The RECORD Cohort Study (“Residential Environment and CORonary heart Disease”, http://www.record-study.org) was used for the analyses 25 . As previously described, 7290 participants were recruited between March 2007 and February 2008 26 27 28 29 . The participants benefitted from a free medical checkup offered every 5 years by the French National Health Insurance System for Salaried Workers to all working and retired employees and their families. Participants were recruited without a priori sampling during these 2-hour–long preventive checkups conducted by the Centre d’Investigations Préventives et Cliniques in 4 of its health centers, located in the Paris metropolitan area (Paris, Argenteuil, Trappes, and Mantes-la-Jolie). Eligibility criteria were as follows: age 30 to 79 years; ability to complete study questionnaires; and residence in one of the 10 (out of 20) administrative divisions of Paris or 111 (out of 1301) other municipalities of the metropolitan area selected a priori. Of the persons selected for participation, 83.6% accepted to participate and completed the data collection protocol. A previous study showed that, due to the absence of random sampling, our sample was not representative of the population of the municipalities from which it was drawn: people with a high education level, people living closer from the examination centers involved in the recruitment, and people living in affluent neighborhoods and in low density neighborhoods had higher rates of participation in the study 30 .

All participants underwent a physical examination, completed questionnaires, and were geocoded with accuracy based on their residential address in 2007–2008. Research assistants rectified all incorrect or incomplete addresses with the participants by telephone. Extensive investigations with local Departments of Urbanism were conducted to complete the geocoding. Spatial coordinates and geographic codes of the street, block, and block group were searched for each participant. Precise coordinates and block-group codes were identified for 100% of the participants. The study protocol was approved by the French Data Protection Authority, which supervises any use of personal health data in France and provides ethical allowances for the processing of these data in biomedical research.

To account for geographic autocorrelation in each outcome, we estimated multilevel logistic regression models with participants nested within IRIS administrative neighborhood (the median number of residents in 2006 in these neighborhoods was 2536, interquartile range: 2161, 3115). The statistical analyses involved several steps (each time for the 4 outcomes of sport practice over the previous 7 days):

1. We estimated models including only age and sex as explanatory variables to assess variability in each outcome between IRIS neighborhoods.

2. We then included the other individual adjustment covariates and the weather variables into the models.

3. Associations of the spatial accessibility to facilities with sport practice were tested by adding to the models obtained in step 2 either the distance to the nearest facility or the number of facilities in the 1 km street network radius neighborhoods (the potential neighborhood confounders were not included at this step of the analyses). As a sensitivity analysis on the distance beyond which a facility may be considered as more difficult to use, separate models including each a spatial accessibility variable with a different distance threshold were compared.

4. We then tested one by one the other neighborhood characteristics (related to the socio-demographic and service environment), in models adjusted for individual and weather covariates (each neighborhood variable in a separate model).

5. Finally, we progressively combined into one model the spatial accessibility variables and the other contextual variables that were independently associated with each outcome (one of the aims was to assess whether the associations between the accessibility to facilities and sport practice identified in step 3 persisted after adjustment for other neighborhood variables).

All models were estimated with Markov chain Monte Carlo simulation using WinBUGS 1.4.3.21 33 .

As shown in Table 2, simultaneously considering the different individual socio-demographic factors, the odds of practicing a team sport in the previous 7 days increased with the level of human development index of the country of birth. Moreover, the odds of using a swimming pool, the odds of practicing a racket sport, and the odds of fitness exercise in the previous 7 days increased both with the level of education of participants and with household income (Table 2). Regarding weather variables, a high minimum temperature increased the odds of using a swimming pool whereas a high minimum temperature decreased fitness practice. No other weather variables were associated with the outcomes.

^a OR: Odds ratio, 95% CrI: 95% Credible Interval.

^b Models adjusted for age, sex, marital status, individual education, occupation, homeownership status, perceived financial strain, household income, and the level of human development of the country of birth.

^c Models further adjusted for minimum temperature.

As shown in Table 3, only the spatial accessibility to a swimming pool was associated with the practice of related activities (mostly swimming) over the previous 7 days, after adjusting for individual socioeconomic and weather covariates. The probability of such a practice was higher when the distance to the nearest pool was less than 1 km. As shown in Figure 2, a detailed sensitivity analysis indicated that the strength of the relationship between a binary variable for distance to the closest swimming pool and the practice of a related activity increased from 500 m (as a cutoff for the distance variable) to 1000 m and then decreased after, suggesting that the distance effect was best captured when a cutoff of 1000 m was used. Furthermore, these analyses suggest that an increase in the odds of using a swimming pool was only detected when there were two or more swimming pools in the 1 km radius residential buffer.

^a OR: Odds ratio, 95% CrI: 95% Credible Interval.

^b Models adjusted for age, sex, marital status, individual education, occupation, homeownership status, perceived financial strain, household income, and the level of human development of the country of birth.

^c Models further adjusted for minimum temperature.

No effect of spatial accessibility was detected for the other 3 groups of sports (team sports, racket sports, and fitness). However, when the practice of sports in one’s (perceived) residential neighborhood was examined as the outcome rather than the overall practice of sports, the presence of a facility at less than 1 km from the residence was associated with the odds to practice in one’s residential neighborhood both a racket sport and activities related to a swimming pool.

Few associations were identified (Table 4). The probability of practice of a racket sport and fitness in the previous 7 days increased with neighborhood income, after taking into consideration the individual socioeconomic covariates. Moreover, the probability of practice of a team sport was lower when the overall density of services in the area of residence was high. However, none of the contextual factors was related to swimming (which outcome is therefore not reported in Table 4). The observed relationship between the spatial accessibility to a swimming pool and the practice of swimming or related activities (Table 3) was therefore not confounded by the other neighborhood variables examined.

^a OR: Odds ratio, 95% CrI: 95% Credible Interval.

^b Models adjusted for age, sex, marital status, individual education, occupation, home ownership status, perceived financial strain, household income and the level of human development of the country of birth.

^c Model further adjusted for minimum temperature.

The main strength of the present study is the correspondence that we were able to establish between indicators of spatial accessibility to facilities and the practice of sports, and the fact that four categories of sports were separately examined. Other strengths of the study include the large sample accurately geocoded over the Paris Ile-de-France region; the exhaustive and high-quality Census of Sport Facilities that allowed us to assess the type and location of sports facilities; the fact that we measured the spatial accessibility to facilities with two complementary approaches that took into account the street network (distance to the nearest facility and count of facilities around the dwelling 4 ); and the examination of neighborhood factors that may confound the associations between the spatial accessibility to facilities and the practice of the corresponding sports.

However, a major study limitation is that the RECORD population is not strictly representative of the Paris metropolitan region, even if we a priori selected a panel of municipalities from the region to ensure the presence of people from all socioeconomic backgrounds. As noted in the Methods section, neighborhood factors affecting the odds of participation in the RECORD Study were identified 30 . In an attempt to reduce the magnitude of bias, these neighborhood characteristics were taken into account for the adjustment of the relationships of interest in the present study. Another limitation is that the findings were based on cross-sectional data, not allowing us to demonstrate that the relationships between spatial accessibility to facilities or neighborhood variables and sport practice are attributable to a causal effect of the former on the latter. Our data do not allow us to rule out confounding due to selective residential migration (people willing to practice a particular sport may choose to live nearby the corresponding facilities). Finally, we were unable to take into account the transportation modes available to the participants as a potential modifier of the relationships of interest between the spatial distance to sport facilities and the use of such facilities.

Independent effects of individual education and household income were found for the practice of racket sports and fitness and for the use of a swimming pool, while none of these factors were related to the practice of collective sports. Socioeconomic status has often been found to be associated with the level of physical activity 21 34 35 . An interest of the present study is to demonstrate independent associations of two individual socioeconomic indicators with the practice of various (but not all) sports. Regarding household income, cost is an established barrier to healthy behaviors in general and it is known that the cost of access to sport facilities can be a barrier to their use 36 37 38 . This is confirmed in our French sample by the association with household income for 3 of the 4 categories of sports investigated (for all categories except team sports because people commonly have free access to soccer fields for example). However, for these 3 categories of sports, the associations with individual education were stronger than those with household income, suggesting that in addition to costs, a number of intrapersonal or cognitive variables also significantly matter. Our interpretation is that a higher education level is related to a better knowledge of benefits or risks associated with behaviors, to a more accurate knowledge of health recommendations, to a higher priority given to health, to more positive attitudes towards health promotion, and possibly to a greater ability to convert intentions to comply with health recommendations into action 39 . A previous study of ours already documented strong individual education effects on the practice of jogging 40 .

This study revealed that the probability of swimming or practicing a related sport was associated with the spatial accessibility to a swimming pool. The combination of our sensitivity analysis on the distance to the closest swimming pool and of our analysis of the count of swimming pools accessible in the immediate neighborhood allowed us to characterize in a very precise way the spatial accessibility effect. The sensitivity analysis is based on the idea that the use of a too short or too long radius for the spatial accessibility variable would result in the dilution of the association of interest. The approach indicated that the association was the strongest when accessibility to a swimming pool was measured within 900 m or 1000 m from the residence. An interpretation is, for example, that spatial accessibility to swimming pools within 1500 m leads to a weaker association because it also includes swimming pools located too far away to be regularly used. Thus the sensitivity analysis indicates that swimming pools located within 900–1000 m are those which most optimally (though weakly) increase the probability of utilization. This finding is coherent with the idea that 1000 m correspond to a 10 to 15 minute walk in an urban setting and encompass the easily accessible resources 19 41 42 .

No spatial accessibility effect, however, was documented for the other categories of sports examined. A possible explanation is that team sports, racket sports, and fitness can be practiced without a facility (in private or public gardens or open spaces for example) while swimming requires access to a pool. Another explanation for the lack of association between the spatial accessibility to facilities and the practice of team or racket sports may be that these sports require the coordination among multiple people. People may have to cover longer distances to practice such sports because they also have to take into account the spatial accessibility to the facility for their partners.

A previous publication based on data of the Ile-de-France region found that the type of sport facilities spatially accessible depended on the urbanicity degree and on the degree of socioeconomic disadvantage of the neighborhoods 13 . However, our findings suggest that, in the relatively well served Ile-de-France region, such disparities in the spatial accessibility to sport facilities do not have a major impact on utilization, except perhaps for swimming pools.

The study was based on the assumption that beyond spatial access to sports facilities, various contextual characteristics may affect the practice of sports. It seemed useful to include these factors in the analysis to detect possible territorial disparities in practices and to examine whether the association identified between the spatial accessibility to a pool and the corresponding practice persisted after taking into consideration these multiple factors. An important finding is that 2 out of the 4 sports examined (racket sports and fitness) were practiced less frequently in socially disadvantaged neighborhoods, after adjustment for individual socioeconomic characteristics. The present study does not directly indicate whether residents of disadvantaged neighborhoods have a lower overall practice of sports, or whether they practice different sports, but the individual and contextual effects documented here suggest that the former alternative may be true.