3.1. General Situations of Active Travel in Nanjing
Basic descriptive statistics and frequencies were used to describe the sample (
Table 2 and
Table 3). Different from Western countries, especially in the U.S., where private car use makes up more than 80% of all out-of-home trips [
26], active transport is still the dominant transport mode in Nanjing, which accounts for about 60% of total trips. Two interrelated reasons might explain this. First, car ownership rates in China, though fast growing, are still lower than the rates in most Western countries. According to the World Bank, in 2011, car ownership was 83 per thousand in China and 809 per thousand in the U.S. In 2008, car ownership was 544 per thousand in Germany and 37 per thousand in China [
7]. Second, the differences in the built environment between Western countries and China might also be a reason. Muller’s (1995) four-stage model of intra-metropolitan transport and urban growth assumes that the dominant transport system in a certain era shapes the spatial structure of the city in the U.S [
27]. In a similar vein, the land use pattern in most Chinese cities is developed and has accommodated non-motorized transport modes. The land use pattern, characterized with high density and land use mix, offers residents more opportunities to travel short distances than their Western counterparts who live in built environments shaped by fast transport modes. Therefore, walking and cycling are more prevalent in Chinese cities.
There are also considerable differences in the transport mode use for different purposes. For subsistence activities, the use of active transport is the lowest, and private car usage is the highest, while for maintenance activities, active transport makes up more than 70% of total trips. The difference actually manifests the importance and temporal-spatial rigidity of various activities and the consequential strategy people utilize to access these spatial opportunities. Given the importance of subsistence activities, people generally search for employment in the entire city, and therefore, the travel distance for employment is longer than for maintenance activities, which usually can be found locally. The differences in the temporal-spatial fixity of different activities are also important reasons for the variance in transport mode usage. Obviously, the differences in transport mode use necessitate separate analyses of the influences of the built environment on mode choices for different travel purposes.
3.2. Binary Logistic Regression Models for Subsistence, Maintenance and Discretionary Travel
Binary logistic regression analyses were conducted to examine the effect of the built environment on active travel choices for subsistence, maintenance and discretionary travel purposes, respectively (
Table 4,
Table 5 and
Table 6). Since the whole data are hierarchical, such that the survey data and demographic attributes are at the individual level while the data of built environments are at a higher level (TAZ), multilevel regression modelling might be more suitable for the analysis. However, due to the limited number of TAZs surveyed in the data, it cannot meet the model requirement for the least numbers of units of a high level (as a rule of thumb: there should be at least 30). We still use binary logistic regression, but keep in mind that the results are to some extent biased. Whether or not to use active travel (
i.e., public transport and private cars are converged together as the not taking active travel category) were used as dependent variables. Given that the idea of the study is to examine which built environment attributes have more profound influences on walking and cycling, three models are explored: a model with only socio-demographics (Model A), a model with variables of the neighborhood form (Model B) and a complete model with all variables (Model C). The pseudo R-square statistic, Nagelkerke’s ρ
2, indicates the percentage of the variance of the dependent variable that can be explained by the independent variables [
7] and “Sig.” in the table refers to significance.
It is worth noting that we checked the multicollinearity properties of the explanatory variables to avoid spurious and erroneous modeling. For that purpose, the variance inflation factor (VIF) was used. All of the selected variables had VIF values between 1.0 and 2.0, except for the “percentage of bike lanes”, for which VIFs ranged between 2.37 and 2.58, depending on the model specification. A commonly-accepted cutoff point signaling multicollinearity problems in logistic regression is 2.5 [
27], so we decided not to exclude any selected variables in the modeling to maintain consistency across the models.
As shown in
Table 4,
Table 5 and
Table 6, Nagelkerke’s ρ
2 of models for subsistence travel is 0.115 (Model A), 0.139 (Model B) and 0.209 (Model C). Nagelkerke’s ρ
2 of models for maintenance travel is 0.079 (Model A), 0.134 (Model B) and 0.207 (Model C). Nagelkerke’s ρ
2 of models for discretionary travel is 0.075 (Model A), 0.129 (Model B) and 0.173 (Model C). Apparently, the overall model fit is lowest for discretionary trips. This might be because leisure trips are more determined by other personal preferences and values, which are not included in the model [
28].
However, when calculating the contribution of the built environment for different travel purposes, it is found that for subsistence travel, the contribution of the built environment is the lowest (the contribution of the built environment for subsistence activities is (0.209 − 0.115)/0.209 = 45%; the contribution for maintenance activities is (0.207 − 0.079)/0.207 = 62%; the contribution of the built environment for discretionary activities is (0.173 − 0.075)/0.173 = 0.57%). People usually choose employment opportunities from all over the metropolitan area, while when people make shopping or leisure trips, they are more inclined to use local facilities. Since all of our built environment variables are residential area-based, the influence of these local built environment indicators are therefore more pronounced for maintenance and discretionary activities and less influential in determining the mode choice for subsistence activities.
Nagelkerke’s ρ2 also indicates that the influence of the neighborhood form is not as strong as the street form on active travel choices. For subsistence activities, Nagelkerke’s ρ2 increases by 0.024 when neighborhood form variables are included and increases by 0.070 when street form variables are taken into account, suggesting that street form variables can explain more variance of the dependent variable than the neighborhood form. A similar situation can be found in the models of maintenance activities and discretionary activities.
Regarding specific variables, very limited significant impacts of the neighborhood form are observed while almost all of the influences of the street form are statistically significant. Population density shows significant influences on the active travel choice in Model B of subsistence and discretionary activities. In line with previous studies [
29], the higher the density is, the more likely that people will choose active travel. However, when street form variables are controlled, the impacts become insignificant. Surprisingly, the land use mixture shows no significant influences in all models, which is in contrast with the findings from Western countries. The design dimension of the neighborhood form—distance to the nearest green/open space—demonstrated profound influence on active travel. In the six models, it exerts significant influences, even when the characteristics of street form are controlled. Given that open/green spaces, especially medium- and small-sized parks, are far less prevalent in most Chinese cities than in Western countries [
15], good accessibility to open/green spaces could considerably encourage the use of active transport.
Almost all of the characteristics of the street form show significant influences on active transport. The density of street crossings is positively related to active travel. The areas with a high density of street crossings are generally the areas constructed at a walkable human scale [
30]. In addition, more street crossings mean more blocks, which makes driving cars inconvenient. The diversity factor of the street form—the percentage of roads with separate bike lanes and the percentage of roads with sidewalks—also exerts a significant impact on active travel choice. Similar to the findings from Western cities, the more numerous the separate bike lanes and sidewalks, the higher propensity for the residents to use active transport [
3]. In addition to the direct effects that the infrastructure has on making active travel more convenient and fluid, the indirect effects that they have on making travelers feel safe cannot be ignored. The findings highlight the importance of the pedestrian/cycling infrastructure in promoting active transport. Regarding the influence of the design aspects of the street form, the analysis results of all three models demonstrate consistently negative relationships between illegal parking on the street and the use of active travel, suggesting that having to traverse walking/cycling environments without major barriers is a crucial factor for promoting active travel.
Unlike the findings in Western cities [
2,
5], building setbacks seem positively related to active travel in our analysis, except for subsistence travel. Chinese cities are characterized by high density with intensely constructed buildings, so that the setbacks are usually much smaller than in Western cities. The buildings in China make the streets crowded and not as easy to walk through. Large building setbacks make more spaces for walking and cycling and, therefore, are positively related active travel. However, too large a building setback, generally related to less diversified and tedious streetscapes, may have negative influences on walking/cycling, as confirmed from Western evidence. In other words, the relationship between building setbacks and active travel may be quadratic rather than linear. The different effect of building setbacks on active travel could also result from potential confounding factors, such as the average width of sidewalks/bike lanes and streets and the volume of traffic, which are not included in this analysis.
In addition to built environment factors, individual socio-demographics also exert notable influences on active travel. Compared to men, women are more likely to use active transportation for subsistence travel. The limited car access of women and short commuting distance might account for the differences. However, for discretionary travel, the influence is insignificant. One possible reason might be that women and men tend to have more joint recreational activities, and therefore, the differences in mode choice are minor. Given the special divisions of household tasks in Chinese households caused by special cultural norms and ideology, which will influence the related travel-activity patterns between male and female heads, the gendered differentials in active travel deserve further research. Generally, age is positively related to active transport, though some of the effects are not significant. People with high educational levels seem more inclined to use motorized transport than active transport, especially for subsistence travel. Highly-educated people tend to have more professional jobs and special requirements for goods and recreational facilities that cannot be found readily in the proximity of home. The influence of income is rather moderate, which is probably caused by the relationship with the educational level (a high education level is related to high income level). Household size only has significant effects on active travel choice for subsistence activities when street form variables are not controlled. The presence of a pre-educated child tends to reduce the propensity to use active travel for recreational activities. For the sake of convenience and safety, this result makes intuitive sense. Household car ownership, as expected, is highly negatively related to active travel for all travel purposes.