Introduction

Walking and cycling are modes of transportation that could potentially replace short to medium distance trips previously made in motor vehicles. These modes of transportation can contribute to achieve the recommended levels of physical activity, reduce sedentary behaviors, improve air quality, and reduce greenhouse emissions, thus providing direct and indirect economic benefits.1–3 Due to their health benefits, low cost, and potential to reach a great portion of the population, walking and cycling have inspired research to better understand correlates of active transportation.4–7

Although evidence has shown that personal factors contribute to active transportation,4 built environment characteristics can also play an important role in this behavior.8 For example, people who reside in areas with higher population density, land use diversity, higher connectivity of streets and trails, access to places for walking, attractive environments, and higher access to public transit have a greater chance of walking or biking.9–12

Despite the relative consistency of those results, how the environment is associated with physical activity varies according to local and cultural characteristics. However, nearly all studies on factors associated with transport physical activity have been conducted in high-income countries.9,12 Therefore, to better understand the relationship between the built environment and physical activity patterns, specific socioeconomic (SES) and cultural characteristics should be considered.

There is growing research from low-income and middle-income countries in Latin America about the correlates of physical activity for transportation. In Bogota, Colombia, higher street density, connectivity, and access to public transportation have been found to be associated with walking for transportation.13 In addition, bicycle use was associated with higher density of streets and lower rates of traffic accidents.13 Among older adults from the same city, higher street connectivity and steep terrain were negatively associated with walking for transportation.14

The evidence from Brazil shows that perceived attractiveness, safety, and proximity to destinations are associated with walking as a means of transportation among adults15 and older adults.16,17 These studies, however, used self-perceived measures, which have shown a divergent association with transport physical activity when compared with objective measures of the environment.18 In addition, objective measures of the environment are scarce in Brazil, resulting in limited evidence.19,20 Thus, the goal of this study was to assess the association between objectively measured features of the built environment and walking and cycling as forms of transportation in the city of Curitiba, Brazil.

Methods

Sample Population

A geographically diverse and representative sample of adults from Curitiba, Brazil was interviewed by phone in the year 2008. Curitiba is the capital city of Parana state, located in the south of Brazil. With a population of 1,851,215 inhabitants and 435 km2 of area, Curitiba is the eighth biggest city in the country. Curitiba is classified as a city of high human development (Human Development Index = 0.85) and has also been recognized worldwide by its efforts to promote sustainable growth and mass public transportation.15,21

The detailed sampling procedures are documented elsewhere.19 Briefly, participants were asked for their address to spatially locate their homes. From the total of 2,097 respondents surveyed, 1,206 provided complete address information. Institutional review board approval was obtained prior to data collection from Pontiff Catholic University of Parana in Curitiba and Washington University in St. Louis. All participants of this study provided verbal consent before the phone interview began.

Outcome Variables: Physical Activity for Transportation

For the assessment of physical activity, the International Physical Activity Questionnaire, Long Form (IPAQ)22 was used. In Brazil, IPAQ is widely used19,23,24 and has shown good reliability. For this study, only questions related to walking and bicycling for transportation were used. Time spent walking as a means of transport (in minutes per week) was classified according to two different criteria. Previous studies have shown that the association of built environment characteristics varies according to physical activity levels.25,26 Hence, the subjects were classified as any walking (≥10 min/week) and walking at recommended levels (≥150 min/week).27 Cycling for transportation (in minutes per week) was categorized as at least once a week for ten consecutive minutes versus no use due to its low prevalence.

Covariates: Built Environment Variables

Built environment variables were obtained from the Institute of Urban Planning and Research of Curitiba. To calculate the density indicators, a radius of 500 m around each individual's residency was used. Although other studies have analyzed greater buffers sizes (800 m,28 1,000 m,29 and 1,600 m30) when the outcome is bicycle use, the environmental characteristics of bigger buffers (1,000 m) have been highly correlated with those observed in smaller buffers (500 m).31 Furthermore, in denser cities like Curitiba, the increase of buffer size could result in reduced variability of the environmental characteristics at the individual level as well as overlap of urban areas.30 Thus, a 500-m-radius buffer was used for bicycle use.

For the accessibility measures, the shortest distance to the selected built environment features (e.g., bus stop, bike path) was used. Both buffer and shortest distance were determined, considering the street network in order to acquire a more precise measure.32 Furthermore, network buffers show more consistent associations with environmental characteristics when compared with circular (crow-fly) buffers.33

Population density was computed as the number of inhabitants per square kilometer based on data from the 2000 national census. As the number of inhabitants was available only at the census tract level, the intersect tool of ArcGIS was used to select census tracts that were wholly or partially contained within the 500-m network buffer. If a census tract was only partly contained within the buffer, inhabitants' unit counts were prorated according to the percentage of the census block area contained within the buffer. Finally, population density was computed as the number of inhabitants divided by the number of kilometers within the 500-m buffer.

Area income was available at the census tract level using data from the 2000 national census. The variable was estimated as the average mean income (in Brazilian Reais) of the census tract sectors contained within a 500-m network buffer area.

Density of public transport was estimated by separately considering the number of “conventional” bus stops and the number of Bus Rapid Transit (BRT) tube station stops within a 500-m network buffer from the participant's home. Accessibility was calculated using the shortest distance from these features to the participant's home, considering street network.

The presence of traffic lights was used as an indicator of traffic safety (e.g., traffic calming device). The number of traffic lights contained within the 500-m network buffer area was computed for each participant household.

To calculate land use mix, we used land use patterns based on 32 types of parcels classified into 5 categories (residential, commercial, industrial, recreational, and empty or other). The proportion of commercial or residential areas and the associated entropy were calculated based on available methodology,34 considering the 500-m network buffer surrounding each residence.

Considering street patterns, for each individual 500-m network buffer, the following variables were computed: (a) streets density (in meters per square meter); (b) average length of the streets; (c) number of blocks; (d) proportion of dead-end streets; and (e) proportion of street intersections (≥4 segments). The density of bike paths (in meters per square meter) was estimated within a 500-m network buffer, and the accessibility was computed based on the shortest distance to a bike path (in meters).

The terrain slope was calculated using the city's contours, which were converted into an irregular triangular network.35 Average slope of the triangles within the 500-m buffer was calculated in percentage values,14 with 0° indicating a 0 % slope (lowest) and 45° indicating a 100 % slope.

All the variables were classified into tertiles, except for the number of bus stops, traffic lights, and bike lane density, which were dichotomized. The variables were categorized due to no linearity on the logit (Table 1).

Table 1 Descriptive statistics of built environment variables (Curitiba, Brazil; n = 1,206)
Full size table

Sociodemographic and Health Status Variables

Sociodemographic variables assessed in the study were sex, age (18–34, 35–54, and ≥55 years), education (did not complete high school, completed high school, some college), marital status (single, married, and other, which included widowed/separated/divorced), car ownership (yes versus no), and body mass index (BMI) (underweight/normal weight versus overweight/obese) based on self-reported height and weight (Table 2).

Table 2 Sociodemographic characteristics by physical activity outcomes (Curitiba, Brazil; n = 1,206)
Full size table

Statistical Analysis

Descriptive statistics were calculated. Colinearity among the built environment variables was tested using Pearson's correlation coefficient, considering r ≥ 0.5 and p < 0.05 as criteria for exclusion from the multivariate model. Binary logistic regressions were used to analyze the association between physical activity outcomes (any walking, walking at recommended levels, and bicycling) and built environment variables. The final multivariate models included environmental variables with p ≤ 0.15 entered together, adjusting by sociodemographic variables (sex, age, BMI, educational, marital status, and car ownership). All analyses were performed in Stata 9.2 using sample weights for gender and age and considering the design effect.

Results

The final sample was composed mostly of women and middle-aged individuals (35–54 years old). The sample was equally distributed with regard to education; a large proportion was married and had at least one car at home. Almost half of the participants in the study were classified as overweight or obese (Table 2). Only a fraction of the analytical sample size was classified as in any walking and cycling for transportation (0.05 %) or walking at recommended levels and cycling (0.03 %) (data not shown).

More than half of the participants walked at least 10 min/week, and approximately one quarter was classified as engaged in recommended levels of walking (≥150 min/week) as a form of transport. The proportion of people who were engaged in any walk was lower among males, older adults, those reporting having a car in the household, and among those classified as overweight/obese. The prevalence of walking for transport at recommended levels was lower among individuals of higher education, those reporting having a car in the household, and those who were married or widowed/separated/divorced (Table 2).

After considering confounding variables, the number of BRT tube stations (OR = 1.43, 95 % CI = 1.13–1.79) and residential (OR = 1.25, 95 % CI = 1.02–1.53) and commercial (OR = 1.47, 95 % CI = 1.13–1.91) area proportion were positively associated with any walking for transportation (Table 3). Participants living in areas with higher income showed 44 % lower likelihood to any walk (p < 0.01) when compared with residents of lower-income areas. In the multivariate model (Table 4), after adjusting for confounding variables, only greater accessibility to a bike path was associated with less walking. People residing up to 367 m from a bike lane had 20 % lower odds of walking at recommended levels (95 % CI = 0.64–1.00).

Table 3 Crude and adjusted logistic regression model showing association between objectively measured features of the built environment and walking for transportation (Curitiba, Brazil; n = 1,206)
Full size table
Table 4 Crude and adjusted logistic regression model showing association between objectively measured features of the built environment and walking for transportation in recommended levels (Curitiba, Brazil; n = 1,206)
Full size table

Only one in ten participants used bicycles as a means of transportation and was higher among males, young adults (18–34 years old), lower education levels, living alone, normal weight status, and not owning a car (Table 2). After adjusting for confounders (Table 5), residing in areas of higher income (OR = 0.26, 95 % CI = 0.08–0.81), higher number of traffic lights (OR = 0.27, 95 % CI = 0.09–0.88), mixed land use (OR = 0.52, 95 % CI = 0.31–0.88), and residential area proportion (OR = 0.53, 95 % CI = 0.34–0.83) were associated with bicycling for transportation.

Table 5 Crude and adjusted logistic regression model showing association between objectively measured features of the built environment and cycling for transportation (Curitiba, Brazil; n = 1,206)
Full size table

Discussion

Overall, individual characteristics were more consistently associated with active commuting than were environmental variables. A greater number of individual and environmental variables were associated with walking and cycling at least 10 min/week when comparing with walking 150 min/week or more. Walking for at least 10 min/week was associated with sex, age, education, car ownership, BMI, neighborhood income, number of BRT stations, and residential and commercial area proportions. However, higher volume of total walking (≥150 min/week) was associated only with education, marital status, car ownership, and distance to nearest bike path. Finally, cycling for transportation was associated with sex, age, marital status, car ownership, neighborhood income, traffic lights, and land use.

Our results indicate that environmental variables associated with walking vary according to the amount of walking, as observed in other studies.36,37 Data from this study suggest that variables such as the proportion of commercial and residential areas are associated with any walking for transportation, but not at the recommended levels. A recent study showed that adults living in highly walkable areas of Curitiba (as measured by an index composed by residential density, density of intersections, and land use mix) are more likely to walk as a mean of transport at recommended levels.38 These data indicate that the combination of different environmental characteristics may be needed to encourage greater levels of physical activity.39,40 However, the comparison between the studies needs to take into account that some different methods were applied in both studies. In the present paper, the use of individual buffer increases the variability between the environmental variables. On the other hand, the work from Reis et al.38 considered characteristics in the census tract level that could reduce the variability among the environmental variables. However, the selection of census tract was done to increase the contrast between high and low walkability and, consequently, could favor finding differences on the walking level.

Our results also showed that the availability of public transport was associated with walking for transportation. Studies from cities such as Bogota (Colombia), New Jersey (USA), and Perth (Australia) support the hypothesis that people who use and/or have access to public transport are more likely to walk and be more physically active than those who do not.25,41,42 As reported elsewhere,25,26 the use or access to public transport was associated with some walking, but not with the recommended physical activity levels. Additionally, this association was specific to certain types of transport system (e.g., BRT), as reported in other studies.25,41,43 A potential explanation is that the BRT is faster than the regular bus system. This could motivate the people to expend more time walking until the BRT tube station stop to save more time to reach the destination using the transit system. Also, these types of transportation are more disperse and available as compared to the regular bus system. On average, in Curitiba, the nearest bus stop is located <200 m from a destination, which would require only 3 min of brisk walking. On the contrary, BRT stations are located at 1,850 m on average, hence helping explain the associations we found. The finding of BRT is very relevant given that more than 100 cities worldwide have implemented this system that was first implemented in Curitiba.44

Regarding built environment features, a higher density of bike paths was negatively associated with walking for transportation at the recommended levels. In Curitiba, such structure connects city parks and is used for leisure purposes rather than daily commuting. Therefore, participants residing near bike paths are likely to live closer to parks, which are typically large areas that may hinder accessibility to nearby destinations. Supporting this assumption, a study by Mass et al.45 identified that local residents living in neighborhoods with more green areas walk and bike less frequently due to the great dispersion of destinations when compared with commercial areas. A study from Bogota, Colombia also found that residents who lived nearby urban parks were less likely to participate in the recreational program that occurs on Sundays and holidays called the Ciclovia program.13

This study showed that bicycling for transportation is associated with a greater number of built environment characteristics. In Curitiba, areas with higher number of traffic lights, land use mix, and residential areas and lower density of bike paths were generally found in high-income areas and in the city's downtown, where people own cars and are less likely to use bicycles. Areas with these characteristics also have higher levels of motorized vehicle traffic, also associated with less bicycle use.46,47 These factors could be associated with more traffic accidents within high SES neighborhoods which in turn have been associated with less biking.13 These findings, despite being opposite from those in the literature, may be reflections of existing social and cultural differences between countries. As mentioned, bike lanes in Curitiba connect parks and are more often used for leisure. As such, cyclists rely on the use of streets where they share the same space with cars, motorcycles, and buses, creating unsafe environments and adding tension to the constant conflict between active commuters and drivers. Brazil is known to have one of the most hostile traffic environments of the world.48 Between 2000 and 2008, road traffic mortality per 100,000 population was between 18 (northeast region) and 30 (center west region) per year,49 higher than countries such as Japan (5 out of 100,000), USA (13 out of 100,000), and Canada (9 out of 100,000).50 In 2006, 22 % of traffic-related accidents in Brazil involved cyclists.51 The proportion of deaths involving cyclists in Brazil is about 5 %, also higher compared to the USA (2 %) and Canada (3 %).50

However, the present study did not consider some important environmental variables that often are not available in Geographic Information System (GIS) databases (e.g., availability and quality of sidewalks, garbage accumulation) or cannot be evaluated through objective measures (e.g., safety perception) and that might be important to better understand some cultural factors related to our results. For example, studies in Brazil have found that perceived safety52 and esthetics53 were negatively related with walking for transportation. Similar results were also observed in low-income countries.54,55 In this sense, future studies in culturally and economically diverse regions should consider a more comprehensive framework including objective and perceived measures of the built environment, since the evidence emerging from high-income countries might not reflect the contextual characteristics from other regions.

This study is one of the first in Brazil and Latin America to investigate the relationship between objective measures of the built environment and active commuting. To this date, the studies in Brazil have only analyzed perceived environments,15–17 and studies using objective measures of the built environment have focused on leisure time physical activity20 or have not included environmental measures at the individual level.38 The study included a sample population of the city that was geographically well distributed and with good variability in built environments. The conclusions of this study, however, should take into consideration some limitations. First, it was not possible to determine a causal relationship between the associated factors. The measure used to assess the practice of physical activity has limitations inherent to self-reported data such as comprehension and recall bias. This problem can be further complicated with measures of sporadic activities, such as physical activity. In addition, IPAQ only captures bouts of at least 10 min, which is usually more than the usual commuting to and from public transport. This might have undermined our ability to detect associations between walking and the availability and accessibility of bus stops. Even though objective measures of the built environment were used, data were based on GIS, which were not originally gathered for physical activity research purposes.

Conclusion

This study shows that features of the built environment were associated with walking or bicycling for transportation among Curitiba residents. Several observed relationships (e.g., number of BRT stations and walking) show face validity and are leads for future studies. In countries of low and middle income, the built environment may affect active commuting despite the individual characteristics of commuters. Findings from this and other recent studies reflect the contextual factors that might be needed to be included in further research such as access to public transportation,13 number of plazas and public facilities,20 and crime-related and traffic-related safety.52,54