Evaluating the effectiveness of earthquake evacuation training compared to traditional methods with a multiplayer serious game

Abstract

Earthquakes are among the most challenging natural disasters to manage due to their unpredictable nature, the extensive destruction they cause and the vast geographies they affect. The seismically active location of Türkiye and the immense risk underscored by the February 6, 2023, Kahramanmaraş earthquakes have proven that human behavior and preparedness levels are as critical as the structural integrity of buildings in disaster management. Field research indicates that half of all injuries result from incorrect behaviors, while 34% of deaths occur during the evacuation process. From a cognitive psychology perspective, high stress during a disaster disrupts decision-making mechanisms, triggering "freezing" or "fight-or-flight" responses. This creates a "Knowledge-Action Gap" that prevents theoretical knowledge from being translated into practical reflexes under pressure. Existing disaster education practices remain insufficient to bridge this gap due to organizational costs, passive participant structures and their inability to simulate the social dimensions inherent in real evacuation processes, herd mentality, social proof and leadership seeking. This thesis aims to fill this critical void in the literature by comparatively examining the effectiveness of a multiplayer earthquake evacuation simulation developed with the Unity game engine against traditional seminar-based training methods. A systematic review of the literature conducted as part of this study revealed that more than 90% of the 14 core studies examined leave the user isolated in a virtual environment or limited to agents of Non-Player character. By moving disaster education beyond an individual process into a networked architecture that allows for real-time social interaction, this study empirically investigates the impact of group dynamics on learning outcomes. An experimental study was conducted with a control group receiving traditional seminar instruction and an Experimental Group engaging with the multiplayer serious game. Data were collected using knowledge tests, self-efficacy scales, perceived training effectiveness items, the DASS-21 and Beck Anxiety Inventory psychometric scales and automated behavioral logs retrieved via the Unity game engine, which served as objective ground truth. Regarding knowledge acquisition, both groups reached near-perfect scores at all measurement stages, producing what the literature defines as a "Ceiling Effect." No statistically significant between-group difference was detected, confirming that both methods are equally effective at reinforcing declarative safety knowledge for participants with prior earthquake training exposure. The more revealing finding lies beyond these scores: automated in-game logs demonstrated that theoretical knowledge does not guarantee enacted behavioral efficiency. Even participants who answered every knowledge question correctly attempted to move hundreds of times during the mandatory Drop-Cover-Hold phase. This behavioral manifestation of acute stress response was accompanied by a universal attempt to exit toward the evacuation route before the shaking phase had ended, revealing the precise point where procedural instinct and trained protocol diverge. The simulation provided a structured environment in which participants could encounter and correct these behavioral defaults safely. The most statistically robust finding of the study concerns perceived training effectiveness. The experimental group reported significantly higher overall perceived effectiveness compared to the control group (p = .024), with the strongest advantages in decision-making improvement (p = .001) and learning from mistakes (p = .041). These dimensions, specifically real-time decision-making under pressure and the revision of behavior through corrective feedback, are precisely those that passive instructional formats are structurally incapable of developing and which experiential, active-learning environments are specifically designed to address. The most original contribution of this study to disaster psychology literature is the contrasting effect of the two training methods on anxiety levels. The control group demonstrated a significantly larger reduction in BAI anxiety scores following training (p = .013), consistent with the deactivating nature of passive instruction. However, the present study interprets this complete anxiety resolution as a form of functional complacency rather than superior preparedness: an individual whose vigilance has been fully extinguished by a seminar enters a real earthquake without the physiological readiness that enables rapid protective action. The Experimental Group, by contrast, maintained a controlled residual activation state at both post-test and retention, demonstrating a pattern consistent with the Yerkes-Dodson principle that more closely approximates the arousal conditions of a real seismic emergency. Both groups' distress scores remained within non-clinical ranges throughout the study, confirming the psychological safety of the simulation. Finally, the social learning hypothesis, which represents the most innovative dimension of the thesis, was confirmed by the multiplayer experience data. The Vicarious Learning sub-dimension reached strong statistical significance (M = 4.07, p = .002), confirming that observing others' correct behaviors and witnessing others' mistakes functioned as primary channels of behavioral knowledge acquisition. The Social Presence sub-dimension also reached significance (M = 3.89, p = .047), while the Panic Contagion item fell below the neutral midpoint (M = 2.11), indicating that the social environment was experienced as informative and collaborative rather than anxiety-inducing. Six of nine participants explicitly adopted peer-following as their primary navigation strategy and collective responsibility emerged as the most consistently articulated theme across independent qualitative responses. These findings demonstrate that evacuation is a social phenomenon rather than an individual one and that multiplayer platforms are uniquely positioned to replicate the crowd dynamics, leadership seeking and group decision-making processes that define real-world emergencies. In light of these findings, the integration of multiplayer simulation technologies into disaster education curricula, particularly for digitally literate populations, is strongly recommended.