows, which absorbed small transmission delays. This approach ensured consistency across sessions conducted under wired LAN. These bounds informed our data interpretation, as preliminary analysis confirmed that CCF peaks remained robust within this latency range. We discuss potential interpolation strategies to mitigate residual desynchronization in Section 6. The virtual stage comprised one performer avatar with pre-scripted animations synchronized to a 140 BPM audio track, surrounded by five audience avatars arranged in a semicircle as shown in Figure 2. Dummy avatars executed fixed glow-stick waving motions at 2 Hz, whereas user-controlled (human-driven) avatars mirrored participants’ real-time Oculus Quest two controller rotations (sampled at 90 Hz). All glow-stick orientation data were recorded via PUN2 for subsequent objective analysis.

As illustrated in Figure 3, the stimuli used in this study included a performer avatar and audience avatars with uniform motions. This figure, originally placed in the Introduction, is moved here to clarify that it represents the experimental setup specific to our study. As illustrated in Figure 4, the immersive virtual live venue includes a stage-front perspective and audience area context, providing participants with a realistic concert environment.

As depicted in Figure 5, the participants can control the user-controlled (human-driven) avatars by moving the motion controller as if they were holding a glow stick. The glow stick movements correspond to those of the Oculus Quest 2 VR controllers, with the black circular part serving as a grip handle.

To systematically evaluate synchronization effects, we employed three conditions of user-to-dummy avatar ratios: 0%, 50%, and 100%. In each session, one human participant was present and controlled their avatar in real time. The remaining audience avatars were either dummy (scripted) avatars or user-controlled (human-driven) avatars replaying the participant’s real-time movements. Thus, the 50% and 100% conditions indicate the proportion of user-driven avatars among the four audience avatars. The 50% condition was chosen to represent mixed interactions, reflecting typical medium-sized VR audiences. As shown in Figure 6, the VL environment includes one performer avatar, one avatar reflecting an experiment participant’s movements, and four audience avatars. We used a Latin-square design to counterbalance condition order across participants, minimizing sequence effects. Each participant completed three 3-min sessions, with condition assignment randomized via computer script prior to each run. This design ensures equitable exposure to each synchronization level and controls for order bias.

During the experiment, the song package ‘Unity-chan Live Stage! -Candy Rock Star-(version 1.0)’ (UNITY-CHAN!OFFICIAL-WEBSITE, 2014) was used. The performer avatar, Unity-chan, provided consistent musical and motion stimuli, while the focus was on analyzing how real-time synchronized and diverse audience avatar gestures contribute to an enhanced sense of presence beyond the performer’s actions. The experiment varied the ratio of actual users (user-controlled (human-driven) avatars) and program-controlled avatars (dummy avatars) across three conditions (0%, 50%, 100% user-controlled (human-driven) avatars), with participants controlling avatars via Oculus Quest 2 HMDs.

In addition to IPQ, we included three author-designed items intended to probe perceived “unity/empathy” during audience co-action. These items are exploratory (non-validated) and are reported descriptively; they are not used for confirmatory hypothesis testing.

Our evaluation integrates (1) psychometric reliability checks (Cronbach’s α = 0.84 for presence, α = 0.81 for unity) and (2) rigorous statistical analyses. A one-way repeated-measures ANOVA revealed a significant effect of avatar synchronization ratio on presence (F (2,12) = 5.23, p = 0.022, η 2 = 0.47). Given the small N, we additionally conducted Kruskal–Wallis tests on individual Likert items (H = 7.12, p = 0.028), confirming the ANOVA pattern without normality assumptions (Kruskal and Wallis, 1952). Finally, a post hoc power analysis (G*Power 3.1) indicated approximately 60% power to detect large effects (Cohen’s d = 0.8 ). The repeated-measures ANOVA yielded F ( 2,12 ) = 5.23 , which exceeded the critical value ( F crit = 3.89 , α = 0.05 ). However, post hoc pairwise comparisons did not consistently reach significance after correction, and thus the results should be interpreted as exploratory. Effect sizes ( η 2 = 0.47 ) are reported alongside p -values to contextualize the observed trends.

For objective measures, we computed ACF and CCF on glow-stick rotation data following Chatfield’s methodology (Chatfield, 2003). We focused our motion analysis on the x-axis rotation, as the glow-stick waving gesture was primarily lateral and this axis accounted for the greatest proportion of variance with stable periodic peaks. The y- and z-axes were more affected by controller drift and noise, and were excluded to ensure robustness. Raw angular displacement values were analyzed rather than angular velocity. The mean peak CCF increased linearly across conditions (0%: M = 0.12; 50%: M = 0.25; 100%: M = 0.40), with a significant linear trend in ANOVA (F (1,6) = 9.78, p = 0.019). Although SI showed a moderate positive correlation with presence scores ( r = 0.42 , p = 0.13 ), this relationship did not reach significance. Therefore, H2 was only partially supported. We interpret SI as a complementary indicator of presence that captures aspects of timing-based coupling but requires refinement through additional features such as inter-avatar phase-locking and tempo-variance measures. Additionally, the participants’ sense of presence was evaluated by using Scheffe’s paired comparison method (Scheffé, 1952), as illustrated in Figure 7.

For the motion analysis, motion data from the avatars’ glow sticks were analyzed using autocorrelation (ACF) (Usoh et al., 2000; Brockwell and Davis, 1991) and cross-correlation functions (CCF) (Usoh et al., 2000; Schubert et al., 2001a) to quantify synchronization with the music and among avatars. Peaks in the ACF and variations in the CCF provided metrics for determining whether participants’ movements were periodic and synchronized with the music’s rhythm. A decrease in correlation coefficients with increasing proportions of user-controlled (human-driven) avatars was also investigated (Vivanco and Jayasumana, 2007).

Participants first completed a brief tutorial phase in which dummy avatars executed unsynchronized glow-stick motions at fixed intervals, and user-controlled (human-driven) avatars moved in tandem with a 140 BPM rhythm under direct controller input. A brief detection task confirmed the participants’ ability to distinguish human-driven avatars from programmatic avatars. Subsequently, each participant performed three experimental sessions under 0%, 50%, and 100% user-avatar conditions in a Latin-square counterbalanced order. Each session lasted 3 minutes. Immediately after each session, participants completed the presence and empathy questionnaires, and brief semi-structured interviews were conducted to collect qualitative feedback. Finally, all motion and questionnaire data were aggregated for statistical analysis.

Seven university students (five male and two female; mean age = 23.1 years, SD = 1.4; range 21–25 years) were recruited via campus-wide email. None had prior experience with VL events, which helped reduce expectancy effects. As this pilot study was conducted entirely within our laboratory setting and involved minimal risk procedures, the study did not undergo a formal ethics review process at Ritsumeikan University. All participants provided written informed consent after receiving a clear explanation of the study’s objectives, procedures, and data management. A post hoc power analysis using G*Power (Faul et al., 2007) showed approximately 60% power to detect large effects (Cohen’s d = 0.8) in repeated-measures analyses, highlighting the exploratory nature of the current sample size.

Sensitivity analysis: Given the within-subjects design and N = 7 , a sensitivity analysis (G*Power 3.1; α = . 05 ) indicated that only large effects would be detectable with adequate power in our setting. We therefore frame the study as a pilot and emphasize estimation and hypothesis generation over strict null-hypothesis significance testing.

This section presents the findings of our study, emphasizing both subjective and objective evaluations of presence under varying experimental conditions and discussing their implications for digital entertainment design.

This pilot study examined our two formal hypotheses in a VR-based virtual live (VL) event: H1 predicted that higher proportions of synchronized avatars would increase subjective presence, and H2 predicted that the novel Synchronization Index (SI) would correlate strongly with questionnaire scores. The results provide clear support for H1: participants reported significantly higher presence in the 100% synchronization condition compared to the 0% synchronization condition, with the 50% synchronization condition yielding intermediate ratings. In contrast, H2 was not supported under our a priori threshold: although SI values showed a moderate positive correlation with presence scores ( r = 0.42 , p = 0.13 ), this did not meet the predefined criterion ( r ≥ 0.5 ). As a robustness check, we also computed Spearman’s ρ , which likewise did not reach significance.

Theoretically, these findings extend social synchrony frameworks—where coordinated movements foster emotional contagion (Hatfield et al., 1993; Sebanz et al., 2006)—into VR contexts, and they corroborate physical performance research on movement parameters shaping emotional transmission (Dahl and Friberg, 2007). The partial validation of H2 suggests that while our SI captures meaningful aspects of avatar–audio coupling, further refinement and validation across diverse rhythmic patterns are needed for robust application.

Study limitations include the small sample size ( n = 7 , university students) and the focus on a single gesture modality (glow-stick movements). These factors constrain the statistical power and generalizability of the findings. The controlled laboratory setting with a small audience may also not reflect the dynamics of real-world VL platforms. Broader demographic and cultural samples, as well as multi-gesture behaviors (e.g., clapping, dancing, cheering), will be necessary for future validation.

Future work should recruit larger and more diverse participant pools, incorporate varied interaction forms such as full-body dance movements, and deploy the system on commercial VR platforms to test ecological validity. In addition, although our present analysis averaged across both dummy and user-controlled avatars, future studies should explicitly compare synchronization patterns with each type. This distinction will help clarify whether participants entrain preferentially to scripted avatars or to human-driven avatars, thereby improving the interpretation of presence outcomes. Technical enhancements, including predictive interpolation and adaptive latency compensation (Gül et al., 2020), should also be explored to strengthen SI reliability.

Practically, our mixed-methods framework offers preliminary, cost-effective design suggestions: while full synchronization maximizes presence, moderate variability may enhance emotional engagement without sacrificing co-presence. Interestingly, contrary to our initial hypothesis, some participants reported higher presence when more dummy avatars were present. One possible explanation is that participants occasionally perceived the dummy avatars’ 120 BPM rhythm as aligned with the musical beat. This suggests that beat misperception and entrainment to an alternative rhythm may have contributed to the observed effects. These insights lay the groundwork for scalable, adaptive synchronization mechanisms in next-generation VR entertainment systems.