Neighbourhood safety and leisure-time physical activity among Dutch adults: a multilevel perspective

Study population

Cross-sectional data on individual characteristics and leisure-time walking and cycling behaviour were obtained from the Dutch national health survey of 2006 to 2009, which is part of the yearly administered Dutch Integrated Survey on Household Living Conditions (POLS). A random nationwide sample of in total 57,281 non-institutionalized persons was drawn from the national population registry. Selected individuals were approached by an interviewer and asked to participate in an interview and, if 12 years and older, to fill in an additional paper-and-pencil survey on specific health topics, including PA. There was a non-response of 36% for the interview, with an additional 16% non-response for the survey. Due to the age restriction, 16% of the sample was not eligible to complete the paper-and-pencil survey. A total of 25.206 persons of 12 years and older completed the survey.

Cross-sectional data on residents’ perceptions of neighbourhood safety were obtained using the three-yearly conducted Dutch Housing Questionnaire (WoON) of 2006. A random nationwide sample of 113,837 non-institutionalized adults (18 years and older) was approached by phone and asked to complete an interview either by phone, face to face, or on paper. There was a non-response of 42%. A total of 64,005 adults from 3,495 neighbourhoods completed the survey. To assess the safety situation in each neighbourhood at large, we constructed neighbourhood-level safety scores by averaging the scores of all WoON res pondents living in the same neighbourhood. Neighbourhood-level safety data from the WoON survey were linked to individual-level data from the Dutch national health survey using the 4-digit zipcode.

From the POLS data, we excluded respondents younger than 18 (N= 2,237), respondents whose neighbourhood-level safety scores have been based on less than five observations in the WoON survey (N=2,369) and respondents with unrealistic and missing PA scores (N=554 for walking, N=566 for cycling). PA scores were classified as unrealistic if the score exceeded 3360 minutes per week, which equals 8 hours each day of the week. A total remained of 20,046 POLS respondents with valid walking scores and 20,034 respondents with valid cycling scores, living in 2,127 neighbourhoods.

Measures

Statistical methods

To take the sampling design of the POLS survey into account, prevalence of leisure-time walking and cycling were weighted for age, gender, marital status, household size, urbanization, province, and month of survey. The associations of (specific components of) neighbourhood safety with leisure-time walking and cycling were assessed by means of odds ratios derived from multilevel logistic regression analysis. Level 1 represented individuals, level 2 represented the neighbourhoods. A three-stage modelling approach was used to make a stepwise adjustment for age, gender, household composition, ethnicity (model 1), education, household income, household wealth (model 2) and population density (model 3). Each potential confounder was included as a set of dummy variables. Subpopulation analyses were performed using model 3. Potential effect-modification by age-gender, education and neighbourhood population density was investigated by entering interaction terms in the model. Intraclass Correlation Coefficient (ICC) was estimated using Rho to display the proportion of total variance in leisure-time walking and cycling that is attributable to the neighbourhood level. Associations were weighted using a complex survey sample design with sample weights. Results were almost identical to those presented below. For all analyses, statistical significance was set at 0.05. Analyses were carried out using the STATA 11.0 software.

This study was based on secondary analyses of anonymized data by Statistics Netherlands (the “CBS”), with approval of the CBS authority for privacy protection. This authority has the responsibility to guarantee that all activities of the CBS are in strict agreement the Dutch laws for the protection of privacy of residents and subjects.

Limitations

The data available to this study had some potential limitations. First, while many potential confounders at the individual level have been accounted for, not all neighbourhood-level factors have taken into account that may have confounded the association between safety and PA. For example, an area with mixed land use and nearby shopping and recreational facilities may stimulate leisure-time walking, while such an area might at the same time be relatively unsafe. In the present study, confounding by such factors may have concealed an association between safety and leisure-time walking. In order to address this potential source of bias, we have adjusted our analyses for population density, as this variable may correlate with land use mix and the proximity of facilities. We found that control for population density had negligible effects on the associations between safety and walking or cycling. Future research should aim to control for neighbourhood-level confounders in more detail.

There was a non-response of 36% for the POLS interview and another 16% for the additional paper-and-pencil survey. Non-respondents may have differed in important aspects from the study population. However, this would only have biased the results of this study if non-response varied according to the level of individual PA, as well as the degree of neighbourhood safety, after accounting for confounders. We think this is not very likely, but we can not entirely exclude the possibility of some bias.

Another limitation has to do with the self-reported nature of PA. People may have difficulty estimating their time spent on PA and may be inclined to give socially desirable answers, which might have caused them to overestimate their levels of PA [31]. However, this only becomes a problem if the amount of overestimation differs between safe and unsafe neighbourhoods after controlling for individual-level confounders. Again, we think this is not very likely.

Neighbourhood safety was also self-reported. We chose to use subjective measures of neighbourhood safety, because there is evidence that objective and subjective ratings are poorly correlated [16, 32] and that subjective ratings may be more important in determining PA behaviour [32, 33]. Those in favour of objective measures, have argued that objective measures are needed to avoid bias associated with self-reports, such as the ‘single source bias’ [34]. This source of bias may occur if self-reported data on both determinants and outcomes are collected using the same survey. To avoid this source of bias, we have used two different sets of respondents for measuring neighbourhood safety (the WoON survey) and leisure-time walking and cycling (the POLS survey).

Interpretation of key results

Associations between safety and leisure-time cycling have been found to be strongest for general safety. This suggests that a combination of safety problems may have more impact on people’s engagement in cycling than isolated safety problems. The weakest association with cycling has been found for traffic safety. This supports an earlier study that found traffic safety to have less influence on leisure-time PA than other safety components [6]. The weak association for traffic safety is remarkable as neighbourhood variations in levels of safety were most profound for traffic safety (Table 2). Results of the present study suggest that people are more influenced by social aspects of safety such as crime and social disorder, than by traffic safety.

Subgroup analyses suggested that elderly women were most influenced by general neighbourhood safety in their cycling behaviour. This supports earlier research [22, 35]. Foster et al. [22] suggest that women and elderly may be more sensitive to safety as they feel more physically vulnerable than men and young adults.

The positive association between neighbourhood safety and cycling may in part reflect a safety component that has not been included: fear of bicycle theft or damage. This seems to be a plausible mechanism since bicycle theft rates among bicycle owners in the Netherlands are the highest in the world [24].

Leisure-time walking was not significantly associated with general neighbourhood safety or any of its safety components. An American study that used various objectively measures safety components, yielded the same results [6]. Many other studies, too, did not find an association between safety and leisure-time walking [6–13, 17, 20, 21]. Some evidence of an association was found in other studies, but results were mostly inconsistent [4, 5, 14–16, 18, 19].

Lack of a positive association in the present study might have been due to focus on fairly large neighbourhoods. 4-digit zip code areas have an average of 4.088 inhabitants [36]. Environmental influences of safety on walking may have been more apparent on a smaller scale, since walking is primarily performed in the immediate area around the home. Future research should assess if a focus on smaller areas reveals a more significant association.

Subgroup analyses have shown an inverse relationship between safety and leisure-time walking for young men and for lower educated. Earlier studies support this finding [18, 37, 38]. Ross et al. [38] suggest that unsafe neighbourhoods have a culture of being outside on the streets, which may be especially attractive to these groups.