Investigating Pedestrians’ Exit Choice with Incident Location Awareness in an Emergency in a Multi-Level Shopping Complex

Abstract

Limited studies have investigated pedestrians’ exit choices in an emergency in multi-level commercial buildings. In particular, the comparison between exit choices before and after awareness of an incident location is non-existent in the literature. Likewise, the influence of individual attributes, such as the presence of a child or a companion, on the individual’s exit choice in complex architectural layouts has rarely been studied in the literature. This paper aims to address these knowledge gaps by investigating pedestrians’ exit choice behavior in an emergency at a multi-level shopping complex considering exit choice behavior before and after awareness of incident location and the influence of personal attributes (e.g., presence of a child or companion). A survey of 1271 pedestrians for two hypothetical emergency scenarios in a multi-level shopping center in Tehran, Iran was conducted. A tablet-based simulator of a multi-story commercial complex was designed, and on-site interviews were conducted. In the first scenario, participants were asked to select their preferred exit door at the start of the emergency alarm without being informed about the incident location. In the next scenario, the scene of an incident (fire) was displayed without altering the conditions, and pedestrians were asked to choose their desired exit. The utility models investigated the differences in pedestrians’ behavior before and after awareness of the fire location. The models show differences in pedestrian decisions to evacuate and select the exit when the fire location information was available compared to when only emergency alarm information was available. Further, differences in evacuation strategy between the people who preferred to delay the exit and those who preferred to exit immediately were observed. Participants with children were more concerned about the ease of moving on the route and preferred a less congested route and exit area. Differences in evacuation behavior on the ground floor and other floors were also observed.

1. Introduction

The significant increase in population and the number of various large gatherings and events have made pedestrian crowd management a challenging issue for planners or managers of emergency response. Understanding pedestrians’ decisions in selecting the exit door and exit route is important to plan and manage the pedestrian crowd flow under normal and emergency conditions in public infrastructure such as transportation stations, sports complexes, and religious monuments [1]. Reliable information regarding pedestrians’ behavior and decision making is also important for developing pedestrian crowd simulation models. Erroneous information about pedestrians’ egress patterns in case of an emergency can lead to fatal incidents [2].

As with all human decisions, the choice of route and exit in public infrastructure is unique to each individual. Pedestrians may consider environmental factors, accompanying family or friends, emotions, and crowd size for route planning or exit choice. The characteristics or attributes of individuals, such as age, gender, fitness level, etc., may influence their decision [3,4,5].

A great deal of complexity makes it challenging to model and predict pedestrians’ actions, especially in emergency conditions. Therefore, random utility models have been used in the literature to examine the decision-making behavior of pedestrians in an emergency evacuation [6,7]. Particularly, the focus has been on identifying the parameters or factors that may influence pedestrians’ selection of an exit in the complex geometrical layout such as in train stations [7].

Despite the advances in modeling to simulate the pedestrians’ crowd and develop design solutions to reduce pedestrians’ casualties, experiences from past evacuation incidents show that people’s behavior during an evacuation differs from modeler’s assumptions [8]. For example, an interview conducted with survivors of the World Trade Center twin towers attack indicated that many people had paused for more than 17 min before attempting to leave, contrary to the assumption made by the modelers that people are expected to leave the escape area immediately [9].

Therefore, improving the safety of pedestrians in transportation systems and buildings, especially in emergencies, requires a deep understanding of people’s behavior during evacuation. Understanding pedestrians’ behavior during emergencies can assist in developing suitable design solutions that facilitate the evacuation process [10].

1.1. Objectives and Scope of the Study

Although pedestrians’ exit choice has been examined in transport hubs, such as train stations, the evaluation of emergency exit choice by pedestrians in multi-level buildings, such as shopping centers, is a gap in the literature. In particular, comparison between exit choices before and after awareness of incident location is non-existent in the literature. Likewise, the influence of individual attributes, such as the presence of a child or a companion, on the individual’s exit choice in complex architectural layouts has been rarely studied in the literature. This paper aims to address these gaps by investigating pedestrians’ exit choice behavior in an emergency at a multi-level shopping complex considering exit choice behavior before and after awareness of the incident location and the influence of personal attributes (e.g., presence of a child or companion) in exit choice behavior.

We developed an interactive digital application-based questionnaire survey in Tehran, Iran for this purpose. The questionnaire survey is based on hypothetical scenarios of an emergency evacuation. Although the questionnaire survey is popular in pedestrian behavior analysis, the lack of realism is a notable disadvantage of this method. By utilizing the interactive digital application-based questionnaire survey, it can increase the realism of the questionnaire survey since the participants can interact with the questionnaire in terms of their current location and see the details of exits and other surrounding features in real time. Similarly, using a digital questionnaire can increase the attractiveness of the procedure to the participants and gather information in a short time, which can increase the participation rate and sample size.

There are underlying differences in the purposes and surrounding environment of the train stations and shopping centers. Therefore, it is important to exclusively examine the pedestrians’ exit choices in a multi-level shopping complex. Various studies have shown how the different angled corridors [11,12] and the adjustment of small architectural configurations—e.g., middle exit vs. corner exit [13]—can influence the pedestrians’ movements and outflow. In this study, we systematically investigate differences in evacuation behavior by gender, age, the presence of one’s child, location familiarity, presence of companions, and evacuation start strategy, which have been less studied in human exit choice decision studies. The variation in exit utilities due to these differences might have implications in developing practical information about managing crowds in emergencies and improving the evacuation process.

1.2. Literature Review

The existing literature on pedestrians’ exit choice behavior suggests that several factors may influence their decision. The existing literature shows three different methods for exit discharge simulations: macro-simulation, micro-simulation, and meso-simulation [14,15,16,17]. The micro-simulation model tracks the details of people’s movements and interactions. On the other hand, the macro-simulation model shows the pedestrian’s aggregate behavior based on speed and flow relationships [18]. Finally, the meso-simulation model creates groups of pedestrians (instead of a single pedestrian) with every group allocated its own rules of behavior. The selection of appropriate simulation tools depends on the dimensions of the problem, the available computational power, and other factors [15].

Among these approaches, micro-models have attracted more attention to simulating crowd egress from buildings and transportation systems as they can capture additional delays arising from interactions among individuals, which is important during the evacuation process [19]. There are two main micro-evacuation model categories: deterministic and stochastic [20]. The deterministic approach is simpler to use, but it only explains the average behavior [14]. For this reason, these models cannot explain humans’ behavioral differences under different conditions. The stochastic approaches are suitable ways to simulate uncertain behavior since they can generate various behavioral examples with the same initial conditions [16,20].

The set of decisions consists of different parts. While approaching the destination, one of the most important parts of decision making is route selection. According to the related literature, route selection can influence people’s exit [6,21,22,23]. At the same time, route selection is essentially based on exit selection [24,25].

Exit selection modeling methods can be categorized into probabilistic and deterministic methods [6]. Deterministic approaches have been derived from different decision theories, such as the utility maximization theory [25,26,27,28]. Probabilistic models differ from deterministic models in their ability to exhibit behavioral variability and incorporate this behavioral uncertainty.

Numerous probabilistic methods have been used for the exit selection problem. Initially, they were designed as the exit selection model using the base probability allocation to select each exit in the model, although the base probability varies with conditions and previous choices of each exit. However, the model requires initial information to start the model, which could be challenging [29]. This problem is solved by utility models so that no initial information is needed for modeling. This has made probabilistic utility models highly popular to model pedestrian exit selection and decision making.

In terms of data collection methods, different approaches have been followed in the literature. They can be broadly divided into two main categories: “controlled and pre-designed studies (controlled laboratory studies)” and “studies on realistic conditions (observational studies)”. The laboratory studies include experiments with animals, controlled testing using humans and virtual reality, and hypothetical choice experiments. The observational field study group includes analysis of natural disasters, video analysis of pedestrians population movements in the natural environment, and interviews with survivors after the accident [30].

A comparison between the methods shows that virtual reality and hypothetical situations were better suited to conduct controlled and pre-designed research [30,31]. This provided a ground for choosing the hypothetical choice data collection method in the present study.

Virtual reality and hypothetical conditions are safe and practical methods for studying pedestrians’ behavior and evacuation experiments. In this way, it is possible to examine a wide range of parameters and conditions so that people’s reactions to several stimuli and parameters can be easily investigated. Under these conditions, many types of designs are examined and the optimal ones are selected [30]. One problem is the possibility of people’s different perceptions of real and simulated crowds. However, several studies have shown this concern to be insignificant, and hypothetical approaches can yield patterns and results that are observable in the real world [32,33,34].

A hypothetical approach can be used for exit selection as it is affected by the influence of social status [35,36,37]. In this approach, individuals could be exposed to various pieces of information and subsequently evaluate it. It has been observed that only demographic factors, such as age and gender, could not provide a good explanation for differences in people’s behaviors. Differences in individuals’ exit choice behavior under the same conditions have been reported, probably due to different personality characteristics as a latent variable [32].

The hypothetical choice methods used in other studies, in which discrete choice models or utility models are applied, are used and compared with deterministic approaches (selection based on distance) [6,38]. Research has shown that the presence of lights and flashlights in tall buildings can also influence people’s decision making [39]. There is a need for further research on whether individuals are less inclined to follow populations in unpredictable situations and whether they prefer emotional decision making over collective decision making [40].

In a recent study on supermarket evacuation by experiments and questionnaires [41], it was observed that the pedestrians’ strategy is to minimize their movement and select the closest exit even though a light congestion occurrence is possible. On the other hand, there is a tendency to offer help merely in poor visibility evacuation conditions.

An analysis of pedestrians’ likely behavior in emergency evacuation in a train station by using a questionnaire survey [42] suggests that the pedestrians are likely to be reactive (e.g., waiting for instructions from station staff) rather than proactive (e.g., moving to exit). There were significant differences in the behaviors between males and females, but not among the different age groups.

In the other evacuation study using virtual reality in a single-story building (museum) [43], it was observed that pedestrians exited through the familiar door that they had entered before. This tendency was, however, reduced if the neighbors exited by the unfamiliar door. The social influence (the effect of others’ behavior on one’s decisions) increases with the number of other individuals. However, this finding needs scrutiny in complex public infrastructure, such as multi-level shopping complexes, as the level of familiarity in different locations and building levels may vary, affecting the exit route choices. Especially in commercial complex buildings, the unfamiliar individual may not remember the entry easily since the journey inside could be disorienting due to the use of facilities such as stairs, escalators, and elevators. This may be further complicated by the fact that some of these facilities may not be available during an evacuation (e.g., elevators).

In summary, although there have been many studies on pedestrians’ exit choices, limited studies have been conducted in the complex multi-level shopping complex. In particular, there is a lack of comparative study on pedestrian exit choices considering before and after awareness of the incident location. The use of a digital interactive questionnaire survey to enhance the realism of the participants’ responses has been also limited in the literature. More studies of pedestrians’ exit choices in different types of public infrastructure and geographic regions will increase the confidence in our understanding of behavioral parameters that influence pedestrian exit choices in an emergency evacuation.

The remainder of the article is organized as follows. Section 2 presents the design of an interactive digital application for a questionnaire survey, the selection of the parameters for the survey, the conduct of the questionnaire survey, and the methodology adopted for this study. Next, Section 3 shows the results of incident location awareness on pedestrian exit choice and individual attributes influencing exit choice decisions. Finally, Section 4 draws conclusions and discusses future research directions.

2. Materials, Survey, and Methods

This section first describes how an interactive digital application was developed to overcome the limitations of the paper-based survey questionnaire. The parameters considered in the survey and the steps involved in their selection are then described. It is followed by an explanation of the survey and data collection procedure. Finally, the utility models adopted to analyze the collected data are described.

2.1. Interactive Digital Application

We developed an interactive digital application (app) that can be run on a mobile phone or a tablet for data collection via a survey. Before the survey, the parameters to be measured were determined based on an extensive literature review, focus group discussions, and the pilot interview process. We then used the app to record the participants’ responses to the exit selection. The collected data were then analyzed using the utility models. In the following sub-sections, we provide detailed information on the digital app for the survey, selection of parameters, questionnaire preparation, data collection, and respondents distribution.

The use of paper-based questionnaires is a common practice for pedestrian behavior research. The process is to provide a location plan and a written description. The need for the participant to imagine the situation in this method is one of the weaknesses; in that sense, it lacks realism. This method is also less welcomed by the participants as it is time-consuming. Similarly, there are chances that the questionnaires are not completely filled. Another disadvantage is the effort required to convert questionnaire information into a data file that can be processed and evaluated. Environmental concern due to paper consumption is also a problem.

The interactive digital app for the survey questionnaire is developed to address the above issues. In this app, the map or plan of the site is created with all the relevant geometric details that pedestrians interact with (e.g., exits, obstacles, corridors, etc.). The users can specify their location on the map.

In addition to the details, the app can generate pedestrian crowds in the corridors and routes. The number and density of people can be generated randomly to create low- to high-density conditions. The program automatically saves data related to population density, visibility, and distance from exit doors. After selecting the desired exit by the user, all information is saved in a spreadsheet for processing.

Snapshots of the shopping complex and other design features used for this study, along with interactive floor plans for the shopping complex generated through the digital application, are presented in Appendix A (Figure A1, Figure A2, Figure A3, Figure A4, Figure A5, Figure A6 and Figure A7). The details on the data collection process using this interactive app are presented in Section 2.3.

2.2. Selection of the Parameters

Various parameters were used by researchers regarding the problem of exit choice in the literature, such as flashing lights for emergency exit usage [23], size of the crowding near exits, size of the moving flows, distance to exit, visibility of exit [33], and limited hearing and visibility conditions [44,45]. To have a realistic and reliable number of parameters, we examined the parameters available in the literature and extracted parameters from a pilot survey of participants. Using this combined approach, we extracted relevant parameters for this study, as described in the next paragraph.

First, a pilot study was conducted to examine the factors based on which people might select an exit among multiple options. A total of 50 people were asked (in the form of a focus group discussion) to state their consideration of factors for an escape through an exit in an emergency condition. The participants were selected randomly with an appropriate mix of ages and gender and were asked, “What are the important factors that you would consider in selecting an exit in a building in an emergency evacuation?”. This experiment obtained a list of factors influencing the participants’ decision on exit choices. In the next step, 100 people were asked to score the previously obtained parameters based on importance (participants chosen randomly from different ages and gender). Finally, considering the results from the focus group, findings from the past research and considering the current research limitations, the following factors or parameters were extracted and considered for the survey:

• Distance to exit.
• Distance of fire to exit.
• The crowd size around each exit.
• The crowd size in the participant’s path to each exit.
• The number of people heading to each exit, referred to as demand for each exit in the paper (it was assumed that people usually select the closest exit).
• Visibility of the exit door from an initial location of a participant.

Likewise, some pedestrian attributes were also considered as below:

7.

familiarity with the place;

8.

age;

9.

gender;

10.

having a child or a companion (friends/family members).

During the focus group, several people mentioned that they would help others instead of exiting immediately in emergencies. Therefore, the question “Will you exit immediately?” was also included in the questionnaire. This study was conducted to identify issues that may be important among individuals but are rarely considered in the existing literature.

2.3. Questionnaire Survey

The commercial shopping complex of Kourosh, located in the west of Tehran, was selected as the survey site for this study. This building consists of multiple floors (2 basements, the ground floor, and seven more floors, Figure A1). The building has a large number of shops, restaurants, movie theaters, coffee shops, hypermarkets, and playgrounds. People of different ages and social statuses go to this place daily. Hence, there is a notable diversity of pedestrians, which helps to collect more comprehensive data required for the study.

The reasons for choosing this shopping complex are due to a variety of environmental factors, namely, having multiple floors and a complex layout. Likewise, the other considerations included the presence of a large number of people (an average of more than 10 thousand visitors each day) and the opportunity to obtain a diverse range of people, including teenagers, adults, and the older population.

The survey was conducted in March 2019; lasted for 25 days, from 11:00 AM to 8:00 PM daily. This was conducted to ensure good coverage of variation in genders, ages, individuals’ companion status, and other characteristics so that the data are comprehensive and representative of the population typically expected in a shopping complex. The survey took approximately 4 to 5 min to complete for each participant.

To conduct this research, it was necessary to determine different scenarios and to make decisions in different situations. For this purpose, in the app, plans of the different floors of the Kourosh Complex were imported and used as simulators. The basement, ground level, floor 1, floor 2, and floor 4 were considered in this study as they represented a good coverage of the physical characteristics of different floors of the complex (refer to Appendix A—Figure A1). The app represents exactly the same location of participants. This approach helps in real time decision making and makes the results realistic.

For this questionnaire survey, a 10-inch tablet was used. The questionnaire was shown to the participants on the tablet so that they could receive the relevant information and the assumptions from the picture and make their choices in different scenarios.

The questionnaire was used to collect data in a series of steps. First, general information about the person was entered (Figure A2). After selecting the floor on which they were located (Figure A3), the map of the floor appeared. Each time the program was executed, the distribution and the number of people on the floor were randomly generated (black dot with a white arrow—Figure A4). The black dot (Figure A4) represents the other people on the floor, and a white arrow on each of them represents their chosen exit to evaluate “the demand per exit” in the considered scenario. The map gives every detail of the floor (routes, obstacles, exits, crowd distribution, and direction); hence, the participants were able to make an informed decision.

Before selecting an exit, the strategy to exit immediately was asked to the participant in the app (Figure A3). Then, the participant was asked to select the desired exit in case of an emergency evacuation. The last two steps were repeated by showing the incident location (Figure A6). The participants’ answers were based on the information on the map and explanations from the interviewer about details on the map (e.g., fire location, crowd size, location of exits, distance).

As shown in Figure A2, at the beginning of the interview, general information was entered (age, familiarity with the place, gender, being with family or friends, having a baby or carriage, and the floor on which the person is present). In family groups, only one person was interviewed (adults); in friend groups, more than one person was interviewed separately. By confirming this part, the selected floor plan was then shown (Figure A3).

At this point, the screen presented an emergency. First, the participant was asked if he or she would exit immediately (or delay) under these conditions. This question aims to establish the state of mind during emergencies and examine people’s initial choice of emergency exit. As shown in Figure A3, the display’s background goes flashing red and white.

The participants were told that the sound of the emergency siren was audible and the type of emergency and threat was unknown. After answering this question, the screen became transparent, and the participant could view the current state of pedestrian crowd distribution on the floor and exits. In this case, as shown in Figure A4, a white arrow was displayed on each black dot, showing the targeted exit among the 4 exits for each individual. The participant’s current location was displayed as “X” (Figure A4). In this situation, the participant was informed about the movement of other people and their destinations in the initial moment of emergency.

It is to be noted that the crowd number (size) was produced randomly for each participant. Afterward, the participant selected his or her desired exit for the emergency egress. The exit selection procedure repeated after showing the fire (as incident location, Figure A6) somewhere random in the plan. These sequential steps help to compare the changes in decision-making behavior after awareness of the incident location.

In total, 1271 people were interviewed. As discussed before, due to the attractiveness of the questionnaire and quick process, participants finished all the questions. A wide range of ages (8 to 75) participated in this study. The young students (8 to 18 years old) were accompanied by a school teacher and were interviewed with permission from the school teacher.

Of the 1271 respondents, 48.5% were male, while 51.5% were female. In terms of age distribution, 3.5% of the respondents were under 18 years old, followed by 26.5%, 27.5%, 18.3%, 10.6%, and 3.4%, respectively, for age groups (18 to 25 years), (25 to 35 years), (35 to 45 years), (45 to 60 years), and (above 60 years).

Regarding participants’ distribution on each floor, 24% were on the ground floor and about 19% on each other floor. In terms of companionship, 20.1% were alone, 25.6% were with friends, and 53.6% were with their family accompanying them in the building.

In terms of location familiarity, 23.5% visited the place on less than 5 occasions (low familiarity), 62.2% visited the place more than 10 times (high familiarity), and 14.3% have medium/average familiarity (between 5 and 10 times). Regarding the presence of children, about 20% of the participants had children with them.

2.4. Methodology

In terms of pedestrians’ decision making in an emergency in the shopping complex, they can directly exit the building when they are on the ground floor. In contrast, they have to choose the emergency stairs when they are on other building floors. Participants had the choice of 4 exit doors on each floor, as shown in Figure A4. Since the participants were selected from each floor and the crowdedness was random, the utility models are suitable for investigating pedestrian exit choice behavior.

Economic models based on discrete choice (also known as utility models) were examined to investigate pedestrians’ behavior in decision making and exit selection. Among these models, the multinomial logistic regression model has been deployed in this study. The reason for choosing this model is the need for a comprehensive assessment of pedestrians’ stochastic behavior. Although external instructions and group dynamics may affect pedestrians’ behavior, their behavior is largely stochastic. As stated earlier in the literature review section, utility models have been popular in pedestrians’ exit choice behavior and are able to capture realistic pedestrian behavior [30,33].

The utility of choice is shown by Equation (1) where the “V” represents the deterministic part of utility, and “ε” is the random error component of utility. The parameter “V” is a linear equation related to the parameters of interest (Equation (2)). The coefficients in the equation are specific for each data set and show the best relation between conditions and the selected object. On the other hand, the value and sign of each coefficient show how greatly it affects pedestrian choice. Finally, the probability of each choice in the data choice set is based on Equation (3). For estimating β values, the open-source program (biogeme) is used [46].

$$U=V+\epsilon$$

(1)

$$Vc=\beta c\ast X=AS{C}_{En}+{\beta }_{DIS}\ast EnDIS+{\beta }_{CA}\ast EnCA+{\beta }_{CW}\ast EnCW+{\beta }_{CT}\ast EnCT+{\beta }_S\ast EnS+{\beta }_{FD}\ast EnFD$$

(2)

$$Pr\left(Y=c\right)=\frac{e^{Vc}}{{{\displaystyle \sum }}_{k=1}^K{e}^{Vk}}$$

(3)

The parameters used in the random utility models are:

• ASC_En: Alternative-specific constant, representing the inherent utility of exit.
• β_DIS: coefficient of person distance to exit.
• β_CA: coefficient of the crowd size around exit point.
• β_CW: coefficient of the crowd size in the path to exit point.
• β_CT: coefficient of the crowd size going toward exit point (exit demand).
• β_S: coefficient of the visibility of each exit point.
• β_FD: coefficient of fire location distance to exit.
• EnDIS: the distance from the origin to exit point divided by 100* (exit number n).
• EnCA: the crowd size around each exit point divided by 100*.
• EnCW: the crowd size in the path to each exit point divided by 100*.
• EnCT: the crowd size going to each exit point divided by 100*.
• EnS: the visibility condition of the entrance of each exit (visible (1), invisible (0)).
• EnFD: the distance from the fire location to the exit point divided by 100*.
• *: divided by 100 to make all coefficients’ values in the same order.

In each model, the utility equation is defined for each exit and is obtained by the theory deployed in this method (values that do not reach a significant limit—99% confidence level—are omitted and assumed to be zero).

The overall steps adopted for this study are shown in a flow chart (Figure 1). While these steps are generalizable to study pedestrians’ exit choice behavior in other public buildings or infrastructures, care needs to be taken to ensure that prior investigations have been carried out regarding the relevant purpose and architectural configurations of the buildings or infrastructures.

3. Results and Discussions

In this section, models of emergency scenarios are examined. As mentioned, in this case, people choose their exit in a scenario where a hypothetical danger alarm is activated and the same consecutive scenario where the incident location is known. This section provides the results using all the data and a series of other results obtained by segmenting the data based on various information (floor number, gender, presence of family or friends, exit strategy—rapid (immediate) egress or delayed (wait for) egress).

The reported coefficients in this research all have a confidence level of 99 percent. Please note that some coefficients have been reported as not applicable (NA) as they have a p-value near 0.2 and are thus not reliable.

3.1. Emergency Exit before and after Incident Location Awareness

The first consideration is to examine the choice of all participants in the building concerning the choice of exit. This evaluation is performed when the participants are told that the alarm is activated. In this case, the following physical parameters are considered: distance to exit, crowd size around each exit (which differs from very low to a very high number of pedestrians on the floor randomly), number of people moving to each exit (demand per exit), number of people around the participant’s path to exit, and visibility of exit door.

In the second scenario, we have a new parameter, the fire distance to the exit. Comparison between the coefficients of the two models represents the change in pedestrian exit choice after being aware of the incident location. According to

Table 1. Emergency exit before and after incident location awareness model coefficients.

Scenario	βDIS Person Distance to Exit	βCA Crowd Size around Exit	βCT Crowd Size Going toward Exit	βCW Crowd Size around the Path to Exit	βS Visibility of Exit	βFD Fire Distance to Exit
Before awareness	−4.4	−1.71	−0.66	−2.48	0.48	-
After awareness	−3.05	−1.22	NA *	−0.93	NA *	2.62

* Not applicable (NA) due to p-value near 0.2 and thus not reliable.

, the influence of common parameters weakens, and the fire distance to the exit plays a significant role. Pedestrians tend to go to the far and crowded exit to avoid being close to the fire location.

With awareness of the fire location change, people displayed different behavior in the second phase situation, as evidenced by the parameters’ coefficient changes. Comparing parameters’ coefficients before and after awareness of the fire location leads us to some important observations. First, pedestrians’ tendency to avoid the crowd decreased for evaluating exit choices after the awareness of the incident location. Second, the number of the crowd toward the exit is not significant. Third, the utility for person distance to exit decreased. Finally, the fire distance to each exit had a positive influence on the exit utility.

On the other hand, based on the results, a comparison can be made regarding individuals’ strategies for exit delay in an emergency situation and intentional pre-evacuation time. It is found that about 40 percent of pedestrians tend to deliberately stay in emergency situations for various reasons, but after locating the fire, this percentage decreases to 12 percent. As a result, the optimal parameters for modeling the effect of pedestrians’ awareness of fire location on exit selection are evaluated (Table 1). It is worth mentioning that the inherent utility of each exit door is close to zero and could be neglected, so the utility is estimated by only considered parameters, as shown in Table 1.

3.2. Emergency Exit Choice at a Disaggregated Level

In this section, pedestrians’ exit choice at a disaggregated level is investigated considering the influence of floor level, gender, presence of children, location familiarity, presence of a companion, and evacuation start strategy.

3.2.1. Floors of the Shopping Complex

It is to be noted that the evaluation of differences in exit choice on multi-level floors of the shopping complex is almost non-existent in the literature [30]. In

Table 2. Disaggregated models coefficients for emergency exit choice.

Disaggregated Level	Sub-Group	βDIS	βCA	βCT	βCW	βS
Floors	Basement One	−4.2	NA *	−0.81	−3.83	NA *
	Ground floor	−6.28	−4.00	NA *	NA *	NA *
	Floor One	−4.51	NA *	NA *	−5.9	NA *
	Floor Two	−3.8	3.29	−1.98	−2.84	NA *
	Floor Four	−3.45	NA *	−2.56	−2.96	1.1
Gender	Male	−4.1	−1.67	−0.79	−2.14	0.42
	Female	−4.48	−1.46	−0.55	−3.23	0.50
With child	Yes	−5.49	NA *	−1.53	−4.91	NA *
	No	−4.31	−1.94	−0.52	−2.21	0.46
Familiarity	High	−4.56	−1.67	−0.79	−2.26	0.44
	Average	−5.37	NA *	−1.36	−5.02	NA *
	Low	−4.48	−1.92	NA *	−3.04	NA *
Companion status	Family	−4.98	−1.44	−0.81	−3.38	0.43
	Friend	−4.36	−2.09	−0.76	−1.74	NA *
	Single	−4.87	−2.68	NA *	NA *	NA *
Escape strategy	Rapid exit	−4.96	−1.79	−0.46	−2.91	0.53
	Delayed exit	−4.36	−1.15	−0.10	−2.47	NA *

* Not applicable (NA) due to p-value near 0.2 and thus not reliable.

, the values of significant coefficients of physical parameters in the models are shown. Each parameter belongs to a particular floor shown in a row. “Crowd size around each exit (β_CA)” was significant on the ground floor with a negative coefficient. Given that the ground floor model is a function of this parameter and distance, the high value of “crowd size around each exit” coefficient can be attributed to the direct egress using the exits on this floor. On other floors, the exits are the emergency exit stairs, but the direct exit from this ground floor has drawn people’s attention to the crowd size around the exit. Furthermore, on the ground floor, people have chosen the nearest exit and have made a choice regardless of the crowd size of the route and the exit demand.

Another point to consider is the positive effect of the density coefficient for the second floor. According to surveys performed during data collection, a children’s play area on the second floor obstructs a view of exits one and two (Figure A5). Therefore, this area faces uncertainty regarding emergency escape due to the limited visibility of the exits. Our finding is consistent with other studies where it has been observed that the architectural configurations, such as hidden exits or exits with limited visibility, influence the pedestrians’ exit choice behavior [33,44,45]. As the configuration of the floor obstructs the two exits, the participants chose crowded exits on the other side of the floor where there was no obstruction to the visibility of the exit. This shows that when there is uncertainty on the exit location due to the reduced visibility, pedestrians try to avoid moving toward those exits and instead choose to move toward crowded exits. However, such a positive factor associated with crowded exits does not indicate the tendency of people to choose crowded exits and acts as a latent variable. As it is observed, this coefficient is not significantly affected on the other floors, and more people are evaluating the route congestion and exit demand for their escape decision. This may be due to the large floor area and the presence of a large number of people in the corridors surrounding the exit.

The parameter coefficient “the number of people moving to the exit (demand per exit) (β_CT)” has the most negative effect on the fourth floor. This may be due to the participants being on the higher floor and looking for an exit to avoid other pedestrians and congestion. Regarding the “crowd size in the path to exits (β_CW)”, this parameter has the highest value on the first floor, followed by the basement floors. Being on a floor close to the ground seems to pay more attention to this parameter. Specifically, for the first floor, it is evident that only “route congestion” and “distance” are the factors that influence the choice of the closest and most desirable route, regardless of the competition issue created by the congestion around the exit and the exit demand. The next parameter is the distance (β_DIS). As expected, this parameter significantly affected individuals’ choices across all floors, with a significant coefficient.

Finally, the visibility of each exit point (β_S) does not significantly impact the model. However, on the fourth floor, the visibility positively impacts the model. This may be due to the absence of crowded shops, playgrounds, or restaurants on the fourth floor.

3.2.2. Gender

While some studies suggest that gender does not influence exit choice, other studies have found a gender effect in the evacuation process [30,42]. According to our results shown in Table 2, the behavior of men and women did not differ significantly based on the coefficient of the parameters. One notable observation is that women participants tend to focus on crowding and try to avoid it. In contrast, men participants are more concerned about the number of people heading to each exit and the crowd size around the exit. Regarding choosing between immediate and delayed exits, women tended to choose immediate or delayed exits as equally as men. It is to be noted that no significant effect due to age differences was observed and, therefore, has not been discussed.

3.2.3. Presence of Children

The effect of the presence of a child on the person’s exit choice is missing in the literature. Based on Table 2, having the presence of children eliminates the importance of “the crowd size around the exit (β_CA)” and “visibility of exit door (β_S)”. For participants who had children with them, the “congestion of the route (β_CW)” and “the number of people heading to each exit (β_CT)” had the most negative effect. Similarly, the distance to the exit reduced its utility. These results suggest that people with children are more concerned about the ease of moving on the route and prefer a less congested route and exit area.

3.2.4. Location Familiarity

Participants who visited the place “less than 5 times”, “more than 10 times”, and “5 to 10 times” were considered to be low familiarity, high familiarity, and average/medium familiarity, respectively. As shown in Table 2, participants who were familiar with the place were relaxed while choosing the exits, as shown with low coefficients for high familiarity as compared to medium and low familiarity. Another interesting observation is the significant differences in the impact of different parameters depending on the level of familiarity. For example, only the high familiarity group pays attention to the exit’s visibility. Likewise, the average familiarity group cares more about the congestion than other groups. It is to be noted that these differences are obtained in the case of complete information given to the participant. In several past studies that conducted controlled laboratory experiments to examine the influence of various architectural configurations (e.g., turning, and merging corridors) [47,48,49] on pedestrian behavior and outflow, it has been suggested to optimize the number of repetitions of experiments. It has been argued that the familiarity of the experimental layout with repetitions may influence pedestrians’ behavior. Further, past studies [21,50] have highlighted that pedestrians would move toward familiar places in an emergency and avoid unfamiliar places. Therefore, in case of an emergency evacuation, the differences in the level of familiarity may play a decisive role in the efficiency of the evacuation process.

3.2.5. Individual’s Companions

Whether alone in a public place or with family or friends may affect a person’s exit choice. However, this issue has not been systematically investigated in the existing literature, regarding people’s exit choices in emergencies. According to Table 2, in the case of being alone, crowd size in the path/route congestion (β_CW) and exit demand (β_CT) did not influence the exit selection decision. People seem to pay attention to congestion around the exit and think that they will arrive at the exit faster than others. Hence, they fail to consider the demand for the exit. Further, the coefficient for route congestion with family is nearly two times more than that with friends, which indicates that individuals with their families prefer to choose a more secluded route because they may not want to leave the family behind. Similarly, more crowded routes may increase the risk of getting separated. In contrast, the presence of friends may not create such restraint, and people may separate from friends and seek the best egress strategy for themselves. With friends, exit visibility does not significantly affect peoples’ exit choices. The visibility parameter has more influence in the case of being alone, which is probably due to the person’s ability to concentrate more on finding and observing the exit location depending on the environment.

3.2.6. Evacuation Start Strategy

One of the important issues discussed in the literature is whether people display a proactive strategy by exiting quickly or a reactive strategy by waiting to assess the situation [42]. One issue that needs to be addressed is the way people subconsciously behave. In other words, it is important to recognize the people who are willing to exit quickly and those who tend to delay. By asking people’s opinions in this regard, a model was developed by categorizing the data based on rapid exit and delayed exit. According to the responses, 513 people (40% of the participants) stated that they would wait (delayed exit), while the majority (60%) of the respondents stated that they would exit immediately (rapid exit).

The actual behavior of people in case of an incident may be different from the views they have stated in the questionnaire [42], but the model results, as shown in Table 2, are important because people who have responded that they will not attempt to exit quickly usually will take some time to evaluate the situation and make a choice. Therefore, as shown in Table 2, it can be seen that this group of respondents has a different evaluation of the parameters. When people choose to wait, they may perceive that the crowd size around each exit will vary by the time they leave. Therefore, how the crowd size around each exit changes may influence people’s decision making. At the same time, other factors with less significant influence may be given smaller weights.

4. Conclusions

Understanding the complexity of pedestrian’s exit choice decision making is important for efficient evacuation management during an emergency evacuation at major public infrastructures. This study examined pedestrians’ exit choice behavior in an emergency in a multi-level shopping complex. Specifically, this study analyzed pedestrians’ exit choice considerations before and after awareness of the incident (fire) location. For this purpose, a digital interactive app was designed using the floor maps of the shopping complex. It was observed that the conduction of the questionnaire survey via the interactive digital app was viewed positively by the participants as the survey could be completed with ease and also increased the realism of the survey. It is recommended to use the digital interactive app in such a questionnaire survey in the future.

A total of 1271 individuals, aged between 10 and 75, with an almost equal distribution of men and women, participated in a questionnaire survey, conducted using the digital app. The usual emergency situation starts with the emergency alarm and asking people to egress. However, as shown in this study, it is crucial for emergency planners to understand the change in pedestrian behavior after being aware of the incident location. The models show differences in pedestrian decisions to evacuate and select the exit when the fire location information was available, compared to the situation when only emergency alarm information was available. It is also important to understand how people evaluate their path to the exit since path selection directly affects the exit selection. Based on interviews, the suggested parameters for evaluating the effect of fire on exit utility and pedestrian’s behavior in emergencies are as follows: 1. fire-to-exit distance, 2. fire-to-person distance, and 3. the shortest distance of fire-to-person-to-exit route.

The differences in pedestrians’ exit choices resulting from the complexity of the layout of shopping centers and the characteristics of the escape area of the floors were analyzed. Specifically, the “distance of the exit” and “crowd size” parameters were considered for exit choice. Likewise, the impact of “floor level”, “gender”, “familiarity with the place”, “presence of a child”, “presence of family or friends”, and “rapid exit vs. delayed exit strategy” were investigated.

Most respondents (60%) preferred to exit immediately compared to 40% who stated they would delay evacuating and wait (delayed exit strategy). However, the delayed exit strategy decreased to 12% after locating the fire. Further, the awareness of incident location decreases the influence of various parameters. For example, the model coefficients influence of “crowd size around the path to exit” decreased by 62%, while “crowd size around exit” decreased by 28%. Likewise, some parameters that were significant before awareness of the incident location were no longer significant after awareness of the incident location (e.g., crowd size going toward exit). Not surprisingly, the incident (fire) location distance to the exit greatly influenced pedestrians exit choice (with the model coefficient value of 2.62).

It was found that both characteristics of the layout of the escape area as well as pedestrian’s attributes can influence pedestrian’s exit choice behavior. Some differences in pedestrian exit choice at different floor levels were observed depending on the characteristics and architectural layout of the floor plan. On the ground floor, participants chose the nearest exit regardless of the crowd size of the route and the exit demand, which is different from that observed on other floors where the crowd size of the route and exit demand were critical parameters in exit choice decision. No gender differences were observed in the exit choice decision. The presence of children or the presence of companions affected the exit choice decision. For example, participants with children were more concerned about the ease of moving on the route and preferred a less congested route and exit area. As compared to being with friends, participants with their families preferred to choose a more secluded route as they may not want to leave the family behind. It was observed that the differences in the level of familiarity could be important in the efficiency of the evacuation process. Differences in evacuation strategy between the people who preferred to delay the exit and those who preferred to exit immediately were also observed.

There are some limitations to our study. Responses from the survey comprise self-reported behavioral intentions and not actual behaviors and could be affected by social desirability bias, where people may report imprecisely on sensitive topics to present themselves in the best possible light [37]. Therefore, it is suggested to supplement the survey findings through incident investigation and survivor interviews in future studies. Further, through analysis of pedestrians’ walking behavior in five countries (Iraq, Israel, England, Canada, and France), researchers [51] have highlighted that it is important to consider the impact of cultural differences when developing modeling for crowd evacuation. The present study only focused on Iranian participants; therefore, it is suggested to conduct a similar study in other countries to average cultural bias.

Conducting similar research in other buildings with different characteristics and architectural configurations can provide more useful insights into the pedestrians’ decision of exit choice. Several studies have found that complex architectural configurations, such as turning, merging, and crossing corridors, can influence pedestrian flow in emergency situations [49]. Likewise, partial architectural obstruction, such as a column near an exit, can substantially influence pedestrian outflow in emergency situations [52]. Therefore, future studies can focus on people’s exit choices considering these complex architectural configurations and pedestrians’ perception of obstacles near exit/door location and the functionality of doors. Furthermore, the influence of stress, physical abilities of people, and group behavior can also be examined.

Nevertheless, this study helps to obtain a better understanding of occupants’ decision-making behavior while evacuating a public place in case of an emergency and subsequently, develop engineering or architectural design solutions to improve pedestrians’ evacuation process. For example, engineers can develop technology that can quickly inform pedestrians regarding the location of the incident. This can reduce the delay in egress due to confusion and unfamiliarity with the incident location, as noted in this study. Likewise, architects can design exits that are visible to occupants at any time or ensure that the amenities on the floor do not obstruct the exit. Understanding pedestrians’ exit choice behavior and their behavioral differences in the presence of children or companions can assist the planners or managers of emergency response in developing targeted evacuation management strategies considering the heterogeneity of pedestrian crowds.