This study aims to understand the street-crossing-facility choice behavior of the elderly. The footbridges were classified into three types with different levels of convenience. On the other hand, the safety of the crosswalk was assumed to be highly related to the remaining time of pedestrian green light. An adaptive SP survey was designed to collect choice data from individuals aged 60 years or above, and a multilevel logistic model was developed to analyze the behavior data. The results indicated that convenience and safety exert significant influence on the facility choice behavior. The choice behavior of the elderly toward making trade-off decision between convenience and safety was further discussed through the multilevel logistic regression analyses.
As individuals age, fewer journeys are undertaken as car drivers and a higher number of journeys are undertaken as pedestrians (Dunbar et al. 2004). Owing to age-related decline, elderly pedestrians could encounter higher safety challenges while crossing streets. For example, Asher et al. [
This study attempted to analyze the street-crossing-facility choice problem encountered by the elderly in terms of behavior analyses. Certain street-crossing facilities, such as footbridge, can provide a high safety level for elderly pedestrians by physically separating the pedestrian flow from the vehicular traffic. However, the inconveniences of longer travel distance contributed by up-down movements and of higher body moments requirement may prevent elderly pedestrians from using them. Elder pedestrians are required to make a trade-off choice between safety and convenience. The street-crossing facilities considered in this study were footbridges and signalized crosswalks with countdown timer (unsignalized crosswalks and pedestrian underpasses are unusual in Beijing). Equipping footbridges with escalators can reduce their inconvenience [
A number of studies have indicated that the perceived safety and convenience can significantly impact street-crossing behavior. For example, Havard and Willis [
It is worthwhile to note that previous studies generally considered the utilization of crossing facilities (e.g., crosswalks and footbridges) separately through field survey data; therefore, there is still a requirement for investigating elderly pedestrians’ behavior toward the decision between footbridges and crosswalks. To further explore the elderly pedestrian’s underlying crossing-facility selection behavior, we conducted an SP survey for collecting corresponding behavior data. We recognized that the attributes of street-crossing facilities can be rather involved for the elderly to understand when described by textual representation. Therefore, the traditional paper-based SP survey may not effectively collect street-crossing-facility selection behavior data, particularly if the responders of the survey are elderly. On the other hand, a computer-based SP survey could display attributes of alternatives through multimedia such as videos and 3D images. Multimedia provides a straightforward method to allow responders to accurately understand the features of alternative facilities and thus reduce perception bias. A few researchers have already adopted computer-based SP surveys to collect travel behavior data. For example, Tilahun et al. [
This study focused on elderly pedestrians’ street-crossing behavior and investigated a key problem that has not been discussed in the previous studies. We attempted to understand the behavior of elderly pedestrians toward trade-off decisions between safety and convenience when they choose various types of street-crossing facilities. The remaining paper is organized as follows: Section
We examined three types of footbridges (types I, II, and III), as shown in Figure
Type I, type II, and type III footbridges.
We were mainly interested in understanding the effects of installation of escalator (corresponding to convenience) and of the length of remaining time of pedestrian green light (corresponding to safety) on street-crossing-facility choice behavior. Thus, the experiment scenario for street-crossing-facility choice was designed as follows: the locations of footbridges are as close to crosswalks as feasible in order that the distance between a footbridge and a crosswalk can be omitted in the behavior analysis. To prevent unnecessary complexity in the pedestrian light setting scenario, this study did not consider the waiting time owing to pedestrian red light and illegal street-crossing behavior is also beyond the scope of this research.
An adaptive SP survey was conducted to collect the data related to street-crossing behavior. In the survey, it was assumed that the width of the street that pedestrians were required to cross was 32 m. Accordingly, the remaining time of pedestrian green light was set within the range from 26 s to 52 s (Knoblauch et al. [
An example of survey questions.
Responder ID | Question number | Footbridge | Remaining time (s) | Choice result |
---|---|---|---|---|
1 | 1 | Type I | 32 | Crosswalk |
1 | 2 | Type I | 29 | Footbridge |
1 | 3 | Type II | 34 | Footbridge |
1 | 4 | Type II | 37 | Crosswalk |
1 | 5 | Type III | 37 | Footbridge |
1 | 6 | Type III | 40 | Footbridge |
2 | 1 | Type I | 31 | Crosswalk |
2 | 2 | Type I | 28 | Footbridge |
2 | 3 | Type II | 33 | Footbridge |
2 | 4 | Type II | 36 | Crosswalk |
2 | 5 | Type III | 36 | Footbridge |
2 | 6 | Type III | 39 | Crosswalk |
The remaining time of pedestrian green light in each question was adjusted according to the response to the previous question. For the In the cases where In the cases where In the cases where We limited the remaining time of pedestrian green light to the range from 26 s to 52 s. If the rule required the remaining time in the
For each question, we used a sequence of 3D computer images to display the attributes of the footbridge; we selected eight images for eight points of view and displayed these images to a responder one by one. We did not use actual photographs to display footbridges because the background and the shooting angle of the photographs can impact the behavior survey and raise bias. On the other hand, 3D computer images permit us to maintain a consistent background in all the images and provide a straightforward method to present a footbridge from various points of view.
We used a synthetic video to enable responders to experience the remaining time of pedestrian green light (Figures
Synthetic video: the time remaining for the pedestrian green light is 35 s.
The command line tool that is developed for managing the survey and generating questions (in the screen shot “A” indicates type I footbridge).
Figure
We developed a command line tool to manage the SP survey (see Figure
We enrolled 169 responders aged 60 years or above from four communities located at Haidian district, Beijing (Table
Responders from four communities.
Community | Number of participants | Gender distribution | Average age (year) | |
---|---|---|---|---|
Male | Female | |||
Beijiaoda | 40 | 27% | 73% | 70.1 |
Zaojundongli | 50 | 38% | 62% | 65.4 |
Zaojunxili | 44 | 34% | 66% | 67.0 |
Tiekeyuan | 35 | 25% | 75% | 74.5 |
The adaptive SP survey system can generate questions that are more targeted than random questions. The survey system automatically adjusts the remaining time of green light according to the answer of the last question. After the survey, we analyzed the attributes of the questions. We grouped the questions by the bridge type and calculated the average remaining time for each group (i.e., for each bridge type). Figure
Characteristics of the remaining time of pedestrian green light.
In order to consider the heterogeneity among the responders, this study employed a multilevel logistic model to analyze the behavior data collected by the adaptive SP survey. The model can be formulated as follows:
The combination of the variables
Footbridge |
|
|
---|---|---|
Type I | 0 | 0 |
Type II | 1 | 0 |
Type III | 1 | 1 |
In this study, the choice behavior is described by a utility function with a linear structure. The structure assumes that individuals make a trade-off between the attributes when taking a decision. On the other hand, certain behavior traits that are inconsistent with the assumption of the specified model are likely be observed in the SP survey. Hess et al. [
We identified the nontrader by the following procedure: responder
Demographic information of responders.
All responders | Nontrader | Trader | |
---|---|---|---|
Number of observations | 169 | 20 | 149 |
|
|||
Sex | |||
Male (%) | 32% | 40% | 31% |
Female (%) | 69% | 60% | 69% |
|
|||
Age | |||
Mean | 69.7 | 74.0 | 69.2 |
SD | 8.2 | 7.6 | 8.1 |
Maximum | 89 | 86 | 89 |
Minimum | 60 | 60 | 60 |
We conducted an ANOVA to analyze the variation between traders and nontraders. The result of the ANOVA indicated that the mean age of responders with nontrading is significantly higher than that of responders with trading (see Table
ANOVA analysis for trader and nontrader.
Source of variance | Sum of square | Degree of freedom | Mean square |
|
---|---|---|---|---|
Between | 411.11 | 1 | 411.11 | 6.37 |
Within | 10774.64 | 167 | 64.52 | |
|
||||
Total | 11185.75 | 168 |
The parameters of the multilevel logistic model were estimated using the behavior data after removing the responses by the nontrading responders. We used MCMCglmm package of R project to carry out the estimation [
Table
Estimated result for the multilevel logistic model.
Post mean |
|
|
|
|
---|---|---|---|---|
|
−0.75 | −1.70 | 0.14 | 0.09 |
|
−0.31 | −0.95 | 0.26 | 0.31 |
|
1.33 | 0.76 | 1.90 | <0.001 |
|
1.10 | 0.54 | 1.69 | 0.002 |
|
−0.25 | −0.40 | −0.11 | <0.001 |
|
||||
DIC | 1153.22 | |||
Number of responders | 149 | |||
Number of observations | 894 |
We modified the regression model by removing variable
We repeated the estimation using MCMCglmm. Table
Estimated result for the modified model.
Post mean |
|
|
|
|
---|---|---|---|---|
|
−0.73 | −1.71 | 0.08 | 0.084 |
|
1.27 | 0.71 | 1.79 | <0.001 |
|
1.01 | 0.35 | 1.56 | <0.001 |
|
−0.22 | −0.36 | −0.06 | <0.001 |
|
||||
DIC | 1161.22 | |||
Number of responders | 149 | |||
Number of observations | 894 |
The estimation result of the logistic model implies that if we install bidirectional escalators in a footbridge, the crosswalk requires a longer remaining time for pedestrian, that is, an additional 10.23 s, for it to be preferable (note that
The parameters presented by Table
Now, let us consider a street of width of 42 m as an example. For this case, the contribution of
In this study, the values of the parameters
In addition to the attributes of street-crossing facilities, the effect of personal attributes on the choice behavior was also investigated. It was determined that age exerted a negative effect on the utilization of footbridge; however, the gender exerted no significant impact on the selection decision. The age factor is also related to whether the responders are willing to make a trade-off decision between safety and convenience. The result revealed that a section of the elderly pedestrians did not make the trade-off choice decision. These responders consistently selected crosswalks; their choice decisions were independent of the attributes of the crossing facilities. Interestingly, these responders’ average age is significantly higher than that of the responders who were willing to make the trade-off decisions.
Data collection is a critical issue for choice behavior analysis. This study indicated that the synthetic video can assist researchers in collecting data effectively. The synthetic video provided an effective way to represent temporal-spatial attributes of street-crossing facilities. The responders can undergo realistic experience of the remaining time of pedestrian green light by watching the videos.
In future studies, it is recommended to include additional factors into the analysis. We will further consider the waiting time for traffic signal and the distance between the footbridge and the desired crossing location (see Zegeer, 2002) in future studies. Pedestrian behavior is also subject to weather conditions and traffic volume (e.g., Jensen, 1998; Houten et al., 1999); therefore, the corresponding attributes should be incorporated in the SP survey. Further behavior data on young and middle-aged individuals are to be collected in order to understand the heterogeneous behavior among various age groups.
The authors declare that they have no conflicts of interest.
This work is supported by the National Natural Science Foundation of China (no. 51408035 and no. 71621001) and Beijing Social Science Foundation (no. 16GLB013).