Research on the Characteristics and Influencing Factors of Community Residents’ Night Evacuation Behavior Based on Structural Equation Model

Abstract

Nighttime natural disasters are common, and earthquakes are the most common of these disasters. This study explores the behavior of residents during night evacuations after an earthquake and the factors that influence such behavior. The aim of this study is to improve nighttime disaster relief in residential areas and provide new ideas for renovating and upgrading existing communities. Shanghai is one of the most population-dense cities in China, and it has a fragile built environment. As part of this study, questionnaires were randomly distributed to residents living in Shanghai, and SPSS and AMOS were used to establish a structural equation model to uncover the relationship between factors and the residents’ nighttime evacuation behavior. Some of the results and conclusions were the following: (a) residents had the highest tendency to choose autonomous evacuation and pro-social behavior during night evacuation than at any other time; (b) spatial perception was significantly negatively correlated with residents’ exclusive behavior; (c) herd behavior, autonomous evacuation, and prosocial behavior were significantly positively correlated with social networks. Finally, this study proposes policy and spatial coping strategies for different behaviors during evacuation to enhance the community’s natural disaster prevention capability.

1. Introduction

Earthquakes are among the deadliest natural disasters, often causing devastating damage and loss of life. According to EM-DAT statistics, earthquake disasters killed 752,498 people and injured around 1,574,000 between 1998 and 2018 [1]. In China, the 2008 Wenchuan Earthquake triggered more than 200,000 landslides, resulting in more than 20,000 deaths [2]. Nighttime earthquakes represent a long-term problem around the world. According to statistics, most earthquakes occur between 19:00 p.m. and 6:00 a.m. [3]. The sudden and imperceptible characteristics of nighttime disasters pose great challenges for disaster relief. At nighttime, evacuations may be more inconvenient and perhaps more dangerous [4,5], often leading to more serious consequences. However, the majority of studies that investigate the factors that determine evacuation decisions and behaviors have been limited to daytime evacuations [6,7]. Therefore, the further investigation of nighttime evacuations is of vital importance.

Studies on people’s evacuation behaviors and relevant influencing factors during nighttime earthquakes are quite rare. The few studies on this issue have mainly focused on the built environment [8,9,10,11,12], mental factors [13], and social networks [14]. Yu conducted an exploratory investigation into the mental and behavioral features of nighttime resident evacuations in Tianjin and found that the emergency lighting system was the most important environmental factor [11]. Sun proved that the functions of social networks (i.e., neighborhood assistance, ridesharing, etc.) can become problematic during nighttime evacuations [14]. However, comparably, there are many studies focused on evacuation behavior and its influencing factors. Lyu evaluated the risk of geohazards around Lanzhou in terms of both hazards and exposure [14,15]. Song studied the evacuation behavior of subway station passengers and their influencing factors and categorized this behavior into four types. He also pointed out personal factors, the built environment, and emergency regulations as the main influencing factors [16]. Previous studies have also suggested that individual factors [17,18,19], risk awareness and preparedness [20,21,22], evacuation warnings and evacuation orders [23], route choice [24,25], and social networks [26,27] all affect evacuation behavior. Despite the fact that abundant factors have been considered in studies focusing on evacuation behavior, few have assessed evacuation behavior at nighttime against the backdrop of residential areas. Among the studies that considered nighttime factors, they are limited to investigating the dynamics and mechanisms of evacuations in networks, and the connection between social networks and nighttime evacuation behavior has not been quantified. As such, this study attempts to explore this connection.

The objective of this study is to assess the characteristics of residents’ nighttime evacuation behaviors and to construct a theoretical hypothesis model of the influencing factors of residents’ nighttime emergency evacuation behavior. The study also aims to test how the influencing factors affect one another. The results of this study will help us to better understand residents’ perceptions and how they affect their behavioral intentions. The results also provide useful suggestions for guiding risk communication activities and disaster preparation.

This study was approached as follows: First, we conducted a literature review on built environment perception at night, night evacuation risk perception, social networks, and evacuation behavior at night. Then, we proposed our theoretical hypotheses and constructed a conceptual model. Second, the data sources, sample collection, and research instruments for this study were collected. A questionnaire on residents’ emergency evacuation behaviors and their influencing factors was devised and distributed to 285 residents of Shanghai. We validated the model using SPSS and Amos 24.0 software from IBM (Armonk, NY, USA), assessed the results, and considered the theoretical and practical contributions of this study. Finally, we examined the shortcomings and contemplated the directions for future in-depth research.

2. Literature Review and Hypothesis Development

The research on nighttime earthquakes can be traced back to the 1990s. With the frequent occurrence of nighttime disasters, researchers have gradually realized that the damage caused by nighttime earthquakes is greater than that in the daytime. The influencing factors of evacuation behavior considered in the existing research mainly include three aspects: the perception of the built environment at night, the perception of risk of evacuation at night, and social networks.

2.1. Built Environment Perception at Night

The influence of the built environment at night on people’s evacuation behavior is quite different from that in the daytime. Under the influence of nighttime factors, people put forward different requirements for the area of refuge space [28], emergency power supply, lighting facilities [13,29], and guidance signs of refuges [30]. Nighttime factors can influence urban disaster prevention from the characteristics of psychological behavior changes, vision, and perception at night during disasters [10]. In residential areas, the difference between the spatial imagery at night and in the daytime is mainly reflected in the light received, environmental construction, and the visual perception of the space [11].

Existing studies have revealed the impact of the perception of the built environment on evacuation behavior. In rural areas, when residents have access to reasonable evacuation routes and good, high-quality roads, they will be more willing to evacuate [9].

Combined with existing research, we studied residents’ perceptions of evacuation spaces from four aspects: the distribution of hidden dangers of secondary disasters at night, emergency shelter space at night, evacuation passages at night, and disaster prevention facilities at night. Considering the aging communities and the weak configuration of public facilities, two indicators were also considered: barrier-free design at night and emergency rescue facilities at night. The indicator system for spatial perception is shown in

Table 1. The indicator system for spatial perception.

Criteria	Sub-Criteria	Space Index	Perception Index
Built environment perception at night	Secondary disasters at night	Accessible design	No tripping on steps (where there is a significant height difference) during night evacuation.
		Building facade control	Things outside the house will not be damaged after an earthquake.
	Emergency safe space at night	Spatial scale	There are reasonable shelters in/next to the community where I live.
		Supporting facilities	The house I live in now has strong earthquake resistance.
	Evacuation routes at night	Residential road structure	The neighborhood is well connected to the road outside.
		Effective width of road	Roads in the community are less likely to be congested during evacuation.
		Height–width ratio	Reasonable spacing between houses facilitates evacuation in the event of an earthquake.
	Night disaster prevention facilities	Emergency lighting system	The neighborhood where I live now has plenty of lights at night, so I can see the road clearly.
		Signage facility	The residential area where I live now is fully signposted at night.
		Emergency rescue facilities	The neighborhood where I live now has complete rescue facilities at night.

.

Hypothesis 1 (H1).

Built environment perception at night has a significant effect on evacuation behavior at night.

2.2. Night Evacuation Risk Perception

Risk perception is a psychological concept that focuses on the influence of the experience gained by an individual through intuitive judgment and subjective feelings on the individual’s cognition, and is the individual’s perception and cognition of various risks in the objective world [31]. In existing disaster risk perception studies, respondents were asked to assess the threats, fears, hazards, and concerns posed by earthquakes [32,33]. Slovic used personality theory to endow risk events with “personality characteristics”, such as voluntary characteristics, potential catastrophic characteristics, and controllability characteristics, and designed a bipolar scale [34]. Li Huaqiang proposed the use of familiarity, control, and fear to measure people’s characteristics of risk perception regarding earthquake disasters [35]. Huang Kun introduced three indicators to evaluate the risk perception of rural residents in earthquake evacuation: the probability of disaster occurrence, the severity and urgency of the disaster, and the degree of impact on oneself [36].

Weinstein proposed a reciprocal relationship between evacuation behavior and risk perception, arguing that high levels of risk perception cause people to change their behavior, which in turn reduces their risk perception [13]. Yu Weiwei summarized the psychological and behavioral mechanisms of residents when disasters occurred at night, proposed the action path of “psychological_perception_decision_behavior”, and the influence path of the environment on psychological perception [11].

Based on the earthquake-related risk perception measurement system, this study focuses on the impact of disasters on residents’ perceived influence and fear of an earthquake, modifying the indicators in combination with the specific environment and nighttime factors of the residential area and constructing the following risk perception indicator system (

Table 2. The indicator system for risk perception.

Criteria	Sub-Criteria	Index
Risk perception	Level of perceived influence	Earthquakes will directly affect me.
		An extreme earthquake will have long-term negative effects on me.
		Earthquakes will cause more damage to my home when shaken.
	Level of fear	Darkness makes me panic.
		Fear of loud noises during evacuation.
		Fear of violent shaking.
		Fear of pungent odors.

).

Hypothesis 2a (H2a).

Night evacuation risk perception has a significant effect on evacuation behavior at night.

Hypothesis 2b (H2b).

Evacuation behavior at night has a significant effect on night evacuation risk perception.

Hypothesis 3a (H3a).

Built environment perception at night has a significant effect on night evacuation risk perception.

Hypothesis 3b (H3b).

Night evacuation risk perception has a significant effect on built environment perception at night.

2.3. Social Network

Social networks are divided into two types: individual-centered and social-organization-centered. This research mainly focuses on personal relationship networks (family networks and neighborhood networks) and organizational networks within the community. In the measurement of personal relationship networks, Lin Nan proposed network composition, network heterogeneity, network height, and network difference to measure the resources in the network [37] (pp. 78–98). Bian proposed individual central network measurement indicators: network size, network top, network differences, and network composition [38]. At the community level, social networks include subjects, policies, and other issues of emergency management [39]. The connection between emergency organizations at the community level and the network construction of community emergency coordination will also affect evacuation behavior [40].

In China, most studies on the influencing factors of evacuation behavior only focus on the built environment, while foreign studies further analyze the impact of social networks and social capital on evacuation behavior [41,42]. By quantifying people’s group evacuation behaviors, the fields of physics and mathematics have found that there is a positive effect on evacuation when there is a cooperative relationship between small groups [43]. In addition, in low-visibility environments at night, small groups can alleviate people’s fear of disorientation in groups and improve message exchange rates [44].

The focus of the evacuation simulation has changed from a single individual to a small group, which reflects that social networks affect the behavior of residents. Researchers have confirmed the influence of built-environment factors on social network size [45] and personal network structure [46]. In recent years, some studies have used numerical evacuation modelling and simulation to model the effects of community or communication networks in the context of evacuation decision making [47,48,49], validating the influence of social networks on evacuation behaviors.

Based on existing research, this study proposes three network metrics (

Table 3. The indicator system for social networks.

Criteria	Sub-Criteria	Index
Social network	Kinship network	Family structure
	Neighborhood network	The number that I associate with the neighborhood.
		The frequency that I associate with the neighborhood.
		The purpose that I associate with the neighborhood.
	Organization network within community	The frequency that I pay attention to the popularization of earthquake disaster prevention knowledge in the community.
		The frequency that I participate in community evacuation drills.

).

Hypothesis 4a (H4a).

Built environment perception at night has a significant effect on social network.

Hypothesis 4b (H4b).

Social network has a significant effect on built environment perception at night.

Hypothesis 5 (H5).

Social network has a significant effect on evacuation behavior at night.

2.4. Evacuation Behavior at Night

The existing research is mainly based on the internal evacuation of public buildings. Song Chao divides the evacuation behavior of subway passengers into four types: herd, mutual assistance, exclusiveness, and autonomous evacuation [17]. Prosocial behavior is any action taken to benefit others [50,51], and studies have found prosocial behaviors in residential evacuations [52]. For example, returning home to rescue relatives is a special prosocial behavior [53]. On the basis of the four existing types of behavior, this study regards mutual aid behavior and returning home to find relatives as prosocial behaviors and forms four behavioral types (

Table 4. The indicator system for evacuation behavior.

Criteria	Sub-Criteria		Index
Evacuation behavior at night	Herd behavior		After the earthquake at night, I will listen to the guidance of the community staff.
			After an earthquake at night, I will give up what I think is the correct evacuation route and follow the crowd.
			When I see other people gathering in crowds during the evacuation process, I will do the same.
	Prosocial behavior	Mutual aid	During the evacuation, I will help to remove obstacles to facilitate the evacuation of others.
			If someone asks me for directions, I will guide others to evacuate.
			I will help them up when someone in the crowd falls.
		Return home to find relatives	I will go back home to find relatives.
	Exclusive behavior		When someone in the crowd falls, I will run away from the crowd as quickly as possible.
			I will push with others during the evacuation.
			During the evacuation, I will not obey orders and will try to move myself away from the crowd.
	Autonomous evacuation		After an earthquake occurs at night, I will choose a path that is familiar to me and has emergency lighting to escape.
			During the evacuation process, I will evacuate with my own intuition instead of following emergency guidance.

).

2.5. Theoretical Framework

This study constructs a model of the influence mechanism of social networks, built environment perception, and risk perception towards evacuation behavior choices (Figure 1).

3. Methodology

3.1. Research Instrument and Data Collection

The population of this study was composed of adults aged 18 years and above, mainly because this group has a better ability to assess their behavioral intentions and may have evacuation experience. The data were obtained from field research conducted in Shanghai from March 2022. The main reasons for selecting Shanghai were, first, the fact that it has a high population density with more than 24 million people, which makes residential areas sensitive to earthquakes. Second, Shanghai has widespread and slowly accumulated soft clay layers, making it a typical soft-soil area with a fragile geographic environment [54]. Third, Shanghai is among the cities with the best earthquake shelter planning and resident awareness of personal protection in the country, making it easier to better assess residents’ evacuation behaviors.

The survey used random sampling and convenience sampling methods. We chose three districts with the highest residential population density in Shanghai and randomly distributed questionnaires in each district. The sample was determined based on the number of items in the survey. It is desirable to have a minimum of three variables for each component, and the sample size should be large enough to obtain reliable results [55]. The number of responses should be five times the number of items in the questionnaire; for example, a 50-item questionnaire needs at least 250 responses [56] (p. 28). As the study had 38 items, 190 respondents (38 × 5) were needed for an adequate sample size. In addition, according to Ali et al. [57], the typical sample size for behavioral studies should range between 200 and 500. A total of 285 questionnaires were randomly distributed via social media, and 264 valid questionnaires were recovered.

The design of the scale drew on the findings of Ao, Y. [9] and Song, C. [17]. During the scale design process, pre-research questionnaires were distributed in December 2021 to ensure quality, accuracy, and applicability. Three revisions were conducted to correct the ambiguous and not clearly expressed questions. The official questionnaire was designed using a 5-point Likert scale, with a scale of 1–5 representing responses ranging from “strongly disagree” to “strongly agree”. The questionnaire consists of five parts: basic information of disaster prevention respondents, social network, built environment perception, risk perception, and evacuation behavior choices. The questionnaire is presented in Appendix A.

3.2. Data Analysis (CB-SEM)

Structural equation modelling is widely used in a variety of disciplines, particularly in the social sciences [58]. Structural equation modelling allows the researcher to statistically examine “whether a hypothesized model is consistent with the data collected to reflect [the] theory” [59] (p. 34). In this study, we used covariance-based structural equation modelling (CB-SEM) to test the effect of risk perception, built environment perception, and social networks on intention for nighttime evacuation.

We used IBM SPSS 23.0 and Amos 24.0 software to analyze the data. To examine the quality and structure of the variables, exploratory factor analysis, confirmatory factor analysis (CFA), and structural equation modelling (SEM) were implemented to examine the causal relationships between the constructs.

4. Results

4.1. Reliability and Validity Test

4.1.1. Reliability Analysis

In this study, a total scale containing 30 items was established. In order to ensure the internal consistency of the scale, before conducting an exploratory factor analysis, this study first used the data of 264 samples to calculate the Cronbach’s α value of the internal consistency reliability coefficient of the scale and perform internal consistency checks. The results are shown in

Table 5. Internal consistency coefficient of evacuation behavior and influencing factors.

Index	Item	Cronbach’s α
Distribution of hidden dangers of secondary disasters at night	Q14A1, Q14A2	0.821
Emergency shelter at night	Q14A3, Q14A4, Q14A5	0.831
Night evacuation	Q14A6, Q14A7, Q14A8	0.848
Night disaster prevention facilities	Q14A9, Q14A10, Q14A11	0.869
Influence level	Q6A1, Q6A2, Q6A3	0.843
Level of fear	Q6A4, Q6A5, Q6A6, Q6A7	0.874
Herd behavior	Q15A2, Q15A5, Q15A6	0.830
Prosocial behavior	Q15A7, Q15A8, Q15A9, Q15A12	0.884
Exclusive behavior	Q15A1, Q15A10, Q15A11	0.847
Voluntary evacuation	Q15A3, Q15A4	0.755

.

4.1.2. Validity Analysis

This study uses SPSS to conduct an exploratory factor analysis on the questionnaire to test whether it has good construct validity. Firstly, a factor model adaptability analysis was carried out based on the questionnaire results (

Table 6. KMO and Bartlett test of evacuation behavior and its influencing factors.

KMO and Bartlett Test		Disaster Risk Perception	Built Environment Perception	Social Network	Evacuation Behavior
KMO Sampling suitability quantity		0.939	0.954	0.911	0.834
Bartlett’s sphericity test	Approximate chi-square	1206.643	2130.489	829.730	1234.114
	Degrees of freedom	21	55	21	66
	Salience	0.000	0.000	0.000	0.000

). The test results showed that the KMOs of disaster risk perception, built environment perception, and social network were 0.939, 0.954, and 0.911, and the p value was 0.000, the KMO of evacuation behavior was 0.834, and the p value was 0.000. Both are suitable for exploratory factor analysis.

A principal component analysis method is used, and the factor analysis results for evacuation behavior are shown in

Table 7. Principal component analysis of evacuation behavior.

Composition	Initial Eigenvalue		Extract the Load Sum of Squares				Rotational Load Sum of Squares
	Summary	Percent Variance	Cumulative %	Summary	Percent Variance	Cumulative %	Summary	Percent Variance	Cumulative %
1	11.403	45.611	45.611	11.403	45.611	45.611	7.532	30.130	30.130
2	2.947	11.786	57.397	2.947	11.786	57.397	4.561	18.242	48.372
3	2.082	8.327	65.724	2.082	8.327	65.724	4.338	17.352	65.724
4	0.764	3.058	68.782
5	0.675	2.700	71.482
6	0.650	2.598	74.080
7	0.560	2.241	76.321
8	0.553	2.213	78.534
9	0.482	1.930	80.464
10	0.466	1.864	82.328
11	0.440	1.761	84.090
12	0.425	1.702	85.791
13	0.399	1.598	87.389
14	0.383	1.532	88.922
15	0.361	1.443	90.364
16	0.337	1.350	91.714
17	0.310	1.241	92.955
18	0.288	1.153	94.108
19	0.281	1.123	95.231
20	0.275	1.101	96.333
21	0.225	0.899	97.232
22	0.207	0.827	98.059
23	0.195	0.779	98.839
24	0.180	0.719	99.558
25	0.110	0.442	100.000

.

According to the principal component analysis in the table, three factors were extracted, namely, the disaster risk perception in the hypothesis, the built environment perception, and the social network. The total explained variance reached 64.724%, indicating that the evacuation behavior influencing factors have a good structural validity.

4.2. Descriptive Statistics

4.2.1. Descriptive Analysis of Basic Features

First, descriptive statistical analysis was performed on the basic characteristics of the collected samples, as shown in

Table 8. Basic characteristics of samples.

Basic Features	Classification	Number	%
Gender	Male	133	50.6
	Female	130	49.4
Age	18–30 years old	105	39.9
	30–50 years old	131	49.8
	Over 50 years old	27	10.3
Time at home during on weekdays	All day	22	8.4
	Daytime	9	3.4
	Nighttime	139	52.9
	Sometimes	93	35.4
Approximate time taken to fall asleep each night	20:00–22:00 22:00–0:00 After 0:00	35 161 53 14	13.3 61.2 20.2 5.3
	Other time
Experienced a nighttime disaster or not	Yes No	187	71.1
		76	28.9

.

From Table 8, it can be seen that the proportion of males and females in the survey results is relatively balanced, and that age is mainly concentrated in the age range of 18–50 years old. Regarding the time spent at home on weekdays, most people are at home at night, accounting for 52.9%, and 8.4% are at home during the daytime. When it comes to falling asleep, most people fall asleep before midnight. In this survey population, 71.1% of them have experienced nighttime residential disasters, which provides reliability for the results of this study.

4.2.2. Descriptive Statistical Analysis of Evacuation Behavior

A descriptive statistical analysis of the various dimensions of emergency evacuation behavior in this survey (

Table 9. Values in different evacuation behavior.

Index	Minimum	Maximum	Mean	Variance
Herd behavior	1	5	3.79	1.211
Prosocial behavior	1	5	3.88	1.186
Exclusive behavior	1	5	3.69	1.299
Autonomous evacuation	1	5	3.92	1.184

).

Table 9 shows that the respondents in this study have a relatively high behavioral preference in the four dimensions of the emergency evacuation process, with all items scoring above 3.5. Among them, the behavioral tendency of voluntary evacuation has the highest score of 3.92 points, while the average score of herd behavior is only 3.79 points, indicating that residents are more willing to conduct a voluntary evacuation based on experience in a familiar environment. The score of prosocial behavior is also high, at 3.88 points, among which the average score of mutual aid behavior is 3.91 points. The proportion of residents who help each other after the earthquakes is very high, and the special behavior of returning to their homes to find relatives has an average score of 3.79 points. The score of exclusive behavior is the lowest, at only 3.69 points, which is the behavior with the lowest choice tendency among the four behaviors.

4.2.3. Descriptive Statistical Analysis of Each Dimension of Influencing Factors

In terms of evacuation space perception (

Table 10. Values in spatial perception factors.

Sub-Criteria	Index	Mean	Variance
Distribution of hidden dangers of secondary disasters at night	Accessible design	3.87	1.197
	Building facade control	3.86	1.227
Emergency shelter at night	Spatial scale	3.96	1.123
	Supporting facilities	3.98	1.056
Night evacuation	Residential road structure	3.84	1.078
	Effective width of road	3.98	1.149
	Aspect ratio	3.95	1.156
Night disaster prevention facilities	Emergency lighting system	3.92	1.118
	Signage facility	3.94	1.132
	Emergency rescue facilities	3.87	1.133

), most Shanghai residents have a good perception of nighttime evacuation space efficiency in residential quarters, with an average score of more than 3.8. Among the residents, the perceived score of residential road structures is relatively low, and the variance is small, reflecting the residents’ dissatisfaction with the evacuation efficiency of residential roads; the supporting facilities and emergency lighting systems have the highest scores, indicating that most residents believe that their community is equipped with a good surrounding evacuation site and an emergency lighting system has been established.

For the analysis of the influencing factors of residents’ evacuation risk perception (

Table 11. Values in risk perception factors.

Sub-Criteria	Index	Mean	Variance
Level of perceived influence	Earthquakes will directly affect me.	4.01	1.127
	An extreme earthquake will have long-term negative effects on me.	4.05	1.095
	My home will become damaged when shaken.	4.16	1.039
Level of fear	Darkness makes me panic.	4.15	1.043
	Fear of loud noises during an evacuation.	4.05	1.083
	Fear of violent shaking.	4.13	0.970
	Fear of pungent odors.	4.10	1.128

), in terms of the impact of nighttime earthquakes, the average value of the three indicators exceeds 4 points, indicating that most residents believe that the earthquake will have an impact on themselves and their families, especially on their own houses. Regarding fear factors, darkness, loud noises, shaking, and pungent odors all make residents fearful, and the dark factor has the highest score and greatest impact on people.

In terms of social networks, personal social networks show great differences (

Table 12. Values in social network factors.

Index	Item	Number	Percent
Family structure	Core family	117	41.05%
	Stem family	139	48.77%
	Living alone	22	7.72%
	Other	7	2.46%
Number of neighbors	0	18	6.32%
	1–2	68	23.86%
	3–5	122	42.81%
	6–10	61	21.4%
	11 or more	16	5.61%
Neighborhood frequency	Once a day or more	24	8.42%
	Every two or three days	69	24.21%
	Once a week	118	41.4%
	Once a month	46	16.14%
	Rarely interact	28	9.82%
Purpose of Neighborhood	Gossip	111	38.95%
	Community activity	103	36.14%
	Private matter	39	13.68%
	Other	32	11.23%
Pay attention to the popularization of earthquake disaster prevention knowledge in the community	Never seen	66	23.16%
	Rarely seen	196	68.77%
	Often	23	8.07%
Frequency of participation in community evacuation drills	Participate often	55	19.3%
	Participate occasionally	168	58.95%
	Never participate	62	21.75%

). In terms of kinship networks, most of the respondents are from nuclear families and stem families, accounting for 41.05% and 48.77%, respectively, and a small number of people living alone participated in the study. The second survey accounted for 7.72%. In the neighborhood relationship network, most people communicate with 3–5 people, and the frequency is once a week, mostly for the purpose of nostalgia and community activities. In the survey of social networks in the community, most people have seldom, or perhaps never, participated in evacuation drills for earthquakes; therefore, there is a lack of earthquake disaster prevention knowledge in the Shanghai community.

4.3. Path Analysis

4.3.1. Fitting Test of the Model

In this study, structural equation modeling is used to verify the proposed model. In the structural equation model of the influencing factors of residents’ evacuation behavior during nighttime earthquakes in residential areas, there are four dimensions of evacuation behavior and three dimensions of behavioral influencing factors. In the model, disaster risk perception, built environment perception, and social networks are potential exogenous variables, whereas herd behavior, prosocial behavior, exclusive behavior, and autonomous evacuation behavior are dependent variables. Through an analysis of the confirmatory factors of the structural equation theory hypothesis model constructed above, it is concluded that the model is well-identified. After deleting the paths with insignificant coefficients from the model, we can obtain the final model. The significance of risk perception on the four types of behavior paths is greater than 0.05, which does not directly affect people’s behavior judgment, but the path between spatial perception and social network is significant, showing a certain interaction between them.

Finally, a standardized route map of passenger evacuation routes is obtained (Figure 2).

As shown in

Table 13. Test results for structural equation model fit.

Index	Value	Threshold
χ2	1052.589	The smaller the better
GFI	0.829	>0.9
AGFI	0.807	>0.9
RMSEA	0.051	<0.08 (<0.05 excellent; <0.08 good)

, the factors of each item are in the range of 0.5–0.95, and they passed the significance level of 0.05, so it can be considered that the model is well-established. The fitting degree test results of the adjusted structural equation model of nighttime earthquake evacuation behavior influencing factors are shown in the table below. Among them, χ² and RMSEA achieve acceptable values. Although GFI and AGFI are not greater than 0.9, they also essentially meet acceptable values. Therefore, the model accurately fits these results.

4.3.2. Analysis of the Influence Coefficient

The coefficient calculated by the structural model is the direct influence coefficient between the variables. In order to further analyze different influencing factors on evacuation behavior, it is necessary to multiply the latent variable to a certain evacuation behavior dimension by the regression coefficient of the influence factor to the latent variable to obtain the indirect influence coefficients.

Through analysis and calculation, the influence coefficients of each influencing factor on each dimension of evacuation behavior are summarized in

Table 14. Standardized total effect value of each factor on each dimension of evacuation behavior.

	Exclusive Behavior	Herd Behavior	Prosocial Behavior	Autonomous Evacuation
Spatial perception factor	−0.588
Space 5	−0.703392
Space 9	−0.692966
Space 6	−0.684428
Space 3	−0.683733
Space 10	−0.6834
Space 11	−0.642446
Space 4	−0.641457
Space 8	−0.638376
Space 7	−0.602007
Space 2	−0.494508
Space 1	−0.479808
Social network factor		0.54	0.584	0.587
Society 3		0.4212	0.45552	0.45786
Society 2		0.39906	0.431576	0.433793
Society 6		0.39528	0.427488	0.429684
Society 4		0.3861	0.41756	0.419705
Society 5		0.38016	0.411136	0.413248
Society 7		0.38016	0.411136	0.413248
Society 1		0.36018	0.389528	0.391529

.

It can be seen from Table 14 that the influence coefficient of the spatial perception factor on exclusive behavior is 0.588. In the analysis of each factor, in terms of secondary disasters, factor 1 barrier-free design and factor 2 building facade control have little influence on exclusive behavior, which is no more than 0.5; in terms of emergency shelter space at night, factor 5 supporting facilities have the greatest negative impact on exclusive behavior, with an influence coefficient of 0.70, and a reasonable shelter in the factor 3 community will also effectively avoid the occurrence of exclusive behavior. Among the disaster relief facilities at night, the negative influence coefficient of emergency lighting facilities is the highest, followed by signage facilities and emergency rescue facilities.

The influence coefficients of social network factors on herd behavior, prosocial behavior, and autonomous evacuation behavior were all higher, which were 0.54, 0.584, and 0.587, respectively. Specific to the analysis of each factor, the neighborhood communication network has the greatest impact on the three behaviors. Among them, factor 3 (the frequency of neighborhood contacts) and factor 2 (the number of neighborhood contacts) are the more important influencing factors; in the social organization network, the frequency of participating in evacuation drills is more important. As for the influence factors of the three types of behaviors, the personal kinship network has less influence on the three types of behaviors.

5. Discussion

On the basis of theoretical research on emergency evacuation behavior and its influencing factors and empirical research through questionnaire surveys, this study constructs a theoretical hypothesis model of residents’ nighttime emergency evacuation behavior.

First, four dimensions of evacuation behavior were identified: herd behavior, pro-social behavior, exclusive behavior, and autonomous evacuation behavior. Moreover, three dimensions of influencing factors were identified: spatial perception factors, risk perception factors, and social network factors. Through the reliability and validity analysis and confirmatory factor analysis of the data, it is proven that the four-factor model of evacuation behavior and the three-factor model of behavioral influencing factors can be identified. The results of the four behaviors are partly consistent with Helbing, D. [60] and Lu, W.L. [61], who found that the evacuation behavior of subway station passengers includes self-help behavior, herd behavior, exclusion behavior, and self-evacuation behavior, among others.

Second, the surveyed residents had the highest preference for voluntary evacuation behavior, a higher preference for pro-social behavior, and a lower preference for herd and exclusion behavior. These attitudes appear to conflict with the results of Song, C. [17], who found that during the evacuation of a subway station, passengers will be most likely to exhibit self-help behavior but least likely to exhibit self-evacuation behavior. The reasons for this inconsistency may be that residential areas are where people are more familiar with and where they have solid social networks.

Third, the correlation between the latent variables was discussed, and the spatial perception factor was significantly positively correlated with exclusive behavior. Disaster prevention facilities at night and shelters in the community were two significant spatial influencing factors. The importance of disaster prevention facilities at night and shelters in the community has been proven in previous studies [29,30]. The former study also put forward more important factors such as the lighting facilities and quality of roads [13,29,31,32], etc. However, our study found the most crucial factors that lead to exclusive behaviors, which may help to guide community construction more directly.

Fourth, the results indicate that social network factors and herd behavior, as well as pro-social behavior and voluntary evacuation behavior, were significantly positively correlated, and the frequency of neighbor communication and number of neighbors were two significant social network influencing factors. In the social organization network, the frequency of participating in evacuation drills is a more important influencing factor; in the personal kinship network, it has less influence on the three types of behaviors. This is consistent with former studies, which found that risk-mitigation behavior may be more relevant to shared social norms and rules than individual perceptions in a culture in which one is considered interlinked with others rather than a distinct, independent, and distinguishable being [62,63]. Furthermore, our findings indicate that providing education programs to households may increase pro-social and preparedness behaviors.

6. Conclusions

Our study of the relationship between built environment perception at night, night evacuation risk perception, social networks, and evacuation behavior at night is of theoretical and practical importance. It may help us to understand whether and how these spatial, mental, and social factors influence different types of evacuation behaviors.

6.1. Theoretical Contributions

Our research makes two main theoretical contributions to the literature on evacuation behavior. First, the study extends the traditional Behavior-Based Safety (BBS) theory from industrial applications to disaster relief scenarios. Based on the BBS theory, a series of scales are used to measure residents’ behavior and a conceptual model of the impact of risk perception, built environment perception, and social networks on evacuation behaviors is constructed. This will help us to understand whether and how these three factors influence each type of resident evacuation behavior.

Second, previous studies on evacuation behavior have focused on the general situation, such as daytime shelters, daytime evacuation paths, and daytime risk perception, with less attention being paid to nighttime elements [64,65]. In this study, nighttime factors such as rescue facilities at night and darkness were added to the model, and such factors were empirically verified to assess their significant positive effect on behaviors, which may further affect the efficiency of earthquake evacuations. Our study also suggests that networks within communities, especially during evacuation drills, may have an important influence on resident behavior.

Third, the characteristics of the resident evacuation behaviors summarized in this study provide a theoretical reference for the formulation of community emergency plans. The behavioral influencing factors found in this study can also guide the preparation of comprehensive disaster prevention planning and departmental special planning.

6.2. Practical Contributions

Based on the analysis of the existing evacuation behavior characteristics and influencing factors, researchers can intervene with behaviors by changing the influencing factors. The favorable behaviors of residents during the night evacuation process are strengthened, and the unfavorable behaviors during the evacuation process are weakened so as to achieve a safe evacuation.

6.2.1. Measures for Exclusive Action

Exclusive behavior is an unfavorable behavior for night evacuations, interfering with normal evacuations, so it is necessary to weaken this behavior. The results of this study show that environmental perception factors have a certain influence on the generation of exclusive behavior. If there are good medical and other supporting facilities around the residential area, or if there are reasonable shelters in and around the community, the occurrence of exclusionary behaviors will be effectively alleviated.

In addition, in the nighttime environment, a good emergency lighting and indicator system can reduce the panic of residents and reduce the occurrence of secondary disasters caused by exclusive behavior. The night lighting in the residential area should follow the layout principle of “highlighting the key points and orderly linking” to form a “point-line-surface” emergency lighting evacuation system [18]. In some existing residential areas in Shanghai, the lighting facilities at the entrance to each housing unit are outdated, emergency lighting does not cover every road intersection, and some important intersections and emergency shelter spaces lack emergency lighting reserves, which creates a significant safety hazard. The nighttime emergency indication system includes light signs, sound signs, and tactile signs. Most communities lack systematic, well-lit signs, and signs with sound and sensors are used less frequently, which fails to fully consider the needs of vulnerable groups.

6.2.2. Measures for Prosocial Behavior

The results show that social networks have a strong influence on herd behavior, prosocial behavior, and autonomous evacuation behavior, and a good social network is conducive to the occurrence of prosocial behavior. In a more mature social network, residents often help each other when disaster strikes because of mutual familiarity and trust. The analysis results show that a neighborhood communication network is the main influencing factor. This community can promote neighborhood communication by strengthening the construction of a humanistic environment. It can also promote the formation of the community and help carry out orderly disaster relief and recovery activities.

6.2.3. Measures for Autonomous Evacuation Behavior

A good social network will deepen residents’ familiarity with their surrounding environment, and people are more likely to engage in self-evacuation behaviors in a more familiar living environment. There are various forms of autonomous evacuation behavior. This study focuses on two behaviors: the tendency for familiarization and simplicity, and light tendency and risk avoidance. Autonomous evacuation is not necessarily detrimental to evacuation, and the community needs to guide autonomous evacuation through benign measures. On the one hand, it is necessary to strengthen residents’ knowledge of nighttime emergency evacuation. The community should regularly organize emergency evacuation drills, popularize prevention instructions for community disasters, and guide residents to participate in community disaster prevention practices. Disaster prevention knowledge is popularized in the form of lectures, training courses, broadcasts, etc. In the guidance for disaster prevention, the daily evacuation habits of residents should also be followed. For example, if residents have a tendency to move toward light at night, good emergency lighting should be set up at important intersections and safe entrances and exits. Additionally, the community can disperse the evacuation crowd through lighting to prevent the gathering of crowds.

6.2.4. Measures for Herd Behavior

Social networks play a positive role in promoting herd behavior. Just as in autonomous evacuation behavior, herd behavior may not only lead to congestion at evacuation exits and interfere with evacuation, but may also help guide crowds and speed up evacuation. On the one hand, the effect of herd behavior depends on the relationship between neighbors. A good neighbor relationship makes the small evacuation groups cooperate rather than compete, and the mutual following between acquaintances is conducive to the transmission of refuge information and the care of special groups. On the other hand, it depends on whether the community can carry out timely and effective guidance. For the community, it is necessary to improve the community disaster early warning system and safety management team system, and make the system connected with the city’s early warning platform. At the same time, the community should also establish a nighttime emergency plan to ensure the normal operation of the communication command system during nighttime disasters.

6.3. Shortcomings and Future Directions

Despite the contributions of this study, there are certain shortcomings. First, the quantitative nature of the study only explains the causal relationship between the studied variables. Thus, a qualitative study is encouraged to provide further information on the characteristics of the built environment, networks of community members, and their risk perception of the earthquake. Moreover, future research may pay more attention to social networks outside of the community and describe their mechanisms of action on residents’ evacuation behavior. Secondly, the conceptual model we built only explains a small amount of the variance of the influencing factors and actual behaviors, while some previous studies claimed that it was possible to explain a large proportion of the variance in predicting different behaviors in various situations using the integrated Behavior Model (IBM). Compared with that model, this study set a few scales regarding behavioral intentions, including attitudes toward the four behaviors, the perceived norms, and personal agency [66,67]. Therefore, future studies may use IBM to predict evacuation behavior more precisely. Third, there are limitations regarding the selection of questionnaire samples. The sample is limited to specific regions in Shanghai, and there is no confirmation with regard to the degree of compatibility with other cities. Furthermore, there are limitations in the selection of methods. The method used in this study is a questionnaire survey (Appendix A), and the survey objects include residents who have not experienced disasters. Therefore, the evacuation behavior selected by the questionnaire may be different from the actual behavior after the disaster occurs. Thus, it is essential to analyze the behavior of residents in other cities besides Shanghai, and these analyses can be combined with the interviews of accident witnesses.