Introduction

In recent years, the rapid development of China’s high-speed railroad has brought great convenience to people’s travel. More and more people choose high-speed rail as a way of traveling. However, this great passenger flow will bring pressure to the railway station. Emergency evacuation is the key to guaranteeing people’s lives and property safety in this crowded public transportation. In addition, the behavior of passengers during the evacuation process will directly affect efficiency and safety. Non-adaptive emergency evacuation behavior, specifically, the behavior of individuals in the evacuation process to achieve their own purposes to harm others and hinder the group evacuation1. It even increases the risk and complexity of the evacuation process. Moreover, it will seriously affect the efficiency.

At present, non-adaptive evacuation behavior has received extensive attention in the field of emergency evacuation research. Wang Yanyun et al.2 applied the KAP theory to reveal the effects of personal characteristics and safety attitudes on the non-adaptive evacuation behavior of subway passengers. It was also found that gender, age, and prior fire experience significantly affected non-adaptive behavior. Wang et al.3 revealed the remarkable effects of stressful environments and fire severity on non-adaptive behavior by constructing a group panic scale. And proposed optimizing evacuation efficiency by managing group panic. Based on the attitude-behavior process model, Qi Xiaoyun and Liu Jie4 studied the relationship between passengers’ attitudes, risk perception, and non-adaptive evacuation behaviors in airport fires. He Yerong et al.5 pointed out that non-adaptive behavioral motivation is the key to preventing and controlling the occurrence of non-adaptive evacuation behaviors by studying the dynamic evolution of non-adaptive evacuation behaviors under emergencies through the system dynamics model. A dynamic security atmosphere is an important factor in inhibiting non-adaptive evacuation behavior. Yang et al.6 pointed out that risk communication has a negative effect on non-adaptive evacuation behavior. Shiwakoti et al.7 found that males are more likely to adopt competitive behavior than females during emergencies, while female passengers are more likely to exhibit conformity behavior. Cheng Fangming et al.8 explored the relationship between security risk perception and non-adaptive evacuation behavior through theoretical models and questionnaires. Kinateder et al.9 and Mackay10 explored risk perception and evacuation behavior in a fire evacuation.

In summary, while current research on non-adaptive evacuation behavior has primarily explored influencing factors from individual characteristics and risk perception, a significant gap exists concerning the psychological mechanisms through which spatial perception influences behavior, especially within the unique context of railway stations. This gap is critical, as recent studies on evacuation in railway stations, while valuable, have predominantly focused on the simulation and optimization of passenger flow, often modeling behavior without deeply investigating its cognitive antecedents11.

These simulation studies rightly identify modern railway hubs not as simple buildings but as complex infrastructural systems with multi-level structures and intricate layouts12,13. However, this very complexity creates significant psychological challenges for passengers. Research in environmental psychology confirms that such complex architectural environments can provoke adverse reactions like anxiety and stress, which negatively impact a passenger’s sense of safety14. For many travelers, particularly those unfamiliar with the station, these are temporary spaces navigated under time pressure. Lacking a pre-existing cognitive map, these passengers must rely on an “inference strategy”—using memories of similar environments to navigate—a process that is cognitively demanding and prone to error15. While existing evacuation models effectively simulate crowd movement, they have paid less attention to how these individual cognitive and emotional pressures—born from navigating a complex space—translate into the group-level panic and non-adaptive behaviors that are central to this study. Therefore, understanding this psychological pathway is essential for developing truly effective evacuation strategies and represents a significant and underexplored area in public safety research.

This paper is based on the environment-behavior theory and the Stimulus-Organism-Response (S-O-R) theory. The S-O-R framework posits that environmental stimuli (S) influence an individual’s internal cognitive and emotional state (O), which in turn drives their final behavioral response (R). This framework is applied to investigate the influence of passengers’ spatial perception on non-adaptive evacuation behavior in railway stations. Group panic and security atmosphere are introduced as mediating variables to explore the interactions among these factors through a structural equation model. The goal is to provide directional advice for reducing non-adaptive evacuation behavior and developing effective evacuation guidance strategies to improve overall evacuation efficiency.

Non-adaptive evacuation behavior

In emergencies, people will produce stress psychology and trigger negative emotions. To ensure the safety of their own lives or property under certain emotions, people may make irrational behavior8. Such as inertia behavior, risk-taking behavior, competitive behavior, return behavior and companion behavior. These behaviors occurring in the evacuation process that are detrimental to the safety of others are collectively known as non-adaptive evacuation behaviors, which are one of the main causes of catastrophic accidents in crowds16. Therefore, reducing the non-adaptive evacuation behavior of passengers in railway stations is an important way to improve the evacuation efficiency of railway stations effectively.

Spatial perception

According to Kaplan’s17 environment-behavior theory, individuals acquire information about the environment through exploration and interaction, thereby forming a cognitive map. This cognitive map helps individuals to locate and select escape routes in emergency situations quickly. Spatial perception is a person’s intrinsic response to space, an awareness of the relationship between the surrounding environment and oneself18. Sensation and perception are the basis for the formation of spatial perception. Sensation is the way to obtain external information, while perception is the further processing of this information. Among them, the processing of information is especially critical in emergency evacuation. Cognitive theory states that individuals process external information through figurative and abstract cognition, including spatial organization and interpretation. This can directly affect an individual’s spatial perception ability in emergencies19. Therefore, reduced spatial perception may lead to irrational evacuation decisions in emergencies. As a result, non-adaptive evacuation behavior occurs.

According to environmental behavioral science, there is an interaction between environment and behavior. Spatial perception is not only affected by the shape of the environment, but also can in turn affect the behavioral responses of individuals. When studying the emergency evacuation behavior of people in railway stations, Li Shuqing20 et al. pointed out that spatial perception positively affects emergency evacuation behavior. Railway and subway stations are characterized by multi-level buildings and complex structures. Good spatial perception is crucial for recognizing safe paths and emergency exits. It can effectively reduce panic and misleading behavior and improve evacuation efficiency20. The process of perceiving and comprehending the spatial structure, size, direction and distance and other attributes of the railway station involves extracting and processing information from the railway station environment. Ultimately, the formation of mental images affects travelers’ evacuation behavioral decisions21.

H1

Spatial perception has a significant negative effect on non-adaptive evacuation behavior.

Group panic

Group panic is defined as a catastrophic psychological state that occurs in a crisis situation. This psychological state is a negative emotional response involving multiple influences. This can result in the creation of high-risk behaviors3. Wang considered group panic as a leading factor in the occurrence of non-adaptive evacuation behaviors. Evacuation efficiency should be optimized by managing group panic. In densely populated public places such as railway stations, emergencies are prone to trigger group panic. Due to emotional contagion among group members, panic can spread rapidly. It exacerbates the chaos of the evacuation process and affects the cognition and behavior of most travelers. For example, group panic may cause travelers to concentrate their attention on escape in emergencies such as fire. This would create a group convergence effect and neglect the safety exits and evacuation paths, resulting in non-adaptive evacuation behaviors.

Spatial perception is an individual’s cognition and comprehension of the surrounding environment, which is closely related to group panic. Clearer spatial perception helps individuals to better recognize the safe evacuation paths and emergency exits. This reduces group panic and the occurrence of non-adaptive evacuation behavior. S-O-R is a model in cognitive behavioral science. Based on the S-O-R model, spatial perception as a stimulus (S) affects the individual’s psychological response (O), resulting in the generation of group panic, which in turn affects the individual’s response (R) and influences the evacuation decision judgment. This ultimately leads to non-adaptive evacuation behavior.

Based on the above analysis, the hypothesis is proposed:

H2

Spatial perception has a significant negative influence on group panic.

H3

Spatial perception influences non-adaptive evacuation behavior through the mediating role of group panic.

H4

Group panic has a significant positive effect on non-adaptive evacuation behavior.

Safety climate

A safe atmosphere is a true reflection of the safety conditions of an individual’s environment22. According to the environment-behavior theory, the security atmosphere as a factor of the environment influences the safety behavior of individuals23. In a public place such as a railway station, a security atmosphere will promote travelers’ compliance with emergency evacuation regulations. It will also affect travelers’ participation in evacuation safety22. And it can improve travelers’ sense of security and enhance their trust in the environment. This reduces the occurrence of panic and non-adaptive evacuation behavior in emergency situations. In the case of perceived high-risk and insufficient security atmosphere, travelers are more likely to panic. Thus, impulsive or irrational evacuation decisions are taken. On the contrary, a clear and supportive security atmosphere can effectively reduce such behaviors and motivate travelers to adopt a more rational and practical evacuation path.

In addition, there is a strong link between safety climate and spatial perception. Safety climate can play a mediating role between spatial perception and non-adaptive evacuation behavior. Ye Gui24et al. found a significant correlation between safety perception and unsafe behavior. External stimuli can also ultimately affect safety behavior by influencing human perception and cognition. According to the S-O-R model, sound spatial perception increases the security atmosphere of travelers. This leads to the formation of favorable behavioral decision-making, which in turn reduces the occurrence of non-adaptive evacuation behaviors. In response to the above analysis, the following hypotheses are proposed:

H5

Spatial perception has a significant positive effect on security atmosphere.

H6

Spatial perception influences non-adaptive emergency evacuation behavior through the mediating role of security atmosphere.

H7

Safety climate has a significant negative effect on non-adaptive evacuation behavior.

The relationships hypothesized above are visually summarized in the theoretical model shown in Fig. 1. The model’s central purpose is to test how passengers’ Spatial Perception influences their Non-adaptive Behavior. It posits a direct negative influence of Spatial Perception on Non-adaptive Behavior (H1) and Group Panic (H2), and a direct positive influence on security atmosphere (H5). Furthermore, the model introduces two key mediating variables: Group Panic and security atmosphere. It is hypothesized that Group Panic positively influences Non-adaptive Behavior (H4) and acts as a mediator (H3). Conversely, security atmosphere is hypothesized to negatively influence Non-adaptive Behavior (H7) and acts as a mediator (H6). Collectively, this model provides a comprehensive framework for testing both the direct and indirect pathways through which passenger spatial perception affects their evacuation behavior.

Fig. 1
figure 1

Theoretical model.

Full size image

Research design

Questionnaire design

This paper is based on the existing scale and combines the characteristics of the railway station to modify the design of the questionnaire. The questionnaire measures the relationship between spatial perception, group panic, security atmosphere and non-adaptive emergency evacuation behavior. The questionnaire is in two sections: ① Influencing factors of non-adaptive evacuation behavior of passengers in railway stations and their measurement variable; ② Basic information of the investigated persons. The spatial perception measurement refers to the studies in literature20 and literature25. According to the characteristics of the spatial environment of the train station, seven subjects were designed from the three aspects of familiarity perception, information perception, and spatial layout perception. security atmosphere Reference22 designed five questions based on three aspects of personnel security atmosphere, management security atmosphere and infrastructure security atmosphere. Group Panic Measurement Reference8 Designed 5 measures based on 4 aspects: information recognition ability, group convergence effect, emergency environment stimulation and accident severity. Non-adaptive Emergency Evacuation Behavior Reference8 designed four topics in three categories: herd behavior, competitive behavior, and turnback behavior.

The individual attributes of the respondents should be collected from the demographic information of the passengers. General attributes such as passengers’ knowledge of emergency evacuation and frequent participation in evacuation drills should be asked. Parameters include gender, age, education level, and emergency experience. Specific questions are shown in Table 1. The questionnaire consists of six sections. The first part is personal information. The second part investigates the spatial perception dimension. The third part is from the risk perception dimension. Part four is surveyed from the dimension of the security atmosphere. Part five investigates the group panic dimension. Part six is surveyed from the dimension of non-adaptive emergency evacuation behavior. Likert 5-point scale26 was used from part 2 to part 6, with 1 to 5 indicating very non-compliant, non-compliant, fair, compliant, and very compliant, respectively.

Table 1 Measurement question items for model latent variables.
Full size table

Sample and data collection

This paper adopts the online questionnaire, using the “WenJuanXing” platform to distribute the official questionnaire. Finally, 342 questionnaires were collected, and 284 were valid, with an effective response rate of 83.04%. As shown in Table 2, in terms of gender, 97 men accounted for 34.15% and 187 women accounted for 65.85%. Regarding age, 8 people aged 18 and below accounted for 2.81%. 188 people aged 19 to 35 accounted for 66.19%. 72 people aged 35 to 50 accounted for 25.35%. 12 people aged 50 to 60 accounted for 4.22%, and 4 people aged 60 and above represented 1.61%. In the case of education level, 10 people were 3.52% in junior high school and below, 29 people were 10.21% in high school or junior college, 193 people took up 67.95% in bachelor’s degree or junior college, and 52 people occupied 18.30% in master’s degree and above. In the knowledge level of safety and emergency response, the number of people who are very unaware is 3, accounting for 1.05%. The amount of people who do not understand is 26, accounting for 9.15%. The 82 people, or 28.87%, had a general understanding. The figure of 143 people (50.43%) is aware of it. 30 people understood it very well, with 10.56%. The total number of people with evacuation experience is 260 (91.54%), and the number of people without evacuation experience is 24 (8.45%). The sample quality of this research is favorable and has a good representation.

Table 2 Basic information on investigated persons.
Full size table

Reliability analysis

In this paper, exploratory factor analysis and validation factor analysis of questionnaire reliability and validity were conducted using SPSS 26.0. Before factor analysis, the internal consistency coefficient (Cronbach’s α) was calculated to test the internal reliability of each factor. When the value of Cronbach’s α is greater than 0.7, it indicates a high internal consistency of the measurements and a high degree of reliability. Exploratory factor analysis refers to the KMO test and Bartlett’s spherical test of the questionnaire. The closer the KMO value is to 1, the stronger the correlation between the variables and the higher the validity. The test results are shown in Table 3. The Cronbach’s α coefficients of each latent variable in are higher than 0.7. This indicates that the measurement scale in this paper is reliable. KMO and Bartlett’s statistics were used to test the validity of the questionnaire scale, and the KOM test value was 0.889, which was greater than the standard value of 0.700. The significance level Sig. value was 0.000 < 0.005, which indicated that the correlation between the variables was high, and it was suitable for factor analysis.

Table 3 Results of reliability and validity tests.
Full size table

SPSS26.0 was used to conduct exploratory factor analysis on the sample data. The maximum variance method was applied to rotate the variables. Then 4 variables were selected that could be used as principal factors. As shown in the Table 4 is the result of principal component variance interpretation after scale rotation, the initial eigenvalues of the four variables are greater than 1. The cumulative variance ratio is 54.562%, which is greater than 50%, indicating that the scale dimensionality division is basically reasonable.

Table 4 Results of principal component variance interpretation after scale rotation.
Full size table

The common factors were extracted by orthogonal rotation method. Table 5 is the factor analysis rotated component matrix, including four components and loading values. Among them, the loading values of items C1 and D1 are less than 0.4, and items C1 and D1 are deleted. The loadings of the remaining items are all greater than 0.4, which basically meets the requirements of model validation.

Table 5 Factor analysis rotated component matrix.
Full size table

Validated factor analysis refers to the use of combined reliability (CR) and average variance extracted (AVE) to evaluate the convergent validity of the questionnaire. The CR value is greater than 0.7, indicating that the set of question items can consistently explain the corresponding latent variables. The AVE value is greater than 0.5, indicating that the questionnaire scale has a good convergent validity. The CR value is greater than 0.7, and the AVE value is larger than 0.5, which means that the scale has a good convergent validity. A validated factor analysis was performed utilizing AMOS 28.0. As shown, the SMCs of question items A6, B2, B5, C1, C5, D1, D5, and D8 were 0.33, 0.34, 0.30, 0.31, 0.32, and 0.16, respectively, which were less than 0.36. They did not fulfill the requirement and will be deleted. Analyze the model after deleting the nonconforming question items. The mean-variance extraction scale AVE was obtained. All of them were greater than 0.400 and reached the acceptable range. The scale combination reliability CR values, were all greater than the standardized value of 0.700, indicating that the questionnaire scales have favorable convergent validity. Meanwhile, according to the conclusion of the literature27, distinguished validity means that the correlation coefficient between the latent variables is less than the open square of the AVE value. It demonstrates that there is a correlation between the variables. Table 6 presents the results of discriminant validity testing, which evaluates the independence between latent variables by comparing their correlation coefficients with the square roots of their corresponding Average Variance Extracted (AVE) values (bold values along the diagonal). According to the validation criteria, discriminant validity is established if the correlation coefficient between two latent variables is smaller than the minimum of their respective AVE square roots. For instance, the correlation coefficient between Spatial Perception and security atmosphere (0.641) is lower than their AVE square roots (0.653 and 0.718, respectively), confirming discriminant validity. Similarly, the correlation coefficient between Group Panic and Non-Adaptive Evacuation Behavior (0.505) is below the smaller AVE square root (0.690 vs. 0.801), meeting the requirement. Although the correlation between Spatial Perception and security atmosphere (0.641) approaches its AVE square root (0.653), it remains within acceptable limits. All correlation coefficients between latent variables are lower than the minimum AVE square roots of their corresponding pairs, demonstrating distinctiveness in measurement and robust discriminant validity. Additionally, the Composite Reliability (CR) values exceed 0.7, further supporting the reliability of the measurement model.

Table 6 Distinguishing validity tests.
Full size table

Model analysis

Model fit test

The goodness of fit is an important index used to test the degree of fit between the observed data and the model. A better fit indicates that the model explains the sample data to a higher degree, and the model is more consistent with the actual situation28. AMOS28.0 was used to test the fit of the model. The test results are shown in Table 7. The fit indicator X2/df value is 2.132, which is in the standard value between 1 and 3. With the ideal fit. According to the literature29, it is known that when the RMSEA value is less than 0.080, it usually means that the model has good fitness. In Table 7, RMSEA is 0.063, less than 0.080, indicating a good fit. GFI, CFI, IFI and TLI were 0.910, 0.933, 0.934 and 0.920, respectively, which were greater than 0.9, and the results fit well; AGFI and NFI were 0.879 and 0.882, respectively, which were greater than 0.8 and close to 0.9, all of them are within the acceptable range, indicating that the model fits well.

Table 7 Structural equation model fit test.
Full size table
Fig. 2
figure 2

Structural equation model fitting results.

Full size image

Model path coefficient test

As shown in Fig. 2, the model was tested for hypotheses using AMOS28.0 to obtain the simulation fit results. The results of the path coefficient test of each latent variable are shown in Table 8. As shown in Table 8; Fig. 2, the standardized path coefficients of spatial perception and group panic on non-adaptive evacuation behavior are 0.441 and 0.456, respectively. P-values are less than 0.001, reaching the significance level. It shows that they have a significant positive effect on non-adaptive evacuation behavior. Therefore, hypothesis H1 is not valid and H4 is valid. The standardized path coefficient of security atmosphere on non-adaptive evacuation behavior is -0.461. P-values are less than 0.001 and reach the level of significance. It shows that a safe atmosphere has a significant negative effect on non-adaptive evacuation behavior. Therefore, hypothesis H7 is not valid. The standardized path coefficients of spatial perception on group panic and security atmosphere are 0.335 and 0.641, respectively. P-values are less than 0.001 and reach the level of significance. It shows that spatial perception has a significant positive effect on group panic and a safe atmosphere. Therefore, hypotheses H2 and H5 are valid.

Table 8 Path coefficients between variables.
Full size table

Mediation effect test

According to the literature30, Mediation Effect refers to the existence of a mediating variable (M) between the independent variable (X) and the dependent variable (Y), through which the effect of the independent variable on the dependent variable is transmitted and realized. In short, the independent variable indirectly affects the dependent variable through the mediating variable. This paper takes group panic and security atmosphere as mediating variables. Test the mediating effect of group panic and security atmosphere between spatial perception and non-adaptive evacuation behavior. Currently, the commonly used methods include the stepwise test, multiplier product method, and Bootstrap method. Most scholars believe that it is more scientific to use Bootstrap method to test the significance of mediating effect. Therefore, this paper uses Bootstrap method with 5000 samples to test the significance of mediation effect. The results of the mediation effect test are shown in Table 9. The bias-corrected 95% confidence interval (BC) and percentile 95% confidence interval (PC) do not contain 0, indicating that the mediation effect is significant. According to the unstandardized estimates of the path coefficients in Table 8, the total effect of safety climate on emergency evacuation behavior is 0.474. Among them, the direct effect of spatial perception on non-adaptive evacuation behavior is 0.700. The mediating effect of group panic between spatial perception and non-adaptive evacuation behavior is 0.243. The p-value is less than 0.01. This indicates that group panic plays a positive partial mediating effect between spatial perception and non-adaptive evacuation behavior. Therefore, hypothesis H3 is valid. The mediating effect of a security atmosphere between spatial perception and non-adaptive evacuation behavior is -0.469. The p-value is less than 0.001, which indicates that the security atmosphere has a negative partial mediating effect between spatial perception and non-adaptive evacuation behavior, and therefore, hypothesis H6 is valid.

Table 9 Bootstrap-based significance tests for mediation effects.
Full size table

Discussion and implications

According to the model fitting results, there are direct and indirect effects between spatial perception and non-adaptive evacuation behavior. Group panic is a mediator variable in the path of spatial perception, and non-adaptive evacuation behavior has a significant “magnifying” effect. And the security atmosphere plays a significant “narrowing” effect. In this paper, three paths between spatial perception and non-adaptive evacuation behavior are constructed.

  1. (1)

    Direct path from Spatial perception to non-adaptive evacuation behavior

    The positive effect of spatial perception on non-adaptive emergency evacuation behavior is significant, which is contrary to the initial research hypothesis. This finding suggests that in a high-threat environment, heightened awareness does not linearly translate to more adaptive behavior. A more robust theoretical explanation can be found in the concepts of information overload and hypervigilance. In an emergency, a passenger with high spatial perception is processing a massive amount of data: the locations of multiple exits, the perceived time to reach them, the distribution of crowds, and the content of public announcements. This can lead to information overload, a state where an individual’s cognitive resources are insufficient to process the volume of incoming information. When overloaded, individuals are prevented from systematic processing and critical thinking, shifting instead to heuristic processing or intuitive thinking, which can lead to poor decisions. This cognitive impairment can result in well-documented paradoxical evacuation behaviors, such as the “faster is slower” effect, where individual attempts to move faster lead to collective gridlock and slower overall egress31. Simultaneously, in a threatening environment, heightened awareness can escalate into hypervigilance, a state of being constantly “on guard” and sensitive to potential threats. This perpetual fight-or-flight mode floods the brain’s frontal cortex—the center for decision-making—with stress hormones, producing emotions so intense that they can “knock out any logic or reason”. An individual becomes over-reactive to every potential threat, such as the chaotic movements of others or blocked exits, which can trigger panic and ultimately non-adaptive behaviors like pushing or freezing. This “dual-edged sword” effect of spatial perception provides a compelling explanation for why individuals who are more aware of their complex surroundings may, paradoxically, be more prone to panic and maladaptive actions during an evacuation.

  2. (2)

    Pathways to the construction of group panic

    The effect of spatial perception on group panic is positive and significant. It indicates that stronger spatial perception will lead to more panic. This is also contrary to the research hypothesis. Increased spatial perception does not decrease group panic, and instead increases it. Group panic has a significant positive effect on non-adaptive evacuation behavior. This is consistent with the findings of Wang et al.3. In emergencies, group panic lowers people’s psychological defenses. It is more likely to lead to non-adaptive evacuation behavior. Overall, the positive mediating effect of group panic in the mechanism of spatial perception on non-adaptive evacuation behavior is significant. Thereby, the logical paradigm of S-O-R is verified to exist in the influence mechanism of spatial perception on non-adaptive evacuation behavior. It indicates that the influence of spatial perception on non-adaptive evacuation behavior is based on the external environment stimulating the final behavioral pattern of human physiology or psychology. In terms of the research hypotheses of spatial perception on group panic as well as non-adaptive evacuation behavior are not valid, this path is the opposite of the hypotheses. The reason for this may be the greater the spatial perception of travelers, the more likely they are to experience group panic. Thus, non-adaptive evacuation behavior occurs.

  3. (3)

    The path of constructing safety climate

    Spatial perception has a positive and significant effect on safety climate. It demonstrates that improving travelers’ spatial perception increases travelers’ security atmosphere. This also validates the research hypothesis. The significant negative effect of security atmosphere on non-adaptive evacuation behavior reveals that a good security atmosphere can reduce the occurrence of non-adaptive evacuation behavior. Meanwhile, this paper verifies that there is a negative mediating effect between security atmosphere spatial perception and non-adaptive evacuation behavior. Both spatial perception and non-adaptive evacuation behavior are affected by the security atmosphere. A safe atmosphere can reduce the positive influence of spatial perception on non-adaptive evacuation behavior.

The findings of this study offer significant practical implications for the design and management of railway stations, focusing on two primary and interconnected intervention points: managing passengers’ cognitive load by “appropriately reducing spatial perception,” and leveraging the mediating role of a strong security atmosphere.

First, the model suggests that “appropriately reducing spatial perception” during a crisis is a key strategy. This does not mean making passengers unaware, but rather managing their cognitive load to prevent the information overload and hypervigilance that our results show can lead to non-adaptive behavior. In an emergency, passengers need the right information, not all the information32. This can be achieved through dynamic signage simplification. Stations should utilize intelligent digital signage that can adapt to a crisis. During an emergency, these systems can be programmed to deactivate non-essential displays and use clear, high-contrast graphics and animated arrows to show only the safest and most direct evacuation routes. This targeted, simplified information reduces the mental effort required for wayfinding under stress.

Second, and crucially, these interventions are most effective when they leverage the mediating role of the security atmosphere. Our model shows that a strong security atmosphere is a powerful buffer against panic and non-adaptive behavior. This is because a trusted environment reduces a passenger’s perceived need to independently scan and analyze every potential threat, thus mitigating the negative effects of high spatial perception and hypervigilance. Station managers can actively cultivate this atmosphere by ensuring visible staff presence and clear communication. During an emergency, the visibility of calm, authoritative staff is critical for instilling public confidence and preventing panic33. When passengers see trained personnel providing clear, consistent, and timely instructions, their trust in the official guidance increases. This encourages them to follow the simplified information provided by dynamic signage, rather than succumbing to hypervigilance. This direct human guidance reinforces a sense of order and control, directly strengthening the security atmosphere and ensuring that the “reduction” of complex spatial information is perceived as helpful guidance, not a lack of information.

Conclusion

This paper elucidates the complex mechanism through which spatial cognition influences maladaptive evacuation behaviors in railway station crowds. The study’s primary contribution is the mechanistic validation of the “dual-edged sword effect” of spatial perception in evacuation contexts. Paradoxically countering initial theoretical assumptions, heightened spatial awareness was found to positively correlate with both group panic and non-adaptive behavior. However, this detrimental relationship is substantially mitigated by the reinforcing presence of a strong security atmosphere. These findings establish novel empirical perspectives for public safety theory and offer tangible guidance for the design and management of complex transit hubs.

Limitations of the study

Despite its contributions, this study has several limitations that must be acknowledged. First, the sample has a significant gender imbalance, with female respondents comprising 65.85% of the participants. This demographic skew may impact the generalizability of the findings. Existing literature suggests that gender can influence key variables in our model. For instance, women are epidemiologically twice as likely as men to develop panic disorder and tend to report higher levels of fear and panic in response to threatening stimuli34,35. This suggests that the strong positive path that identified from spatial perception to group panic may be particularly pronounced in our female-skewed sample. Conversely, the research indicates that men are more prone to competitive behaviors during emergencies36, a key component of non-adaptive actions. Therefore, our model may underestimate the prevalence of competitive non-adaptive behaviors in a gender-balanced population. Future research should aim to recruit a more representative sample to validate these findings and explore potential gender-specific differences in the model’s pathways. Second, the demographic skew toward younger, highly-educated cohorts leaves insufficient representation of elderly passengers and those with lower educational attainment, potentially obscuring critical behavioral variances. Third, the reliance on self-reported measures introduces the risk of social desirability bias. Finally, the exploratory factor analysis yielded a cumulative variance explained of 54.562%. While this figure meets the acceptable threshold of > 50% for social science research, it also indicates that a substantial portion of the variance in non-adaptive evacuation behavior is explained by factors not included in our model. These could include specific environmental variables (e.g., lighting levels, signage clarity, dynamic crowd density) and other individual psychological traits (e.g., personality, general anxiety levels), which are known to influence behavior in emergencies.

Future research directions

The findings and limitations of this study open several avenues for future research. The paradoxical positive link between spatial perception and non-adaptive behavior warrants deeper investigation, potentially through experimental methods that can isolate the causal mechanisms of information overload and hypervigilance. To address the limitations identified, future studies should aim to recruit a more representative and gender-balanced sample to validate these findings and explore potential gender-specific differences in the model’s pathways. Furthermore, future research could integrate VR-enhanced experimental paradigms with multimodal psychophysiological monitoring (e.g., EEG spectral analysis, galvanic skin response) to enable dynamic mapping of the triadic interaction framework encompassing spatial cognition, affective states, and behavioral manifestations. This would enable a dynamic mapping of the triadic interaction framework encompassing spatial cognition, affective states, and behavioral manifestations, providing a more granular understanding of passenger behavior under stress.