Study design and sampling
This cross-sectional study was part of the European Commission-funded SPOTLIGHT project [21], with data obtained from five urban regions: Ghent and suburbs (Belgium), Paris and inner suburbs (France), Budapest and suburbs (Hungary), the Randstad (including cities of Amsterdam, Rotterdam, The Hague and Utrecht) in the Netherlands, and Greater London (United Kingdom) [22]. Sixty neighbourhoods were randomly sampled according to their level of residential density and socioeconomic status (SES), and four types of neighbourhoods were obtained: high SES/high residential density, low SES/high residential density, high SES/low residential density, low SES/low residential density [22]. A random sample of the neighbourhoods’ adult inhabitants was then invited to participate in the survey between February and September 2014. A total of 6037 individuals were recruited (10.8%, out of 55,893 invited adults). Local ethics committees in each participating country approved the study and all participants gave informed consent. Further descriptions about recruitment of participants, sampling and characteristics of neighbourhoods, are provided elsewhere [22].
Measures
Participants completed a survey on their socio-demographics, perceived characteristics of their social and physical environment, energy balance-related behaviours, self-rated weight, height and health status, and perceived barriers to healthy behaviours [22]. In addition, the objective physical neighbourhood environment was characterized using a validated virtual audit tool, as previously described [23].
Dependent variable: Leisure-time physical activity
Questions from the validated self-administered long version of the International Physical Activity Questionnaire (IPAQ) [24] were used to collect data on leisure-time PA in the last seven days. Leisure-time physical activities include walking for leisure (light intensity) and moderate-to-vigorous physical activities like aerobics, running, or cycling. We used three dependent variables to account for the variety of physical activities that can be performed at outdoor recreational facilities: total leisure-time PA, leisure-time walking, and leisure-time moderate-to-vigorous PA, expressed in minutes per week. The IPAQ showed good reliability (Spearman’s correlation coefficients around 0.8) and acceptable criterion validity (median ρ = 0.3) for adults included in a 12-country study [24]. Given the inability of accelerometers to distinguish between domains of physical activity, no information is available with regard to the separate criterion validity of the questions about leisure-time physical activity.
Independent variable: Availability of outdoor recreational facilities
As previous studies have indicated possible mismatches between objective and subjective measures of neighbourhood facilities [25, 26], we used an objective as well as a subjective measure of availability of outdoor recreational facilities in the neighbourhood.
For the objective assessment of outdoor recreational facilities, we performed a Google Street View-based virtual audit [23]. Outdoor recreational facilities were defined as any man-made or natural outdoor environments where people can exercise, play sports, or recreate in any other way; e.g. parks, soccer courts, outdoor fitness areas or skate parks. Data were collected by trained researchers for 4486 street segments in 59 neighbourhoods (Google Street View data were not available at the time of the virtual audit for one Hungarian neighbourhood) [27]. Availability of outdoor recreational facilities was defined as the percentage of street segments in a neighbourhood with these facilities present. We subsequently classified neighbourhoods into either having at least one outdoor recreational facility available or having no outdoor recreational facilities available, in order to enable comparisons between this objective measure and the subjective measure. As 87.9% of the participants objectively had at least one outdoor recreational facility available in their neighbourhood, we also divided the variable ‘percentage of street segments in a neighbourhood with outdoor recreational facilities present’ into quartiles and performed sensitivity analyses with this new variable.
For the subjective measure, we asked participants whether ‘open recreation areas (such as parks or playing fields)’ were present in their neighbourhood, and if present, whether they had used them in the last month. Answering options were: (1) present and used, (2) present and not used, and (3) not present. We performed analyses with these three categories separately, as well as with a dichotomized variable representing recreational facilities to be ‘present’ (options 1 and 2 combined) or ‘not present’ (option 3), to allow comparison with the objective measure.
Self-selection variables
We also asked respondents about factors that influenced their decision to live in that neighbourhood. We separately asked respondents about eight factors (e.g. cost of housing, family/friends living nearby) that might have influenced their decision. We used the item ‘It is close to recreation facilities, parks or sports facilities’ as an indicator for direct self-selection in relation with the research question asked here.
For the indirect self-selection measure, we used education level as a socioeconomic indicator. Education was self-reported in the survey with multiple but differing categories in each country [22]. We combined these categories to classify the education level of participants as either higher (college or university level) or lower (below college level).
Covariates
Participants reported their birth year, gender and self-rated health, which was measured with a Visual Analogue Scale ranging from 0 (very unhealthy) to 100 (very healthy) [28]. Urban region was also used as a covariate. We tested the effect of adjusting for season or month in which the survey was completed, but since this changed the results little, we decided to present the associations unadjusted for season or month.
Motivations for the (non-)use of outdoor recreational facilities
If participants indicated that recreational facilities were available in their neighbourhood, and they used them, we asked them about the most important motivation for their usage. Seven answering options were available: ‘distance from home’; ‘it is on my route’; ‘price’; ‘my family/friends go here’; ‘I like to go here’; ‘parking’; ‘other’ (only one option could be chosen).
If participants indicated that outdoor recreational facilities were present, but they did not use them, we asked them about the most important motivation for the lack of use. Five answering options were available: ‘too far from home/work/school’; ‘it is not somewhere I would normally go’; ‘price’; ‘parking’; ‘other’.
Statistical analyses
After the exclusion of individuals for whom no objectively measured data on the physical neighbourhood environment were available (N = 838), a sample of 5199 participants was included in the analyses. Descriptive statistics were used to provide insight into participants’ characteristics. ANOVA, Chi-square, and Mann-Whitney U tests were used to assess if there were statistically significant differences between groups with and without available outdoor recreational facilities (for the subjective and objective measure).
All variables were examined for non-response, with percentages ranging from < 1% (age) to 23.7% (preference for neighbourhoods with recreational facilities). Multiple imputations were performed, under the assumption that missing values were missing at random (i.e. missing values are dependent on observed data and not on unobserved data) [29]. Thirty imputed datasets were created by Predictive Mean Matching, based on the percentage of missing values.
The dependency of observations within neighbourhoods and countries was evaluated and revealed relevant clustering of individuals within neighbourhoods. Because of the non-normal distribution and high proportion of zeros in the variable leisure-time PA, negative binomial regression analyses were conducted using generalised estimating equations (GEE) with an exchangeable structure [30] and having the neighbourhood level as grouping variable. The coefficients and 95% confidence intervals (CI) generated from the multivariable GEE negative binomial regression analysis were exponentiated to represent rate ratios and their respective CIs. Rate ratios can be translated into the difference in minutes of leisure-time PA per week between those with and without outdoor recreational facilities available by multiplying the rate ratio with the median leisure-time PA of the reference category.
Age, gender, self-rated health and urban region were first tested as effect modifiers by adding interaction terms to the model. Since none of them were significant (p < .10) effect modifiers, they were added to the model as confounders.
We assessed the association between subjective and objective availability of outdoor recreational facilities with total leisure-time PA, adjusted for confounders (Model 1). In Model 2, we added the indirect self-selection variable education. In Model 3, we replaced the indirect self-selection variable by the direct self-selection variable preference for neighbourhoods with recreational facilities. In Model 4, both self-selection variables were added. To quantify the contribution of self-selection variables to the association between the availability of outdoor recreational facilities and leisure-time PA, we calculated the percentage change in coefficient between Models 2,3 and 4 with Model 1.
To examine if use of recreational facilities was more strongly associated with leisure-time PA than (perceived) availability alone, we assessed the association between perceived availability and use of recreational facilities in the neighbourhood. In a last step, we described the most common motivations for (not) using recreational facilities in the neighbourhood using pie charts.
In a sensitivity analysis, we used only complete cases to ensure robustness of findings. These results were comparable to the analyses with imputed data (see Additional file 1: Table S1). In addition, we repeated the analyses with a dichotomized measure of the objective availability of outdoor recreational facilities with quartiles of availability of outdoor recreational facilities, in relation to all three leisure-time PA measures (see Additional file 1: Table S2). Analyses were performed using IBM SPSS statistics for Windows V.23.0. P < 0.05 was considered statistically significant.