Ding et al. International Journal of Behavioral Nutrition and Physical Activity 2013, 10:57 http://www.ijbnpa.org/content/10/1/57 RESEARCH Open Access Perceived neighborhood environment and physical activity in 11 countries: Do associations differ by country? Ding Ding1,2,3*, Marc A Adams1,4, James F Sallis1, Gregory J Norman1, Melbourn F Hovell2, Christina D Chambers1, C Richard Hofstetter2, Heather R Bowles5, Maria Hagströmer6,7, Cora L Craig8, Luis Fernando Gomez9, Ilse De Bourdeaudhuij10, Duncan J Macfarlane11, Barbara E Ainsworth12, Patrick Bergman13, Fiona C Bull14, Harriette Carr15, Lena Klasson-Heggebo16, Shigeru Inoue17, Norio Murase18, Sandra Matsudo19, Victor Matsudo19, Grant McLean20, Michael Sjöström21, Heidi Tomten22, Johan Lefevre23, Vida Volbekiene24 and Adrian E Bauman3 Abstract Background: Increasing empirical evidence supports associations between neighborhood environments and physical activity. However, since most studies were conducted in a single country, particularly western countries, the generalizability of associations in an international setting is not well understood. The current study examined whether associations between perceived attributes of neighborhood environments and physical activity differed by country. Methods: Population representative samples from 11 countries on five continents were surveyed using comparable methodologies and measurement instruments. Neighborhood environment × country interactions were tested in logistic regression models with meeting physical activity recommendations as the outcome, adjusted for demographic characteristics. Country-specific associations were reported. Results: Significant neighborhood environment attribute × country interactions implied some differences across countries in the association of each neighborhood attribute with meeting physical activity recommendations. Across the 11 countries, land-use mix and sidewalks had the most consistent associations with physical activity. Access to public transit, bicycle facilities, and low-cost recreation facilities had some associations with physical activity, but with less consistency across countries. There was little evidence supporting the associations of residential density and crime-related safety with physical activity in most countries. Conclusion: There is evidence of generalizability for the associations of land use mix, and presence of sidewalks with physical activity. Associations of other neighborhood characteristics with physical activity tended to differ by country. Future studies should include objective measures of neighborhood environments, compare psychometric properties of reports across countries, and use better specified models to further understand the similarities and differences in associations across countries. Keywords: Physical activity, Built environment, Neighborhood environment, International, Generalizability, Moderator * Correspondence: melody.ding@sydney.edu.au 1 Department of Family Preventive Medicine, University of California San Diego, La Jolla, California, USA 2 Graduate School of Public Health, San Diego State University, San Diego, California, USA Full list of author information is available at the end of the article © 2013 Ding et al.; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Ding et al. International Journal of Behavioral Nutrition and Physical Activity 2013, 10:57 http://www.ijbnpa.org/content/10/1/57 Background Physical inactivity accounts for a substantial proportion of the global burden of non-communicable diseases [1-3]. Population-level physical activity varies greatly by country [4-6]. The reasons for such variation are not well understood. As postulated by ecological models [7,8], physical activity is affected by multiple levels of influence, including the built and social environments [9,10]. Empirical evidence suggests that neighborhood design features, such as land use mix, are related to physical activity, primarily walking [11-13]. Recreation environments, such as parks and exercise facilities, are associated with leisure-time and overall physical activity [14]. Findings regarding neighborhood traffic, crime, and aesthetics are equivocal [12,15,16]. To date, most studies examining associations between built environments and physical activity were conducted in single countries, primarily the USA and other highincome countries. Review papers have identified this as a limitation and called for more geographic diversity in study locations [14,17,18]. An international comparison approach is important to advancing the theoretical foundation and empirical evidence of the field. Theoretically, most studies on built environments and physical activity are based on ecological models, which postulate crosslevel interactions of influence [8,19]. Conceptually, countries represent unique macro-environments as a result of socio-historical and cultural processes [8,20-22]. Attributes of macro-environments are likely to modify the associations between neighborhood environments and physical activity, but this has rarely been tested. Empirically, comparisons of associations across countries provide tests of generalizability in an international setting. Recently, researchers from the USA, Australia, Belgium, and Sweden conducted studies with comparable designs to examine the association between neighborhood walkability and physical activity [20,23-25]. Most findings from these studies supported similar associations across countries, suggesting evidence of generalizability of some associations, such as the association between objectively measured neighborhood walkability and accelerometry-based physical activity. However, such comparisons should be expanded to a larger geographic area, particularly including lower and middle income countries. The present study addresses generalizability through country-specific analyses of associations between attributes of neighborhood environments and physical activity. Data were collected from 11 countries on five continents using common methodologies, making it possible to compare associations across countries [6]. Similar patterns of associations indicate evidence of generalizability. Distinctive patterns of associations suggest country as a potential Page 2 of 11 moderator for the association between neighborhood environments and physical activity. We hypothesize that there is generalizability across most countries, in which activity-friendly neighborhood attributes (e.g., mixed land use) are positively associated with physical activity. We expect, however, that the consistency of associations across countries will differ for some attributes of neighborhood environments that may vary widely (e.g., residential density, transit access) or be more subjective (e.g., safety from crime). Methods Sampling and procedures The International Prevalence Study was a collaborative international project. The primary aim of the study was to determine nationally representative prevalence of physical activity for international comparisons. Investigators were invited to participate, but needed to demonstrate capacity and agree to follow rigorous protocols to ensure comparability of data collection methods across countries. A description of the research protocols and inclusion criteria was published elsewhere [6]. Of the 24 countries that expressed interest, 20 met the inclusion criteria and conducted data collection. Eleven countries included an environmental survey: Belgium, Brazil, Canada, Colombia, Hong Kong (Special Administrative Region of China), Japan, Lithuania, New Zealand, Norway, Sweden, and the USA. Of these countries, Brazil, Colombia, and Lithuania are upper-middle-income countries and the rest are high-income countries/regions [26]. Informed consent was provided in verbal or written format from all participants and ethics approval was obtained in each participating country. Sampling, recruitment, survey translation/adaptation, and data collection followed established protocols while allowing for minor modification in local settings (e.g., using random digit dialing or computer-assisted telephone interview) [6]. In each country, the study sample was required to be 18–65 years of age (18–40 in Japan) and representative of the overall population in a country or a significant region within a country (i.e. population of > 1,000,000). Households were randomly selected within each country/ region, and individuals within households were selected at random or by most recent birthday. The data collection was conducted in spring or fall 2002/2003 to reduce seasonal variation in physical activity. Questionnaires were either self-administered or administered by interviewers through phone or face-to-face interviews. Current analyses were restricted to participants living in towns or cities with populations ≥30,000 because the environmental measures were not suitable for rural neighborhoods. Demographic characteristics and other descriptive statistics of the analysis sample were presented in a previous paper [27]. Ding et al. International Journal of Behavioral Nutrition and Physical Activity 2013, 10:57 http://www.ijbnpa.org/content/10/1/57 Measures In all non-English speaking countries, surveys were back-translated to English and approved by investigators before data collection. Environmental attributes Attributes of neighborhood environments were measured using items from the Physical Activity Neighborhood Environment Survey (PANES) [27,28]. The testretest reliability of the questionnaire was supported in several countries [28-30]. Each single item of the questionnaire was validated against a relevant multi-item subscale of the abbreviated Neighborhood Environment Walkability Scale (NEWS-A) (Spearman correlations: 0.27 - 0.81) [29]. Neighborhoods were defined as the area within a 10-to 15-minute walk from home. Seven common items were asked in all 11 countries and were used in the current analysis. Participants reported the main type of housing in their neighborhood (e.g., apartment, townhouse, single family home) as a proxy measure for residential density. Having shops and other retail destinations in the neighborhood was used as a marker for land-use mix. The presence of transit stops (e.g. bus stops or train stations) near home was asked because public transportation often involves walking [31]. Questions were asked about the presence of sidewalks, bicycle facilities, and free or low-cost recreation facilities (e.g., parks, public swimming pools) as they provide opportunities for physical activity. Participants reported whether crime in the neighborhood made it unsafe to go on walks at night, as a marker for personal safety. The original response options ranged from 1 (strongly agree) to 4 (strongly disagree) and were recoded as “strongly agree/agree” vs. “disagree/strongly disagree,” with the exception of housing type that was dichotomized to contrast detached singlefamily homes (i.e., lower residential density) from the rest (higher residential density) [27]. Based on the literature [15,19], we hypothesized that higher residential density, the presence of shops, transit stops, sidewalks, bicycle facilities, low-cost recreation facilities near home, and better personal safety were positively associated with physical activity. We reversed the coding when necessary to reflect the expected direction of associations. Physical activity The International Physical Activity Questionnaire (IPAQ) short format was used to assess the frequency and duration of past-week walking, moderate-intensity, and vigorous-intensity physical activity that lasted for at least 10 minutes. Questions were designed to measure physical activity across all domains. Evaluation of the short IPAQ in 12 countries concluded that the questionnaire had good one-week test-retest reliability and fair-to-moderate criterion validity when compared against accelerometer total Page 3 of 11 counts [32]. When used to classify achieving physical activity guidelines or not, the short IPAQ was found to have acceptable specificity but low sensitivity [33]. The IPAQ questions were used to determine whether participants met the recommended level of physical activity, defined as 75 minutes/week of vigorous physical activity or 150 minutes of moderate physical activity accumulated in a week through any combination of walking, moderate, or vigorous physical activities [34]. Data analysis Country-specific analyses Data from each country were pooled and weighted to account for differential probabilities of sample selection within each country and to improve sample representativeness. Logistic regression was used to examine the association of each environmental variable with meeting physical activity recommendations. To examine whether the association of a neighborhood attribute with physical activity differed by country, a neighborhood attribute × country interaction was included in each model. A significant interaction suggests that the association between an environmental attribute and physical activity was not equivalent across all countries, and therefore countrystratified analyses were warranted. Forest plots were used to display the odds ratios and 95% confidence intervals for associations in each country. All models were adjusted for age and gender as they were the only common demographic variables. We conducted sensitivity analyses by repeating the regression analyses with additional key covariates (educational attainment and car ownership) in countries where these data were available (nine countries collected data on educational attainment, seven countries on car ownership). Statistical analyses were conducted in 2012 using SPSS 19.0 (SPSS Inc.). Post-hoc power analyses Because statistical power is a common concern in research on environments and physical activity, post-hoc power analyses were conducted to aide interpretation of non-significant associations [35]. Statistical power was calculated based on four key variables: the prevalence rate of the exposure (i.e., the environmental attribute), the prevalence rate of the outcome (i.e., meeting physical activity recommendations), effect size (as measured by odds ratio), and sample size. An association with a significance level at p=0.05 was equivalent to the critical value for rejecting the null hypothesis, which was also equivalent to having 0.50 power. Those significant at p<0.05 had more than 0.50 power and those nonsignificant had less than 0.50 power. Statistical power increases with increases in effect sizes and sample sizes and decreases as the prevalence rates of the exposure and outcome deviate from 0.50. Ding et al. International Journal of Behavioral Nutrition and Physical Activity 2013, 10:57 http://www.ijbnpa.org/content/10/1/57 Results Neighborhood attribute × country interactions were significant in all models tested. Therefore, analyses were stratified by country. Country-specific associations are presented in Figure 1. Country-specific associations Residential density Higher residential density was associated with higher odds of meeting physical activity recommendations in Norway; however, the association was in the opposite direction in Japan. Odds ratios in Hong Kong could not be calculated due to the lack of variance in the main housing type (only 3 out of 1100 lived in neighborhoods where the main type of housing was single-family homes). Page 4 of 11 Bicycle facilities Having bicycling facilities present in the neighborhood had a significant association with higher odds of meeting physical activity recommendations in Hong Kong (OR=1.83; 95% CI: 1.19, 2.82; p=0.006), Japan (OR=1.36; 95% CI: 1.04, 1.79; p=0.026), and the USA (OR=1.31; 95% CI: 1.06, 1.62; p=0.013). This association, however, was inverse and significant in Brazil (OR=0.68; 95% CI: 0.50, 0.93; p=0.014). Low-cost recreation facilities The presence of free or low-cost recreation facilities in the neighborhood was only significantly associated with higher odds of meeting physical activity recommendations in Hong Kong (OR=1.54; 95% CI: 1.03, 2.30; p=0.036) and Lithuania (OR=1.78; 95% CI: 1.25, 2.54; p=0.002). Associations in most countries had wide confidence intervals that overlapped 1. Shops near home In most countries, the association of having shops near home and physical activity was positive as expected. Associations in Brazil (OR=1.57; 95% CI: 1.05, 2.35; p=0.027), Hong Kong (OR=1.80; 95% CI: 1.09, 2.97; p=0.023), Japan (OR=1.60; 95% CI: 1.18, 2.17; p=0.002), and New Zealand (OR=2.00; 95% CI: 1.26, 3.18; p=0.003) reached statistical significance (p<0.05). Associations approached significance (0.052), but weak in most other countries. One potential explanation is that because transportation mode was not measured, it is unknown how much physical activity in each country was attributable to public transit use. Particularly in countries like the USA where public transit use is rare [37], the contribution of transit use to overall physical activity could be trivial. Another possible explanation for the lack of association is that in most countries access to public transit was Page 7 of 11 highly prevalent (more than 90% of participants reported having a transit stop near home in eight countries). The lack of variance could result in underestimated associations. To enrich current data and improve variance, future studies should examine additional aspects of public transit, such as pricing, frequency, and quality of service [39]. Only a small number of associations that involved access to recreation facilities were significant. However, most nonsignificant associations were in the expected direction. Because the effect sizes were small, the power for detecting significant associations was limited. This finding suggests that the presence of parks and other recreation facilities in the neighborhood could be a generalizable but weak correlate of physical activity in most countries. Both residential density and land-use mix are key components of neighborhood walkability [40]. In this study, however, there was little support for the association between residential density and physical activity. This may be because the measure of residential density was only a crude proxy. This may also suggest that land-use mix had more predictive validity and may be a more important component of walkability than residential density. A previous meta-analysis of built environments and travel behavior had similar findings and suggested that density could be an intermediate variable that influenced travel behavior through other variables such as land use mix [13]. Crime is a frequently cited barrier to physical activity, but its association with physical activity has been inconsistent [41]. The current analyses revealed similarly inconsistent findings. The association between crime-related safety and physical activity is complex because different types of crime, timing and context (e.g., day-time vs. night-time), emotional responses, and coping strategies (e.g., constrained vs. protective behavior) may affect physical activity differently. Furthermore, people from different countries and cultures may have different perceptions about safety and cope with unsafe neighborhoods differently. Also, the association between crime and physical activity might be confounded by residential density, a component of walkability. Future studies should test more complex models, compare psychometric properties of crime/safety measures across countries, and adjust for potential environmental confounders. Previous pooled analyses of the 11 countries found that five of the seven environmental correlates were significant, including shops near home, transit stops, sidewalk present, bicycle facilities, and low-cost recreation facilities [27]. The higher percentage of significant associations compared to current analyses was due to more power as a result of a larger sample size and more variability in data. However, a potential drawback of such pooled analysis is that it averages effects that could be different across countries. For example, access to transit stops was significant in the pooled analysis; however, in country-specific analyses this association was close to zero in most countries. A more Dependent variable: meeting physical activity recommendations Residential density Shops near home Transit stops near home Sidewalks present Bicycle facilities Low-cost rec facilities Crime-related safety n Prx Pry Power Prx Pry Power Prx Pry Power Prx Pry Power Prx Pry Power Prx Pry Power Prx Pry Power Belgium 348 0.66 0.74 0.08 0.63 0.74 0.06 0.75 0.64 0.82 0.84 0.69 0.26 0.78 0.76 0.02 0.78 0.72 0.12 0.76 0.76 0.03 Brazil 876 0.12 0.70 0.41 0.85 0.63 0.62 0.95 0.65 0.08 0.25 0.71 0.04 0.33 0.74 0.70 0.28 0.71 0.04 0.35 0.69 0.40 Canada 634 0.40 0.87 0.08 0.67 0.83 0.47 0.85 0.88 0.05 0.80 0.83 0.33 0.68 0.85 0.19 0.86 0.84 0.09 0.79 0.88 0.10 Colombia 2692 0.79 0.86 0.06 0.92 0.88 0.10 0.96 0.92 0.42 0.89 0.81 0.84 0.41 0.86 0.29 0.51 0.86 0.03 0.24 0.86 0.23 Hong Kong 1100 1.00 NAa NA 0.88 0.83 0.64 0.97 0.78 0.53 0.97 0.74 0.78 0.37 0.87 0.79 0.73 0.86 0.54 0.65 0.91 0.42 Japan 1221 0.71 0.63 0.90 0.83 0.45 0.87 0.91 0.32 0.99 0.58 0.44 1.00 0.25 0.54 0.63 0.59 0.56 0.04 0.67 0.49 0.98 Lithuania 1245 0.84 0.87 0.06 0.82 0.89 0.04 0.91 0.86 0.14 0.86 0.81 0.88 0.47 0.87 0.17 0.54 0.85 0.88 0.25 0.88 0.43 New Zealand 797 0.24 0.88 0.09 0.73 0.83 0.85 0.91 0.88 0.04 0.95 0.82 0.32 0.45 0.88 0.10 0.87 0.85 0.21 0.57 0.89 0.19 Norway 500 0.58 0.81 0.69 0.84 0.79 0.47 0.97 0.85 0.05 0.77 0.88 0.11 0.72 0.86 0.03 0.76 0.82 0.32 0.85 0.94 0.55 Sweden 440 0.70 0.80 0.20 0.77 0.78 0.23 0.97 0.77 0.15 0.95 0.80 0.09 0.80 0.87 0.16 0.79 0.81 0.10 0.61 0.85 0.06 USA 2560 0.40 0.84 0.10 0.59 0.83 0.44 0.68 0.84 0.04 0.75 0.84 0.03 0.57 0.82 0.70 0.69 0.83 0.18 0.67 0.83 0.21 Ding et al. International Journal of Behavioral Nutrition and Physical Activity 2013, 10:57 http://www.ijbnpa.org/content/10/1/57 Table 2 Post-hoc power analyses for logistic regression in 11 countries Prx: Pr(x=1), the probability of having a neighborhood attribute (e.g., having shops near home). Pry: Pr(y=1‫׀‬x=0), the probability of having the outcome (meeting physical activity recommendations) given that the neighborhood attribute is not present. Bolded numbers: statistical tests with >0.50 power based on post-hoc power analyses for logistic regression. Underlined numbers: associations that are close to statistical significance (0.05