More specifically, for the travel path built environment attributes' impacts on travel mode selection, some scholars have found that distance has a great impact on transportation mode choice (17, 63–65), and the distance between home and school has a strong negative impact on the choice of walking to school (66). Otherwise, high traffic speeds (especially more than 30 km/h) and volume increase the risk of serious or fatal injuries to children and pedestrians, and may prevent parents from encouraging their children to walk to school (55, 67).The increase in the number of street areas and intersections also negatively affect children's independent walking to school (45).

For Chinese children in rural areas, parents often are concerned that their children will be involved in traffic accidents when walking in areas that lack facilities such as sidewalks. Scholars believe that the implementation of effective pedestrian interventions can reduce the traffic risks that hinder children from walking to school (68). For example, children have a higher proportion of walking to school in the built environment with high pavement integrity (43, 69).

Traffic lights and bicycle infrastructure are one of the main attributes that encourage cycling (70). Bicycle-specific facilities (SBFs), such as safety islands, raised curb pavement and other infrastructure separating motor vehicles and bicycles, can improve the safety of cycling, and such facilities play a good role in promoting cycling (68). Lin's observational study of adolescents found that high shade tree density and high sidewalk coverage encouraged children to walk to school independently (45), and Bosch's study found that the average density of convenience stores along the way was positively correlated with high positive traveling rate (71).

Therefore, on the basis of existing literature research and considering the actual situation of rural construction in Chengdu (in Southwest of China), this study selected eight built environment attributes, namely distance, number of intersections along the school path, sidewalk/bicycle lane, traffic speed, separation facilities between motor vehicles and non-motor vehicles, traffic lights and zebra crossings, green plants along the school path, and shops.

To design these eight attributes value levels, existing studies and the rural area conditions are combined to determine the tailor-made attribute levels for this research.

Several studies by scholars in other countries have shown that distance tends to be the primary concern of most parents. For most students, 1 mile is the maximum walking distance (72), and 2 miles is the maximum cycling distance (73). In addition, Kontou pointed out that walking was the most common way of school travel in urban and rural areas when the distance to school was < 0.5 mile, and the bicycle riding rate peaked when the home-school distance was between 0.5 and 1 mile (38).

Based on the previous research experience and data collection in nearly 100 rural areas in Sichuan Province (where the case area is located), the distance attribute in this study is divided into four levels: < 0.5, 0.5 to 1 km, 1 to 2.5 km, and more than 2.5 km. At the same time, we set the attribute level of the number of intersections passing through as: >5, 3–5, 1–2, and 0.

Considering the different development conditions in rural areas and the limited awareness of rural children on traffic facilities; it is impossible to accurately quantify the traffic speed of rural school paths, lawn, flower beds, street trees and other green plants and shops on the way to school. Therefore, the level of five attributes of green plants such as sidewalk/bicycle lane, separation facilities between motor vehicles and non-motor vehicles, traffic lights and zebra crossings, green plants on the way to school, and shop is set as Yes or No, so as to facilitate the children interviewed to judge and answer.

The traffic speed of the school path is set to two levels: high speed (≥30 km/h) and low speed (< 30 km/h). The attribute design level values are shown in Table 1.

There are many attributes that affect children's choice of school travel mode. Compared with judging directly a specific road segment, the statement choice experiment can control the covariance of the attribute level. Therefore, when other conditions are the same, the results reflect a more basic measure of children's preferences.

The application of stated choice experiments involves the creation of a design that combines the attribute levels in a particular manner. In this study, the two attributes with four levels and six attributes with two levels would result in 4² × 2⁶ different profiles in a full factorial experimental design that involves all possible combinations of attribute levels. In order to make the questionnaire concise and reduce the dependence between different variables, the experimental design method was used for scenario combination. Orthogonal design and uniform design are two commonly used experimental design methods. However, orthogonal design was applied in the questionnaire because it could be used for experiments with a small number of levels ( ≤ 5).

In this study, Statistical Analysis System (SAS) statistical software was used to orthogonal combine eight attributes and corresponding levels in the design, and an orthogonal fractional factorial containing 32 attribute profiles was selected. Choice sets were created by randomly combining these 32 attribute profiles, thereby creating choice sets of two unlabeled alternatives. Each option added the option of “None” to allow the possibility that both options are lower than a certain selection threshold. The 32 choice sets were divided into eight blocks to reduce the burden on the respondents. In the end, each interviewee needed to answer the questions in four scenarios in a questionnaire. In each scenario, there are four different ways of school travel. The modes of transportation are: walking, bicycle, electric bicycle/motorcycle, and private car. The interviewees were asked to choose their favorite route scenario under each mode of transportation. If the interviewees chose “none,” it means they did not like any of them. Table 2 shows an example of a selection set.

After the questionnaire work was completed, the unfilled questionnaires were first screened and valid survey data were entered to test their reliability and validity. In addition, the study also coded the socio-demographic variables and micro-attributes of the built environment for school travel as required by the SP questionnaire data analysis. For example, male gender is coded as 1, female gender is coded as−1. For micro-graphic built environment attributes, three indicator variables were constructed for each attribute with four levels. Each indicator variable corresponds to a category coded as 1, the remaining categories are coded as−1 for all three indicator variables. For each attribute with two levels, two indicator variables are constructed, and the category corresponding to the “yes” indicator variable is coded as 1, and the category corresponding to the “no” indicator variable is coded as -1.

On the premise that walking is selected as the school travel mode, the influence of distance attribute on the first and second levels is 1% significant at the general level, with positive values. This indicates that when children choose to walk to school, they are willing to walk within a distance of < 500 m and a distance between 500 m and 1 km, and the first rank coefficient is greater, indicating that the interviewed children prefer to walk to school within a distance of < 500 m. The influence of distance on the third level is significant and negative at the normal level, which indicates that children do not like walking for the distance between 1 and 2.5 km.

The influence of the number of intersections on the first and second levels is 1% significant at the normal level, with negative values, indicating that children who walk to/from school do not like the sections with the number of intersections between 3 and 5 or more. This is consistent with existing findings that neighborhood built environment characteristics (i.e., major street intersections, retail density, and neighborhood density) are strongly associated with the odds of walking (75–78).

The partial value utility of sidewalks suggests that children who walk prefer the presence of sidewalks, which is consistent with existing research (43). Lack of sidewalks or intermittent sidewalks are considered a barrier because they make walking to school more dangerous, and discontinuous sidewalks force children to cross the street repetitively, increasing the number of children crossing intersections and posing a potential hazard (30). The installation/widening of crosswalks and sidewalk improvements would result in a significant increase in the number of children walking or bicycling to school (71). However, in the actual rural school travel built environment, most rural areas only have sidewalks near school sections, and in some rural areas even all sections are country roads, which indicates the need for sidewalks in future planning.

The effect of the first level of the vehicle speed attribute has a negative significance at the 1% level at conventional levels, indicating that children walking to/from school do not prefer environments with vehicle speeds >30 km/h. This may be due to the fact that a school travel built environment with high vehicle speeds can compromise children's walking safety. Carlson et al. (79) combined the effect of major roads on the odds of walking to school and found that children and parents may seek a route with lower traffic speeds to walk to school. As shown in Table 5, when children walk to school, they prefer facilities with road teeth, fences, green strips, and delineations that separate them from motor vehicle lanes. For children who walk, the effect of having traffic lights and crosswalks was significant at the 5% level, indicating that children who walk to/from school prefer crossing facilities. Greening facilities on the path to and from school indicate that children on foot prefer greenery such as lawns, flower ponds, and street trees on the path. In addition, the study finds that children who walk to/from school prefer to have shops on the way to school or in front of the school, probably because children are younger and like to stay and play at the shops on their way to/from school or buy small items such as snacks, toys, and stationery.

To summarize, for walking to school, the attribute that children care most about is distance, they are most likely to walk to school within 500 m. As the distance increases, their chances of choosing to walk to school decrease. The installation of sidewalk and isolation facilities between motor vehicles and non-motor vehicles will bring a sense of security. The green plants such as lawn and street trees along the school path will provide shade, isolation protection, and increase the sense of visual beauty for children who walk to school, thus promoting children to choose to walk to school, while the high traffic speed will hinder children from choosing to walk.

The effects of the distance attribute at the first and second levels were positive 1% significant at conventional levels, given the choice of bicycle to/from school, indicating that when children choose bicycle to pass school, they prefer to ride a bicycle at distances < 500 m and at distances between 500 m and 1 km. Interestingly, the coefficient of the distance attribute is greater at the second level and the effect of the distance attribute at the third level is not significant at the conventional level, indicating that the interviewed children prefer to take the bicycle mode to pass school within the range of 500 m to 1 km. This is similar to the study by Kontou et al. (38) who found that the rate of bicycling peaked at home-school distances of between 0.5 and 1 mile.

The effect of the first of the number of intersections attribute is significant and negative at the conventional level, which indicates that children who cycle to/from school do not prefer road environments with more than five intersections. The effect at the third level is significant at the conventional level of 5%, indicating that children prefer to take the bicycle in a road environment with 1–2 intersections on the way.

Part of the value and utility of bicycle lanes show that children who travel to school prefer the road environment with bicycle lanes. The effect of the first level of the speed attribute had a negative significance of 1% at the conventional level, indicating that children do not prefer environments with speeds >30 km/h when choosing bicycles to get to/from school, which may be due to the fact that a built environment with too high a speed can affect children's safety through school.

The property of the isolation facilities between motor vehicles and non-motor vehicles shows that children prefer separated facilities from motorized lanes when bicycling to school, and that a built environment with separated facilities makes children feel safer.

For children who bicycle to/from school, the effect of having traffic lights and crosswalks was significant at the 1% level with a high significance coefficient, indicating that children who bicycle to/from school prefer cross-street facilities. The greenery on the path to school showed that children who bicycle to school prefer to have greenery such as grass, flower ponds, and street trees on the path to/from school. Children who bike to/from school prefer traveling paths with shops on their way or in front of the school. This is consistent with established research that higher average density of convenience stores is associated with higher odds of active traveling (71).

It is found that the maximum distance that rural children can accept is 1 km, and the distance they are most willing to ride a bike is between 500 m and 1 km. This may be because the distance of 500 m is short and it is not as convenient as walking to park a bike. During the visit, many children also said that they were not allowed to ride bicycles to/from school alone. Even though they had the ability to ride bicycles to/from school independently, parents do not allow children to ride bicycles to school independently for fear of children's safety. This is because that most of the rural roads have many intersections, with no special bicycle lane, and there is a lack of a safe bicycle environment.

The effect of the distance attribute at the first level was 1% significant at the conventional level and negative when children chose to ride the electric bicycle/motorcycle to/from school, indicating that when children chose to ride the electric bicycle/motorcycle to and from school, they did not prefer to ride the electric bicycle/motorcycle at distances of < 500 m. The effect of the distance attribute at the second and third levels was positively significant at the conventional level, indicating that children preferred to ride the electric bicycle/motorcycle at distances of 500 m or more. The effects of the first and second levels of the number of intersections attribute had a negative significance of 1% at the conventional level, indicating that children who ride to/from school by electric bicycle/motorcycle do not prefer road environments with more than three intersections; and the effect at the third level was not significant at the conventional level.

Some of the value utility of sidewalks/bike lanes suggests that children who go to school on electric bicycle/motorcycles prefer a road environment with bike lanes, possibly because children perceive that electric bicycle or motorcycles can be driven on bike lanes and that bike lanes provide them with safety and convenience.

The effect of the first level of the speed attribute has a negative significance of 1% at the conventional level, indicating that children do not like to choose electric bicycles or motorcycles to pass school in road environments where speeds are generally >30 km/h, and that driving environments with excessive speeds can be dangerous for children. The results show that when they go to school by electric bicycle or motorcycle, they liked to have such isolation facilities as curb, fence, green separation belt and scribing.

The school travel built environment with isolation facilities will enhance children's sense of security. The traffic light and zebra crossing attributes have a 1% level of significance at the first level of impact with a high significance coefficient, which indicates that children who ride to/from school on electric bicycles or motorcycles prefer to cross street facilities. The attributes of green facilities and shops on the path to school indicated that children who traveled by electric bicycle or motorcycle to school preferred the presence of shops and greenery such as lawns, flower ponds, and street trees on the path to/from school.

During the survey, the researchers found that most of the rural school-age children interviewed were transported to/from school by their parents on electric bicycles. In the survey on the preference of the built environment variables for children's school passages by electric bicycles, the highest significance was found for the range attribute of 1 to 2.5 km distance, and they also disliked the road environment with too high speed and too many intersections.

The effect of the distance attribute at the first and second levels is significant at 1% at conventional level and both are negative, given the choice of private car to and from school, indicating that when children choose private car to go to school, they do not prefer to travel by private car at distances < 500 m and distances between 500 m and 1 km, which may be due to children's consideration of factors such as time, cost, and convenience. The effect of distance attribute at the third level is positively significant at conventional level, when the distance to/from school is >1 km, they are willing to go to and from school willing to choose private car as the transportation. The effect of the first of the number of intersections attribute is significant at the conventional level of 1% and negative, indicating that children who travel to/from school by private car do not prefer road environments with more than five intersections and that too many intersections may prolong school travel time; the effects at the other two levels are not significant at the conventional level.

There is no significant difference in the attributes of sidewalks or bicycle lanes, speed, isolation facilities between motor vehicles and non-motor vehicles, and shops, which indicates that these attributes of the built environment for school travel have no impact on children taking private cars. The traffic light and crosswalk attributes have a 1% level of significance at the first level of impact with a high significance coefficient, which indicates that children who travel to/from school in private vehicles prefer to cross street facilities. The attributes of green facilities on the path to school indicate that children who travel by electric bicycle or motorcycle to school prefer to have greenery such as grass, flower ponds, and street trees on the path to/from school.

For children who go to/from school by private car, they prefer a built environment with fewer intersections, traffic lights with crosswalks and greenery, in addition to distance.

The four multinomial logit models analyzing the preferences of the school travel built environment for the four travel modes showed two constant terms. Only the constant term of the choice experimental model in the walking condition was negatively significant, while the constant terms of the remaining three choice experimental models were positively significant. This suggests that in addition to the attributes and levels set by this study, there are other attributes of the school travel built environment that can affect walking or bicycle, electric bicycle, private vehicle, etc.