Objective environmental exposures correlate diﬀerently with recreational and transportation walking: A cross-sectional national study in the Netherlands

Background: Walking is a good and simple to increase people’s energy expenditure, but there is limited evidence whether the neighborhood environment correlates diﬀerently with recreational and transportation walking. Aim: To investigate how recreational walking and transportation walking are associated with the natural and built environmental characteristics of the living environment in the Netherlands, and examine the diﬀerences in their associations between weekdays and weekends. Method and data: We extracted the total duration of daily walking (in minutes per person) for recreation and transport from the Dutch National Travel Survey 2015-2017 (N=66,880), and analyzed it as an outcome variable. Objective measures of the natural (i.e., Normalized Diﬀerence Vegetation Index (NDVI) and meteorological conditions) and built environment (i.e., crossing density, land use mix, and residential building density) around respondents’ home addresses were determined for buﬀers with 300, 600, and 1,000 m radii using a geographic information system. To assess associations between recreational and transportation walking and the environmental exposures separately, we ﬁtted Tobit regression models to the walking data, adjusted for multiple confounders. Results: On weekdays, people living in areas with less NDVI, higher land use mix, higher residential building density, and higher crossing density, were more likely to engage in transportation walking. While recreational walking was negatively associated with NDVI, crossing density, precipitation and daily average temperature, it was positively associated with residential building density. At weekends, land use mix supports both recreational and transportation walking. A negative association appeared for NDVI and transportation walking. Daily average rainfall and temperature were inversely correlated with recreational walking. Sensitivity tests indicated that some associations depend on the buﬀer size. Conclusions: Our ﬁndings suggest that the built and natural environments have diﬀerent impacts on people’s recreational and transportation walking. We also found diﬀerences in the walking-environment associations between weekdays and weekends. Place-based policies to design walking-friendly neighborhoods may have diﬀerent implications for diﬀerent types of walking.


Introduction
Sedentary behavior and physical inactivity are well-known risk factors for developing chronic disease, threatening people's physical and mental health alike [1]. Among the different ways to be physically active, walking is the most popular and results in a range of health benefits (e.g., lower risk of type II diabetes and cardiovascular disease) [2,3]. In the Netherlands, the daily average distance walked by Dutch people is around 800 m [4], although it varies significantly across municipalities, which can to a certain extent be attributed to differences in the environment.
Evidence is mounting that walking in the residential surroundings is influenced not only by people's demographic and socioeconomic characteristics (e.g., age, gender, and income), but also by the natural and built environments (green space, water bodies, air quality, etc.) [5][6][7]. Previous research showed that the availability of green space and water bodies is positively associated with people's walking behavior [8,9]. Built environment factors include land use diversity and urban design features, such as street connectivity, building density, and accessibility. While a few studies found positive associations between walking and land-use diversity, design (street density and intersections), and access to recreational and commercial places [10][11][12], other environmental characteristics (e.g., traffic noise and risk) were reported to be inversely correlated with walking [13,14].
However, these associations between environmental exposures and walking were not consistently found across studies. For example, while [15], [16], and [17] found that land use mix was positively associated with walking, [12] found a negative association. Results were also inconsistent for crossing density [17,18]. These inconsistencies can to a certain extent be attributed to, first, the difference between using perceived and objective measures [5]. While perceived measures were mainly derived from surveys or interviews, which were possibly affected by a recall bias, objective measures were obtained based on geographical information system (GIS) technologies.
Second, it is common to use administrative areas (e.g., census tracts) to delimit people's neighborhood environments. This approach, however, has been conceptually and methodologically criticized [19].
Refined exposure assessments are achieved through GIS-based buffers centered on people's residential locations [20,21] To what extent are potential walking-environment associations sensitive to the buffer size used for delimiting the residential context?

Study design and study population
The study was based on the cross-sectional Dutch National Travel Survey [31]. Each year around 40,000 respondents report their travel behaviors over one 24-hour period by means of a travel diary.
Respondents are assigned to a specific day in order to include seasonal effects. To maximize the sample size, we pooled data from three consecutive years, namely 2015, 2016, and 2017. In total, our nationally representative sample comprised 66,880 respondents.
Ethical approval was not required because only secondary data were analyzed.

Walking behavior
Walking trips were classified as either recreational walking (e.g., for leisure) or transportation walking (e.g., commuting to and from work), depending on the reported trip purpose. For both types of walking, the dependent variable was defined as the total duration of daily walking trips in minutes per person (within a 1000m buffer). Note that trips outside the residential environments (>1,000 m) were excluded.

Natural and built environmental exposures
We included four measures of the residential natural and built environment, following previous studies [8,32]. Based on their residential addresses, respondents' were allocated to their 6-digit postal code (PC6) rather than larger administrative areas as done previously by [32]. Netherlands [37]. Following previous studies [38,39], we grouped a total of 37 land use types into five categories, namely: residential, recreational, commercial, industrial, and others.
The entropy index was computed based on the proportion of the area of each land use category.
Index values range between 0 and 1. A higher value means greater diversity.
3 Street connectivity is related to the design of the street layout and indicates the access to other places [33]. We considered two aspects of street connectivity: >=4-way crossings and cul-de-sacs.
More >=4-way crossings are indicative of a higher street connectivity, which promotes walking [5], while the opposite applies to

Descriptive statistics
Of the 66,880 respondents, 52% were female, 84% were Dutch, 23% were retired (aged > 65 years), and 59% had a household income of 20.000-40.000 euros ( Table   1). The majority (87%) held a driving license, 52% owned a car. 21% had an e-bike, and 8% had a moped.     We also observed that varying the buffer sizes affected the association between the residential environment and walking. For example, a positive association between cul-de-sacs and transportation walking during weekdays was found for the buffer size of 600m, but was absent in the 300m and 1,000m buffers. There was also evidence that land-use mix based on 300m buffers was positively associated with transportation walking at weekends (p < 0.001), but the association was less significant for buffers with 600m (p < 0.01) and 1,000m (p < 0.05) radii.
Similarly, the significance levels of the association of land use mix with recreational walking at weekends varied across different buffer sizes.

Main findings
This study examined the differences in the correlations between environmental exposures and different types of walking. Consistent with earlier research [16,17,38], land-use mix was positively associated with transportation walking. For recreational walking, however, the relation differed between weekdays and weekends. At weekends, residential environments with a high mix of land use encouraged more recreational walking, but on weekdays no association was found. This may be because environments with a high land-use diversity attract fewer recreational walkers on weekdays than at weekends. NDVI was negatively associated with recreational and transportation walking, which is inconsistent with previous studies [8,9,21]. A possible explanation is that less utilitarian destinations (e.g., shopping centers, stores, supermarkets) are located in areas with high levels of vegetation, which results in less transportation walking. Another reason may be that pronounced vegetation (e.g., parks) could reduce perceived safety.
Especially at night, green spaces can be used for criminal activities [42,43], and people prefer not to walk in insecure areas [13,44]. Residential building density, which is correlated with population density [24], was positively associated with recreational and transportation walking, as also found by [11,45] and [46]. However, its association with transportation walking was insignificant at weekends, possibly because many business and commercial facilities are closed at weekends, and therefore fewer people walk for transportation. Congruent with earlier work [17], we found that street connectivity supported transportation walking. However, cul-de-sacs seem to be barriers to recreational walking on weekdays. An explanation could be that most cul-de-sacs are located in suburbs where there are fewer opportunities for recreational walkers, who seek different scenic environments.
Although the built and natural environments were largely and consistently associated with walking across different buffer sizes, we observed that for some environmental factors, the associations varied with different geographical scales. This aligns with work conducted by [47]. For example, while a negative relationship was found between cul-de-sac and recreational walking at weekends within a 1,000m buffer, no associations were observed at smaller scales (i.e., 300 m and 600 m). This confirms the results reported by [23] on other urban form variables (e.g., sidewalks) and regional accessibility [53] were not available.
However, in the Dutch context, most streets are walkable and we think the implications for our model estimates are minor.

Conclusions
In this study we examined differences in associations of people's residential environments with recreation and transportation walking on weekdays and at weekends, using a large sample with national coverage. We provided robust evidence that the associations between environmental correlates and walking differ by weekdays and weekends. Sensitivity assessments across buffer sizes confirmed that our results are reasonably robust. If future studies confirm our findings in a longitudinal setting, our results suggest that it may be more efficient for urban planners and policy makers to take into account walking type and time when developing strategies to promote walking.

Ethics approval and consent to participate
Not applicable

Consent for publication
Statistics Netherlands (CBS) provided consent for publication of the results presented in this paper.

Availability of data and material
Supplementary data and material to this article can be found online at https://github.com/vitality-data-center

Funding
This work was part of the Vitality Data Center project, which in turn is part of the Vitality Alliance (funded by Utrecht University and Eindhoven University of Technology), and was supported by the Open Data Infrastructure for Social Science and Economic Innovations (ODISSEI) in the Netherlands.

Competing interests
The authors declare that they have no competing interests.

Author's contributions
DE developed the research idea. ZW prepared the data together with MH.
ZW carried out the analyses and drafted the manuscript. All authors read, edited, and approved the final manuscript.