Skip to main content

The contribution of area-level walkability to geographic variation in physical activity: a spatial analysis of 95,837 participants from the 45 and Up Study living in Sydney, Australia



Individual-level studies support a positive relation between walkable built environments and participation in moderate-intensity walking. However, the utility of this evidence for population-level planning is less clear as it is derived at much finer spatial scales than those used for regional programming. The aims of this study were to: evaluate if individual-level relations between walkability and walking to improve health manifest at population-level spatial scales; assess the specificity of area-level walkability for walking relative to other moderate and vigorous physical activity (MVPA); describe geographic variation in walking and other MVPA; and quantify the contribution of walkability to this variation.


Data on sufficient walking, sufficient MVPA, and high MVPA to improve health were analyzed for 95,837 Sydney respondents to the baseline survey of the 45 and Up Study between January 2006 and April 2010. We used conditional autoregressive models to create smoothed MVPA “disease maps” and assess relations between sufficient MVPA to improve health and area-level walkability adjusted for individual-level demographic, socioeconomic, and health factors, and area-level relative socioeconomic disadvantage.


Within-cohort prevalence of meeting recommendations for sufficient walking, sufficient MVPA, and high MVPA were 31.7 (95% CI 31.4–32.0), 69.4 (95% CI 69.1–69.7), and 56.1 (95% CI 55.8–56.4) percent. Prevalence of sufficient walking was increased by 1.20 (95% CrI 1.12–1.29) and 1.07 (95% CrI 1.01–1.13) for high and medium-high versus low walkability postal areas, and for sufficient MVPA by 1.05 (95% CrI 1.01–1.08) for high versus low walkability postal areas. Walkability was not related to high MVPA. Postal area walkability explained 65.8 and 47.4 percent of residual geographic variation in sufficient walking and sufficient MVPA not attributable to individual-level factors.


Walkability is associated with area-level prevalence and geographic variation in sufficient walking and sufficient MVPA to improve health in Sydney, Australia. Our study supports the use of walkability indexes at multiple spatial scales for informing population-level action to increase physical activity and the utility of spatial analysis for walkability research and planning.

Peer Review reports


Promoting moderate-intensity walking is the cornerstone strategy of public health efforts to increase population levels of participation in moderate and vigorous physical activity (MVPA) [14]. Walking is low risk [5]; accessible to most people regardless of age, sex, socioeconomic status, or cultural background [2]; and confers health benefits independent of participating in more vigorous forms of physical activity [6]. Walking may be undertaken for recreation, leisure, and health; to move between destinations and origins; and to access services [7, 8]. These latter activities describe utilitarian walking or active transport, and have been a focus for built environment research over the last two decades [9]. “Walkability” is the term used to describe the capacity of built environments to facilitate walking for multiple purposes [8], especially active transport [10]. Walkable neighbourhoods facilitate active transport by reducing distances between origins and destinations, and maximizing the mix of proximal land uses for residential, commercial, educational, and recreational purposes [7, 10, 11].

Walkability is typically operationalized as an index of high-resolution built environment variables within a geographic information system. The most widely utilized and researched of these are the North American Neighborhood Quality of Life Study (NQLS) [12] and South Australian Physical Activity in Localities and Community Environments (PLACE) [8] indexes. Developed in parallel for comparing between North American and Australian populations [8], these metrics comprise four environmental variables: residential density, street network connectivity, land use mix, and retail floor area ratio, which are combined into a total score by summing over either decile ranks [8] or standard (Z) scores [12]. The total score is then divided into sample-specific quartiles representing the relative variation in walkability between units of analysis [8, 12]. The indexes have been adapted and validated for use in other cities and countries (e.g. [8, 9, 1319]), and underpin a large international evidence base demonstrating consistent associations between environmental walkability and levels of moderate-intensity walking that benefits health [17].

Creating local opportunities for transport-related walking through strategic land use and infrastructure developments is a key strategy of many regional development plans [9, 17] such as the Sydney Metropolitan Plan [20], and may contribute to population levels of total daily physical activity [3, 4, 7]. Environmental and policy interventions such as these generally have much smaller effect sizes than those targeting individuals [21] but can have larger population-level impacts because exposure to the built environment is ubiquitous and changes more persistent than interventions with individuals [7, 22, 23]. This has prompted some to recommend using walkability indexes to inform urban design, transportation, and health policy; target infrastructure investments; and evaluate environmental interventions to increase population-levels of physical activity [8, 9, 17, 2427].

Needs assessment, planning, intervention, and evaluation activities to address the health of populations typically occur at larger regional levels [28], which Saelens and Handy have termed macro-level environments [10]. These are distinct from micro-level environments specific to individuals and meso-level environments that are shared by groups of individuals, such as residents within a neighborhood [10]. Greater use of objective walkability indexes has demonstrably progressed our understanding of environment-behavior relations through a focusing of research and inference at increasingly finer spatial resolutions (see reviews [10, 11, 2931]). However, this has contributed to an evidence base derived at geographic scales substantially smaller than those used for population health policy, planning, and intervention; assumes individual-level environment-behavior relations scale to populations; and raised concerns about the utility of micro-level evidence for macro-level health programming [32].

The extent to which micro-level correlates of physical activity manifest at macro-levels is poorly understood [33], under-researched, and limited to journey to work data among employed populations. For example, Frank et al. have reported that the prevalence of employed persons walking to work at the 2000 United States Census was 4-7% higher in the most compared to least walkable block groups of King County and Baltimore after stratifying on household income [12]. We too have observed higher prevalence of income-stratified self-reported walking to work in the 2006 Australian Census for Sydney residents in the most (7.9-11.0%) versus least (2.1-3.0%) walkable Census Collection Districts and adjusted odds of walking to work 2.8-3.3 times higher for the top versus bottom walkability quartiles [17]. Similarly, Kelly et al. have recently reported that the odds of ≥5% of employed block group residents reporting walking to work in the 2004–2009 American Community Survey were 1.6-5.5 times higher for the most compared to least walkable neighborhoods in St. Louis City and County after adjusting for area-level socioeconomic deprivation [34]. These small number of macro-level built environment studies are consistent with a positive association between population-prevalence of utilitarian walking and area-level walkability. However, the behavioral specificity of outcome measures for walking to work among employed persons in these studies limit their generalizability to the broader population and other domains of walking.

Walkability research has also largely ignored geographic variation in behaviors and outcomes, which is essential for framing public policy and action [35]. Spatial variation in outcomes and behavior beyond that explained by demographic, social, and economic factors may indicate additional, unobserved, and geographically varying influences on health and health-enhancing infrastructure [36]. Spatial analysis is also a potent tool for identifying environmental inequalities [37] and may assist in targeting infrastructure upgrades and developments to mitigate environment-related health risks, and promote equitable access to health-enhancing built environments [38, 39]. For example, Huang and colleagues have identified clusters of high active transportation (walking or riding) among residents of Los Angeles and San Diego Counties living in Census block groups with higher population, employment, street, block, and intersection densities; shorter block lengths; and the presence of a bus route [38]. Tamura et al. have also observed significant clustering of prevalence of meeting physical activity guidelines in California, Massachusetts, and Pennsylvania respondents to the 2004 Nurses’ Health Study [39]; however, they found inconsistent evidence for macro-level differences between cluster and non-cluster neighborhoods [39].

Given the limited evidence on macro-level relationships between walking and walkability, and the potential for geographic analysis to inform this research, the aims of the present study were to: (1) evaluate if area-level walkability was associated with population-levels of moderate-intensity walking; (2) assess the specificity of area-level walkability for walking compared to other MVPA; (3) describe geographic variation in walking and other MVPA; and (4) quantify the contribution of area-level walkability to this variation using a population-based cohort living in Sydney, Australia. We hypothesized that: (1) area-level walkability would be associated with population-levels of sufficient walking to improve health but not other MVPA; (2) sufficient walking and other MVPA to improve health would vary geographically in the study area; and (3) area walkability would contribute to this variation but only for sufficient walking to improve health.


Study design and area

We used a cross-sectional, ecological design to investigate geographic variation in physical activity behavior and its relationship to walkability in the Sydney Statistical Division of NSW, Australia [40]. Sydney covers a land area of 12142 km2 and had a population of 4.1 million persons living in 1.6 million dwellings at the 2006 Census [41]. Analysis was undertaken at the Census postal area level, which was the smallest spatial unit at which geographically identified 45 and Up Study data were available from the data custodian. Sydney comprised 260 conterminous postal areas at the 2006 Census [42] with a median land area of 7.6 km2, 5304 residential dwellings, and 13090 residents [41].


Participants were selected from members of the The Sax Institute’s 45 and Up Study, a population-based cohort established between January 2006 and December 2010 to investigate healthy aging among persons 45 years and older living in NSW, Australia [43]. Prospective participants were randomly sampled from the Medicare Australia enrollment database and invited to return a completed consent form and baseline questionnaire via mail [43]. People aged ≥80 years were oversampled by a factor of two; rural and remote populations were also oversampled but these subgroups are not resident within the Sydney area [43]. The 45 and Up Study includes approximately 10% of the NSW population and had a response rate of 18% [43]. This is similar to other international population cohorts that seek consent for data linkage (e.g., [44, 45]), and consistent with the global trend of reducing participation rates in epidemiological studies [46]. Individual-level data were provided by The Sax Institute with 2006 Census postal areas identifiers for all 266848 persons recruited to the study between January 2006 and April 2010 [47]. We limited our analysis to the 115153 persons from this release that were geocoded to the Sydney Statistical Division to correspond with the spatial extent of our study factor.


Individual-level data comprised self-reported responses to the baseline questionnaire of the 45 and Up Study [43], and were used to derive respondent-level physical activity outcome variables and covariates. Postal area-level data included the Sydney Walkability Index (SWI) [17] and the 2006 Index of Relative Socioeconomic Disadvantage (IRSD) [48], which were included as study and covariate factors, respectively.

Outcome variables

We defined three physical activity outcomes: sufficient moderate-intensity walking to improve health, sufficient MVPA to improve health, and high MVPA. Each outcome was derived from self-reported responses to Active Australia Survey [49] questions included in the baseline survey. Participants were asked to report separately the number of times in the last week they had: (i) walked continuously for at least 10 minutes for recreation or exercise or to get to or from places; (ii) participated in vigorous physical activity that made them breathe harder or puff or pant; (iii) participated in other moderate physical activities. Participants also reported the total time they spent doing each of these activities.

Sufficient total moderate and vigorous physical activity to improve health

Moderate and vigorous physical activity was calculated as the sum of total minutes engaged in walking and other moderate-intensity activities plus two times the total number of minutes engaged in vigorous physical activities. Respondents were classified as sufficiently active if they accumulated ≥150 min of MVPA over ≥5 sessions of at least 10 minutes duration. A double weighting is given to vigorous activity to reflect its greater intensity, and a threshold of five sessions operationalizes the Australian physical activity guidelines recommendation that adults be active on most days of the week, and assumes sessions are most likely to occur on separate days [50].

Sufficient walking to improve health

Sufficient walking to improve health used the same frequency and duration criteria as sufficient MVPA but only used responses on walking in calculations. This outcome measure identified respondents that met Australian physical activity guidelines from walking alone. We included this outcome on the basis that walkability indexes have higher specificity for walking behavior—especially for utilitarian purposes—than for MVPA [8, 12, 17].

High moderate and vigorous physical activity

We categorized respondents as highly physically active if they reported participating in ≥300 minutes and ≥5 sessions of MVPA over the previous week. The purpose of this outcome was to identify respondents at or above the upper Australian guideline limit of 300 minutes of MVPA per week [51]. This represents a level of MVPA at which even more health gains are accrued [52] and potentially a minimum level of MVPA required for certain health outcomes such as prevention of weight gain and some cancers [53].

Study variable

The study variable of interest was postal area walkability measured using the SWI [17]. The SWI is an abridged version of the NQLS and PLACE indexes [17]. It has predictive validity for utilitarian walking, and is a cohesive and internally consistent measure of walkability in Sydney, Australia [17]. The index is derived within a geographical information system and operationalizes three measures of the built environment associated with walking:

  1. 1.

    Residential dwelling density—number of residential dwellings per hectare of residential land use

  2. 2.

    Intersection density—number of intersections with three or more road junctures per square kilometre of total land area

  3. 3.

    Land use mix—entropy of five land use classes (residential, commercial, industrial, recreational, and other uses) adjusted for differences in the size of spatial units [54].

Variable values are divided into deciles, scored from 1 (lowest) to 10 (highest), summed to give a total score out of 30, which is then divided into quartiles corresponding to low, low-medium, medium-high, and high walkability [17]. We have previously demonstrated that abridged indexes excluding retail floor space ratio data, which are difficult to source [8, 12] and frequently unavailable [55], have comparable measurement properties to four-variable NQLS and PLACE indexes [17].

We calculated SWI variables within the spatial extents of Sydney postal areas, which have a median land area of 7.6 km2 that approximately corresponds to a radial buffer of 1550 meters. This is within the range of “high resolution” buffers typically used for individual-level analyses, [56, 57] and for which consistent environment-behavior associations have been reported across adult life stages, including older adulthood [56].


We included individual and area-level correlates and determinants of physical activity from the research and 45 and Up Study literatures (e.g. [5864]). Individual-level covariates included sex; age at baseline interview; language spoken at home; education level; relationship status; employment status; health insurance type; smoking status; body mass category; number of chronic conditions ever diagnosed and number of chronic conditions treated in the last month; and physical function limitation and role limitation due to emotional problems sub-scales from the RAND 36-Item Health Survey (Version 1.0) [65].

Area-level socioeconomic disadvantage was measured for postal areas using the 2006 IRSD [48]. The IRSD is a general measure of disadvantage derived from Census variables indicative of low socioeconomic well being such as percent of population ≥15 years with no post school qualification; percent of population unemployed; percent of employed persons classified as laborers; percent of private dwellings with no motor car; and percent of people who do not speak English. We divided the postal area distribution of IRSD scores into quintiles indicating high through to low relative socioeconomic disadvantage.

Statistical analysis

We used a two-stage strategy to model within-cohort prevalence. In the first stage predicted probabilities (\(\hat {Y}_{ij}\)) of achieving sufficient walking, sufficient MVPA and high MVPA to improve health were calculated for each respondent using logistic regression models conditioned on individual-level covariates. These probabilities were summed within the j postal areas to obtain the predicted number of outcomes for each unit adjusted for its underlying respondent structure [6668].

In the second stage postal area prevalence ratios (PR) for physical activity outcomes were estimated using Bayesian Besag, York, and Mollié conditional autoregressive models with Poisson likelihoods [69]. This model is commonly used in epidemiology for small-area disease mapping estimation [70, 71] and decomposes area-specific random effects into a local, spatially structured variance component (s j ) and a global, spatially unstructured (heterogeneity) variance component (u j ) [70, 72]:

$$ log(\theta_{j}) = \alpha + x_{j}\beta + s_{j} + u_{j} + log(e_{j}) $$

where θ j is the relative risk for the j th area; α is the overall relative risk across the study region; x j and β are optional vectors of ecological explanatory variables and parameter estimates, respectively; and e j is an offset representing the expected number of cases in the j th area, which we derive using either the overall prevalence (e j =p×n j ) or sum of predicted probabilities from stage one (\(e_{j} = \sum \hat {Y}_{ij}\)).

The heterogeneity component (u j ) was given a normal prior with mean 0 and precision \(\tau _{u}^{2}\) [70]. The local smoothing component (s i ) was given an intrinsic conditional autoregressive prior [70] with mean \(\bar {s}_{j}\) and precision \({tau}_{j}^{2}\) conditioned on the mean risk in the k surrounding postal areas with intersecting boundaries. Variability of u j and s j were controlled by hyper-parameters \(\tau _{u}^{2}\) and \(\tau _{s}^{2}\) [70], which were given Gamma hyper-priors of γ(0.5,0.0005) [73].

Six models were fit for each outcome. Model 1 (M1) was an unadjusted disease mapping model with expected cases proportional to the overall prevalence (p×n j ). Model 2 (M2) was also a disease mapping model but with individually-adjusted expected cases from stage 1. Models 3-6 were ecological regressions. Model 3 (M3) added SWI to M2; Model 4 (M4) added IRSD to M2; Model 5 (M5) added IRSD to M3; and Model 6 (M6) added effect modification of SWI by IRSD to M5.

Medians and 95% credible intervals (CrI) for each model parameter were summarized from the posterior distribution obtained from two Monte Carlo Markov Chains with over-dispersed starting values. Each chain ran for 2.5 million iterations with every 250th sample retained to reduce autocorrelation and improve convergence. The first half of each chain was discarded as burn-in leaving 10000 samples in total for inference. Model convergence was assessed using trace and autocorrelation plots, and Gelman-Rubin diagnostics [74].

The Deviance Information Criterion (DIC) was used to choose between spatial models [70] and evaluate the importance of area-level variables [75]. We considered models within 1–2 DIC units of the best model (i.e., lowest DIC) as deserving consideration, 3–7 as having less support, and >7 no support [76]. An increase in DIC between nested models was interpreted as support for selecting the variable omitted from the reduced model [75]. We visualized the exponentiated sum of spatial and non-spatial variance components using choropleth maps to identify variation in excess of that attributable to fixed-effect factors. We also calculated the spatial fraction \(\left ({\rho = \sigma ^{2}_{s}/}\left [{\sigma ^{2}_{s} + \sigma ^{2}_{u}}\right ]\right)\) from the marginal variation of random effects to determine the proportion of residual variation due to spatially-structured factors [77, 78].


Weighting of the 45 and Up Study sample is not required to estimate externally valid relative effect measures when non-spatial analyses condition on the variables used to construct post-stratification weights [79]. However, whether weighting is necessary for valid geographical analysis of the cohort is unclear. Unweighted spatial analyses are simpler to implement within a Bayesian disease mapping framework [80, 81] but geographically structured (non) response rates can bias inference [82]. In the context of our study, this potential for bias arises through the estimation of postal area prevalence ratios from sample counts and expectations. We evaluated the need to adjust our sample for response rates by comparing postal area prevalence ratios derived using unweighted and weighted sample data for each physical activity outcome. Post-stratification survey weights were calculated to benchmark the study sample to the Sydney Statistical Division population from the 2006 Census [83], with post-strata formed by 2006 Census postal areas (n=260), sex (male and female), and five-year age groups (45–84 and ≥85 years). We evaluated the performance of unweighted data using scatter plots and Pearson correlation coefficients to visualize relationships and strength of associations with weighted postal area prevalence ratios calculated using both unadjusted and adjusted expected cases.

All data analysis and mapping was undertaken in R 3.2.2 using R2WinBUGS 2.1-21 and sp 1.2-1. Correlation coefficients, t-tests, and general linear models were evaluated at the 5% alpha level and conditional autoregressive models using DIC and 95% credible intervals summarized from posterior distributions.


Complete data were available for 95837 of 115153 (83.2%) cohort members living in 255 of 260 (98.1%) postal areas in Sydney. Respondent counts within postal areas ranged from 0–3481 with a median and interquartile range of 271 and 152-466. Sample characteristics for people included in our study are reported in Table 2. Gender and employment status were comparable to 2006 Census estimates for the study area [83]; however, similar to the full cohort profile [47], our sample was younger, more highly educated, less likely to speak a language other than English at home, and more likely to be living with a partner than the study population.


Median walkability scores for low, low-medium, medium-high, and high walkability quartiles were 5, 13, 19, 27, respectively. Table 1 reports the median, minimum and maximum values for environmental variables by walkability quartiles. Variable values increased monotonically but non-linearly between successively increasing walkability quartiles. The ratios of environmental median values for high compared to low walkability quartiles were approximately two times higher than for medium-high compared to low quartiles, which were approximately two times higher than for low-medium compared to low quartiles. Interquartile ranges for residential density (0.74–7.25, 9.63–15.62, 16.95–22.31, and 36.60–66.91), intersection density (1.94–10.34, 34.28–59.01, 66.87–89.86, and 121.50–203.00), and land use mix (0.002–0.014, 0.020–0.048, 0.047–0.081, and 0.088–0.218) did not overlap when stratified by low, low-medium, medium-high, and high walkability quartiles, and indicates that each variable is contributing to the segmentation of postal area walkability.

Table 1 Median, low, and high values for Sydney Walkability Index environmental variables by walkability quartiles

Prevalence of physical activity outcomes

Prevalence of sufficient walking, sufficient MVPA, and high MVPA to improve health were 31.7 (95% CI 31.4-32.0), 69.4 (95% CI 69.1-69.7), and 56.1 (95% CI 55.8-56.4) percent, respectively. Frequencies, relative frequencies, and prevalence of physical activity outcomes for area-level factors are reported at the top of Table 2. Sufficient walking to improve health displayed the strongest prevalence gradient for area-level walkability followed by sufficient MVPA; the gradient for high MVPA was small and inconsistent. Prevalence of all outcomes increased with increasing area-level socioeconomic disadvantage but the gradient was weakest for sufficient walking to improve health.

Table 2 Sample characteristics and prevalence estimates for physical activity outcomes

Individual-level factors

Sample characteristics and prevalence of study outcomes for individual-level factors are reported in Table 2 and full-model odds ratio estimates in Table 3. Physical activity outcomes were strongly associated with individual-level demographic, social, economic, and health status and behavior factors. For all outcomes, prevalence of sufficient MVPA increased with increasing education level; decreased with increasing numbers of diagnosed and treated chronic health conditions, functional limitation and emotional problems; and were higher for females, people who spoke English at home, or were non- or ex-smokers. These gradients and differences were less pronounced for sufficient walking to improve health than for either sufficient MVPA or high MVPA. This was also observed for age, which displayed inverted U-shape associations with prevalence of MVPA outcomes. Prevalence of sufficient walking to improve health decreased with increasing body mass and was especially low in obese persons; body mass gradients were less consistent for sufficient MVPA and high MVPA. Health insurance type was unrelated to prevalence of sufficient walking to benefit health but strongly related to both sufficient MVPA and high MVPA prevalence. The areas under the curve for fully adjusted individual-level logistic regression models were 61.0%, 66.6%, and 64.9% for sufficient walking, sufficient MVPA, and high MVPA, respectively.

Table 3 Full-model odds ratio estimates for individual-level adjustment variables


Figure 1 summarizes relationships between unweighted and weighted physical activity prevalence ratios for the 255 postal areas for which survey data were available. Left panel plots show prevalence ratios derived using unadjusted expected values, and right panel plots show prevalence ratios derived using expected values adjusted for individual-level factors. Strong linear relationships were observed between unweighted and weighted prevalence ratios for all physical activity outcomes, regardless of the method used to derive expected cases. Correlation coefficients for sufficient walking, sufficient MVPA, and high MVPA were 0.98, 0.96, and 0.97 for unadjusted, and 0.98, 0.94, and 0.96 for adjusted prevalence ratios, respectively. Based on these strong relations, we determined that weighting was not necessary for spatial analyses, and fit disease mapping models to unweighted sample data.

Fig. 1
figure 1

Comparison of unweighted and post-strata weighted postal area prevalence ratios for physical activity outcomes in Sydney Statistical Division. Post-strata were formed by postal areas (N=260), sex (male and female), and five-year age groups (45–84 and ≥85 years). Left panel plots show relationships for prevalence ratios with no adjustment for individual-level factors. Right panel plots show relationships for prevalence ratios adjusted for individual-level demographic, economic, and health factors using logistic regressions

Spatial analysis

Tables 4, 5 and 6 summarize conditional autoregressive models 1–5 for sufficient walking, sufficient MVPA, and high MVPA to improve health, respectively. Mean baseline models (M1) indicated very high levels of clustering with ≥97% of residual variation due to unobserved, spatially structured factors. Differences in effective parameters (pD) and DIC values indicated that the addition of expected cases adjusted for respondent-level variables (M2) simplified models and substantially improved fits over M1 for sufficient walking (Δ DIC =-16.3), sufficient MVPA (Δ DIC =-76.9), and high MVPA (Δ DIC =-76.7). These were the best fitting models for sufficient MVPA and high MVPA, and reduced spatial variation by 84.2% and 82.2%, respectively. The best fitting model for sufficient walking was M5, which included SWI and IRSD, and reduced the DIC and spatial variance by 20.8% and 75.6% over M1, respectively.

Table 4 Conditional autoregressive model summaries for sufficient walking to improve health
Table 5 Conditional autoregressive model summaries for sufficient MVPA to improve health
Table 6 Conditional autoregressive model summaries for high MVPA

Interaction models (M6) found no evidence that IRSD modified associations between SWI and prevalence of sufficient walking (Δ DIC =8.2), sufficient MVPA (Δ DIC =22.9) or high MVPA (Δ DIC =22.7) to improve health. However, there was strong support for the simpler sufficient walking model M3 without IRSD compared to the best fitting M5 (see Table 4). The DIC for M3 increased by only 0.7, had fewer effective parameters, and reduced spatial variation by a further 28.3% relative to M5. There was also strong support for an association between SWI and prevalence of sufficient MVPA model M3 (see Table 5). This model had a slightly increased DIC compared to the best fitting model M2 (Δ DIC =0.5) but fewer effective parameters and reduced postal area clustering by an additional 47.4%.

The left-hand panels of Fig. 2 plot the residual geographic variation in unadjusted mean prevalence rate ratios (M1) for sufficient walking, sufficient MVPA, and high MVPA to improve health, which ranged from 0.79-1.56, 0.87-1.19, and 0.84-1.20, respectively. There is clear evidence for geographic clustering of postal areas with lower prevalence in southern Sydney and higher prevalence in eastern Sydney. Low rates of sufficient walking are also evident in central north Sydney, and clusters of higher rates of sufficient and high MVPA in outer western Sydney. The right-hand panels of Fig. 2 show residual prevalence rate ratios for fully-adjusted models (M5). Residual prevalence rate ratios for all outcomes were attenuated with ranges reduced to 0.90-1.25, 0.98-1.04, and 0.97-1.07 for sufficient walking, sufficient MVPA, and high MVPA, respectively. Despite these reductions, spatial fractions and disease maps indicated residual variance was principally spatial with higher prevalence on the eastern seaboard, and lower prevalence in central and southern Sydney. A north-south band of low prevalence is also evident for sufficient walking to improve health.

Fig. 2
figure 2

Excess prevalence ratios for physical activity outcomes in Sydney Statistical Division. Excess prevalence ratios were derived by exponentiating the sum of the log odds for the spatial and non-spatial random effects. Left panel maps report estimates from unadjusted models (Model 1) using the mean prevalence for the study area to calculate expected cases. Right panel maps report estimates from fully adjusted models including area-level walkability and relative socioeconomic disadvantage (Model 5) with expected cases derived from individual-level logistic regression models


To our knowledge this is the first study to report macro-level associations between walkability and MVPA to improve health using a representative large-scale cohort and geospatial methods. Our results provide support for a positive association between increasing postal area walkability and prevalence of sufficient walking to improve health; weaker support for a positive association between increasing postal area walkability and prevalence of sufficient MVPA to improve health; and no support for an association between postal area walkability and prevalence of high MVPA. These findings are independent of individual-level demographic, social, economic, and health factors, and area-level socioeconomic disadvantage. We also found geographic clustering in prevalence of all MVPA outcomes, with higher rates of sufficient walking, sufficient MVPA, and high MVPA to improve health in the central business district and adjacent east-coast areas, and lower rates through central and southern Sydney. Approximately half of this spatial variation is explained by postal area walkability for sufficient walking and sufficient MVPA to improve health. Taken together, our findings extend individual-level environment-behavior relations between walking to improve health and walkability to population and spatial scales typically used for health planning, intervention, and surveillance; highlight the utility of spatial analysis for informing walkability research and planning; and support the validity of undertaking geographical analysis on the 45 and Up Study cohort.

Walkability and moderate and vigorous physical activity

Syntheses of the research literature consistently report that residents in highly walkable neighborhoods are more likely to participate in MVPA, especially for active transportation, than residents in less walkable neighborhoods (see [10, 11, 8487]). Some of these reviews have additionally concluded that the evidence is sufficiently robust to recommend incorporating built environment factors, including walkability, into urban design, transportation, and health planning [84, 85]. However, this evidence base is largely derived using individual-level studies that measure walkability at micro-environmental spatial scales [28], which has raised concerns about its validity for population-level action [32]. Our results indicate that increasing macro-environmental walkability at the postal area level is positively associated with population-levels of sufficient walking to improve health. Prevalence of sufficient walking to improve health was 12-29% greater in high versus low walkable areas, and 1-8% greater in medium-high versus low walkable areas; no difference was observed between medium-low and low walkability areas. These results indicate that micro-level associations between walkability and walking manifest at macro-level spatial scales that are similar to those used for population health planning and intervention; support the validity of individual-level walkability evidence for informing population-level action to increase walking for health; and extends to middle and older aged populations, with previous research showing increased area-level walkability is associated with higher prevalence of walking in employed populations [12, 17, 34]. Our results also provide helpful information for targeting interventions to increase walking and walkability identified in the New South Wales State Government’s plan for growing Sydney [20].

We observed a monotonically increasing exposure-response gradient between postal area walkability and prevalence of sufficient walking to benefit health. The effect size for high versus low walkability areas was three and seven times greater than for medium-high and low-medium versus low areas, respectively. This suggests a threshold effect whereby high levels of environmental walkability are required to observe an association with population-levels of sufficient walking to improve health. In our study this equated to median values of 46.0 residential dwellings per hectare, 162.5 intersections per square kilometer, and a land use entropy mix of 0.13. We have previously raised the possibility of a walkability threshold in the Sydney metropolitan region [17], and Kelly et al. have recently reported macro-level journey to work results for North American populations consistent with a threshold effect [34]. However, we are unable to preclude the possibility that any threshold is due to scale effects, which can diminish associations as spatial granularity coarsens [88]. For example, Australian research has demonstrated that associations between walkability and individual-level walking for transport attenuate as walkability is measured at increasingly coarser spatial scales [15]. However, the similarity of our results to other macro-level studies conducted at finer geographic resolutions (e.g., [17, 34]), and our matching of outcome and exposure scales, provide some evidence against a spatial scale artefact.

We found support for an association between walkability and sufficient MVPA to improve health after adjusting for individual differences. This finding was somewhat unexpected as walkability indexes are typically specific for utilitarian walking [8, 12, 17], although findings have been increasingly mixed in recent years with both positive [89] and null [15, 90] results reported. Our finding possibly reflects the very high prevalence of walking in our cohort. Sufficient MVPA to improve health was reported by 69.4% of Sydney respondents of which 45.7% attained this from walking alone. A large subset of sufficiently active walkers may also explain other positive findings for MVPA reported in the walkability literature (e.g., [89]), and reaffirms the strategy of promoting moderate-intensity walking to increase population levels of sufficient MVPA to improve health [14].

Our results provide no evidence for a macro-level association between postal area walkability and prevalence of high MVPA to enhance health. We defined high MVPA as ≥300 min per week, which equates to approximately 60 minutes of moderate or 30 minutes of vigorous intensity activity on most days of the week. This outcome was chosen as it is thought to reflect the minimum MVPA required for prevention of weight gain and some cancers [53]. It is likely that persons meeting this threshold would do so through a combination of moderate and high intensity physical activity, and not by walking alone. Vigorous physical activity is most consistently associated with availability of home exercise equipment and convenience of nearby facilities [9194], which are distinctly different environmental factors to those underpinning walkability. As such, this study extends our previous findings [17] on the domain specificity of the SWI in employed populations to the general population aged 45 years and over living in Sydney, Australia.

Geographic variation in moderate and vigorous physical activity

We observed very high levels of geographic clustering for all MVPA outcomes in unadjusted spatial analysis with increased prevalence in the Sydney central business and surrounding east-coast areas, and reduced prevalence through central and southern Sydney. Postal area walkability explained 65.8% and 47.4% of residual geographic variation in sufficient walking and sufficient MVPA to improve health but only 15.5% of high MVPA. These associations were readily apparent in disease maps, and were substantially attenuated by the inclusion of postal area walkability in spatial models. We believe presenting population-level variation in MVPA as disease maps is likely to be especially helpful for identifying and targeting areas that may benefit from infrastructure upgrades or developments, and generating hypotheses for additional research. This is supported by primary health care research showing maps are readily comprehended by decision-makers, and can facilitate the alignment of services, and interventions with population needs [95].

Ours appears to be the first built environment study to quantitatively demonstrate associations between area-level walkability and spatial patterning in population-levels of sufficient walking and MVPA to improve health. Merom and colleagues have previously reported geographic variation in prevalence and increases in prevalence in any walking for Sydney local government areas between 2002–2012 [28]. They also observed highest prevalence in the Sydney business district and adjacent areas, and lowest prevalence through central and southern Sydney. Our findings expand on this research in a number of aspects. First, our study used fully Bayesian hierarchical models to account for, and leverage, spatial autocorrelation to produce “smoothed” effect estimates, valid credible intervals, and partition variation into spatially structured and unstructured components. Second, we used MVPA outcomes that are routinely used to monitor health status in Australian populations. Third, we conducted our study at a much finer spatial resolution in order to maximize between-area heterogeneity and increase the locational specificity of our results [96]. Fourth, we evaluated the proportion of excess spatial variation in MVPA outcomes that was attributable to postal area walkability after removing variation due to individual and area level demographic, social, economic, and health status factors. Together, these differences allowed us to produce robust disease maps over a shorter time horizon and identify regions where walkability may be contributing to population differentials in sufficient MVPA to improve health.

We also observed spatial clustering in high MVPA to improve health but this was unrelated to postal area walkability. Variation in high MVPA to enhance health is most consistently associated with availability of home exercise equipment and convenience of nearby facilities [9194]. There is also some evidence that higher densities of exercise facilities within 1,000 metres of an individual’s residence is associated with increased duration of MVPA and odds of meeting physical activity recommendations [97]. The extent to which the observed geographic variation in high MVPA in Sydney is attributable to the spatial distribution of these and other environmental factors was beyond the scope of our study but could easily be addressed by including density estimates into our analytic framework, and warrants further investigation given the hypothesised benefits of high MVPA for the prevention of weight gain and some cancers [53].

Excess prevalence for all outcomes decreased after adjusting for individual-level factors; however, despite these reductions disease maps remained highly clustered, especially for walking. This finding has important methodological implications because it suggests (1) spatial autocorrelation is an inherent feature of built environment data, and (2) individual differences do not fully explain this clustering. Spatial autocorrelation is a problem for linear regression because it violates the assumption that residuals are independent and identically distributed, which may lead to erroneous inference [98]. Concerns regarding ‘spatial multicollinearity’ in the walkability literature are not new but have focused on the covariation between built environment factors [99]. For example, NQLS-based walkability indexes aggregate data across environmental variables to minimize multicollinearity and leverage their ‘synergy’ [12]. However, this does not account for spatial autocorrelation in the distributions of outcome or study factors, which may be substantial based on our results.

Multilevel analysis using general and generalized linear mixed models (GLMM) provides one solution to account for spatial autocorrelation, and is already used widely in the built environment literature for individual-level analyses where walkability is measured at meso-levels [10]. There are two problems with this approach: first, researchers typically use GLMM only when walkability is measured at a level different to that used for inference, and are unlikely to consider more complex analytical models for apparently non-hierarchical designs; and second, GLMM most often employ a covariance structure that conflates spatial and non-spatial variation within a single variance component [100]. Our study highlights the utility of examining this spatial component for informing policy and planning activities. We therefore recommend the use of spatial models in built environment research to (1) make explicit the expectation of spatial structure in the environment-behavior data under investigation; and (2) identify geographic variation in outcomes to inform population-level programming. We believe the Bayesian disease mapping and ecological regression approach used in our study is especially useful in this regard because it incorporates both individual and area-level factors; is easily implemented in freely available statistical software; and will provide unbiased effect estimates in the absence of spatial variation [101].

Area-level socioeconomic disadvantage and Moderate and Vigorous Physical Activity

Area-level relative socioeconomic disadvantage was not identified as a correlate, confounder or effect modifier in any of our walking or MVPA models. This differs from our previous study that found median household income was independently associated with walking to work, and attenuated associations between SWI and walking to work at the smaller Census Collection District level in Sydney metropolitan region [17]. Kelly and colleagues have reported similar findings for block group prevalence of walking to work in the St. Louis City and County areas of North America [34]. These differences likely reflect methodological improvements in our current study, including adjustment for individual-level factors to account for heterogeneity in the demographic, social, economic, and health characteristics of postal area respondents. Our findings therefore extend empirical observations that area-level socioeconomic status is unrelated to individual-level walking and walkability after adjusting for person-level factors [90] to population-level associations between MVPA and walkability, and reaffirm the potential of walking to address population inequalities in MVPA participation [102].

Implications for policy and planning in Sydney

The Sydney environment will be transformed over the next 20 years through the Plan for Growing Sydney [20], which aims to accommodate a projected population increase of one million people by developing communities that are strong, healthy, and well connected. Creating local opportunities for transport-related walking through strategic land use and infrastructure developments is a key strategy of this plan [20]. Ensuring these developments and upgrades maximize transport, health, and environmental benefits will be a significant challenge for population health advocates. Our study indicates that participation in sufficient walking to improve health is not uniform across Sydney but varies geographically. This structure is independent of individual-level demographic, social, environmental, and economic factors, and unrelated to area-level socioeconomic disadvantage—factors often considered instrumental in urban design, transportation, and health policy and planning (e.g., [20, 103]). The SWI has the potential to inform these decisions by characterizing the walkability of geographic areas with lower than expected MVPA participation rates with a view to prioritizing infrastructure upgrades and developments that support active transportation and walking for other purposes [8, 20].

Strengths and limitations

This study has a number of strengths. First, we linked the SWI to high-quality and geocoded baseline data from the 45 and Up Study, which allowed us to examine area-level associations between walking infrastructure and population-levels of MVPA adjusting for individual-level factors. The 45 and Up Study is a prospective cohort with approximately quinquennial follow-up [47]. As these follow-up data are collected, geocoded, and made available to researchers, they will provide unique opportunities to examine the influence of walkability on the walking behavior and health outcomes of individuals and populations with increasingly sophisticated designs [17]. Second, we measured walkability using the SWI [17], which is derived from NQLS [12] and PLACE [8] walkability indexes. These indexes form the basis of an extensive national and international literature linking increased environmental walkability to individual-levels of sufficient moderate-intensity physical activity to enhance health [17], and provides an international context for our findings and methodological approach for population-level programming and built environment research. Third, we used a Bayesian disease mapping and ecological regression study design that allowed us to quantify the geographic variation in MVPA outcomes attributable to postal area walkability after removing variation explained by other individual and area-level factors. Our approach appears novel in the walkability literature despite its common use in epidemiology for small area estimation problems [104, 105]. A particular advantage of this approach is the capacity to produce smoothed disease maps to communicate variation in geographic risk to politicians, planners, and policymakers. Fourth, we used a spatial scale more proximal to those typically used for population health planning and status monitoring [28] but within the upper range of buffers used for individual-level analyses [56, 57]. However, we caution against interpreting area-level walkability as simply an average of individual-level exposures within the areal unit. Area-level walkability is derived at the spatial scale of analysis. It is a contextual measure of the area’s built environment to which groups, communities and populations are exposed. That is, we maintain area-level walkability is a characteristic of the areal unit and measures an aspect of walkability that is qualitatively different to individual-level walkability. Finally, we observed that unweighted prevalence ratios used in our spatial analyses were analogous to those derived using post-stratification weighting, which provides support for the generalizability of our findings and spatial trends to the Sydney Statistical Division.

Our study is also subject to a number of limitations. First, our study sample included all Sydney respondents to the 45 and Up Study with complete data on selected baseline survey items. Although the 45 and Up Study includes approximately 10% of the NSW Population aged ≥45 years, the cohort is younger, more highly educated, and more likely to speak English at home and live with a partner than the Sydney study base [43]. Our sub-sample reflected these cohort characteristics. Point estimates may be biased by non-response in cohort studies [106, 107]; however, relative measures of effect are generally considered representative of the study base [108112]. The external validity of 45 and Up Study results is supported by research showing that relative effect measures derived from this cohort are consistent with those from population-representative surveys [79], irrespective of sample weighting. The very high correlations between unweighted and weighted postal area prevalence estimates described in our study are consistent with these observations, and provide support for the external validity of our findings.

Second, we used a two-step approach to model associations between postal area physical activity and walkability using ecological spatial regressions that were adjusted for person-level factors by including model offsets derived from individual-level regressions. This approach is often used for spatial analyses where area and individual-level factors cannot be incorporated into a single parsimonious model (see [6668, 113]); however, both individual and area-levels factors would ideally be modeled concurrently. These types of multilevel spatial models are starting to appear in the methodological literature (see [114, 115]) but are not yet easily implemented in standard statistical packages and require distributed computational environments for large data problems [115], which prohibited their use in our study.

Third, although widely used for environment-behavior research, NQLS-based indexes are derived using population-specific cut-points for defining walkability quartiles, which can result in data-dependent exposure categories [116118]. To increase the utility of our findings for planners, policy analysts, and researchers, we have reported the median, minimum, and maximum environmental variable scores associated with walkability quartiles in our study area. These values compare favorably with the limited number of studies that also report environmental values within walkability quartiles (see [21, 54, 119]), and provides support for the international relevance of our findings. A small number of recent studies have directly modeled associations between NQLS environmental variables and minutes spent in moderate-intensity physical activity (e.g., [18, 120, 121]). These studies employed non-parametric generalized additive models (GAM) with spline functions to account for the complex, non-linear relationships between exposure and outcome variables. This was beyond the scope and resources of our study, which would have required fitting computationally intensive penalized spline terms to our already resource-demanding conditional autoregressive models. We therefore recommend that analysts and researchers carefully consider the concordance of built environment characteristics between their target environments and the Sydney metropolitan area before utilising our findings for population-level policy, planning, and intervention.

Fourth, we were unable to evaluate if our results were sensitive to the spatial scale at which area-level walkability is measured because our access to geocoded data was limited to postal area identifiers. The modifiable areal unit problem has the potential to affect all spatial analyses that do not have access to individual-level longitude and latitude coordinates [122]. However, the concordance of our results with previous macro-level studies conducted at finer spatial resolutions (see [12, 17, 34]) supports the robustness of our findings. Fifth, postal areas are Australian Census statistical output units, so their geographic dimensions and populace may not be representative of planning geographies used in other countries. Finally, we used a cross-sectional design which precluded considerations of causality.


Our study appears to be unique in the walkability literature for its population focus and spatial approach. We observed that increasing postal area walkability was associated with higher prevalence of sufficient walking and sufficient MVPA to benefit health in Sydney, and accounts for large proportions of the residual geographic variation in these outcomes that remains after adjusting for individual and area-level demographic, social, economic, and health factors. Our results support the potential of walkability indexes to inform and target population-level programming, especially if local context is incorporated into these activities, and affirms the importance of including “place” in walkability research and planning to ensure the robustness of outcomes. From a practical perspective, our study demonstrates the utility of disease maps for communicating adverse geographic risk of MVPA outcomes and the extent to which this may be attributable to modifiable environmental factors such as walkability. Finally, our findings provide another resource for the NSW Government to use in sustainably growing Sydney by identifying regional opportunities for strategic land use and infrastructure developments to increase population-levels of walking and create built environments that support health.



Confidence interval


Credible interval


Deviance information criterion


Generalized linear mixed models


Index of relative socioeconomic disadvantage


Models 1–6


Moderate and vigorous physical activity


Neighborhood quality of life study


New South Wales


Effective parameters


Prevalence ratio


Social, environmental and economic factors


Sydney Walkability Index


  1. Bull F, Schöppe S, Bauman A, Weston E. International review of policies on physical activity In: Bull F, Bauman A, Bellew B, Brown W, editors. Getting Australia Active II: an Update of Evidence on Physical Activity for Health, Chap. 4. Melbourne: National Public Health Partnership: 2004. p. 69–85.

    Google Scholar 

  2. Lee IM, Buchner DM. The importance of walking to public health. Med Sci Sports Exerc. 2008; 40(7 Suppl):512–8. doi:10.1249/MSS.0b013e31817c65d0.

    Article  Google Scholar 

  3. Sallis JF, Frank LD, Saelens BE, Kraft MK. Active transportation and physical activity: opportunities for collaboration on transportation and public health research. Transport Res Pol Pract. 2004; 38(4):249–68. doi:10.1016/j.tra.2003.11.003.

    Article  Google Scholar 

  4. Hamer M, Chida Y. Active commuting and cardiovascular risk: a meta-analytic review. Prev Med. 2008; 46(1):9–13. doi:10.1016/j.ypmed.2007.03.006.

    Article  PubMed  Google Scholar 

  5. Hootman JM, Macera CA, Ainsworth BE, Martin M, Addy CL, Blair SN. Association among physical activity level, cardiorespiratory fitness, and risk of musculoskeletal injury. Am J Epidemiol. 2001; 154(3):251–8. doi:10.1093/aje/154.3.251.

    Article  CAS  PubMed  Google Scholar 

  6. Bauman AE. Updating the evidence that physical activity is good for health: an epidemiological review 2000-2003. J Sci Med Sport. 2004; 7(1 Suppl 1):6–19. doi:10.1016/S1440-2440(04)80273--1.

    Article  CAS  PubMed  Google Scholar 

  7. Saelens BE, Sallis JF, Frank LD. Environmental correlates of walking and cycling: findings from the transportation, urban design, and planning literatures. Ann Behav Med. 2003; 25(2):80–91.

    Article  PubMed  Google Scholar 

  8. Leslie E, Coffee N, Frank L, Owen N, Bauman A, Hugo G. Walkability of local communities: using geographic information systems to objectively assess relevant environmental attributes. Health Place. 2007; 13(1):111–22. doi:10.1016/j.healthplace.2005.11.001.

    Article  PubMed  Google Scholar 

  9. Giles-Corti B, Macaulay G, Middleton N, Boruff B, Bull F, Butterworth I, Badland H, Mavoa S, Roberts R, Christian H. Developing a research and practice tool to measure walkability: a demonstration project. Health Promot J Austr. 2014; 25(3):160–6. doi:10.1071/HE14050.

    Article  PubMed  Google Scholar 

  10. Saelens BE, Handy SL. Built environment correlates of walking: a review. Med Sci Sports Exerc. 2008; 40(7):550–66.

    Article  Google Scholar 

  11. Ewing R, Cervero R. Travel and the built environment. J Am Plann Assoc. 2010; 76(3):265–94.

    Article  Google Scholar 

  12. Frank LD, Sallis JF, Saelens BE, Leary L, Cain K, Conway TL, Hess PM. The development of a walkability index: application to the Neighborhood Quality of Life Study. Br J Sports Med. 2010; 44(13):924–33. doi:10.1136/bjsm.2009.058701.

    Article  CAS  PubMed  Google Scholar 

  13. Adams MA, Frank LD, Schipperijn J, Smith G, Chapman J, Christiansen LB, Coffee N, Salvo D, du Toit L, Dygryn J, Hino AA, Lai PC, Mavoa S, Pinzon JD, Van de Weghe N, Cerin E, Davey R, Macfarlane D, Owen N, Sallis JF. International variation in neighborhood walkability, transit, and recreation environments using geographic information systems: the IPEN adult study. Int J Health Geogr. 2014; 13:43.

    Article  PubMed  PubMed Central  Google Scholar 

  14. Hankey S, Marshall JD, Brauer M. Health impacts of the built environment: within-urban variability in physical inactivity, air pollution, and ischemic heart disease mortality. Environ Health Perspect. 2012; 120(2):247–53.

    Article  CAS  PubMed  Google Scholar 

  15. Learnihan V, Van Niel KP, Giles-Corti B, Knuiman M. Effect of scale on the links between walking and urban design. Geogr Res-Aust. 2011; 49(2):183–91. doi:10.1111/j.1745-5871.2011.00689.x.

    Article  Google Scholar 

  16. Marshall JD, Brauer M, Frank LD. Healthy neighborhoods: walkability and air pollution. Environ Health Perspect. 2009; 117(11):1752–9. doi:10.1289/ehp.0900595.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  17. Mayne D, Morgan G, Willmore A, Rose N, Jalaludin B, Bambrick H, Bauman A. An objective index of walkability for research and planning in the Sydney metropolitan region of New South Wales, Australia: an ecological study. Int J Health Geogr. 2013; 12(1):61.

    Article  PubMed  PubMed Central  Google Scholar 

  18. Sallis JF, Cerin E, Conway TL, Adams MA, Frank LD, Pratt M, Salvo D, Schipperijn J, Smith G, Cain KL, Davey R, Kerr J, Lai PC, Mitas J, Reis R, Sarmiento OL, Schofield G, Troelsen J, Van Dyck D, De Bourdeaudhuij I, Owen N. Physical activity in relation to urban environments in 14 cities worldwide: a cross-sectional study. Lancet. 2016; 387(10034):2207–17. doi:10.1016/S0140-6736(15)01284-2.

    Article  PubMed  Google Scholar 

  19. Stockton JC, Duke-Williams O, Stamatakis E, Mindell JS, Brunner EJ, Shelton NJ. Development of a novel walkability index for London, United Kingdom: cross-sectional application to the Whitehall II Study. BMC Public Health. 2016; 16(1):1–12. doi:10.1186/s12889-016-3012-2.

    Article  Google Scholar 

  20. NSW Department of Planning and Environment. A Plan for Growing Sydney. Sydney: NSW Government; 2014.

    Google Scholar 

  21. Frank LD, Schmid TL, Sallis JF, Chapman J, Saelens BE. Linking objectively measured physical activity with objectively measured urban form: findings from SMARTRAQ. Am J Prev Med. 2005; 28(2 Suppl 2):117–25.

    Article  PubMed  Google Scholar 

  22. Rose G. Sick individuals and sick populations. Int J Epidemiol. 2001; 30(3):427–32.

    Article  CAS  PubMed  Google Scholar 

  23. Giles-Corti B, Timperio A, Bull F, Pikora T. Understanding physical activity environmental correlates: increased specificity for ecological models. Exerc Sport Sci Rev. 2005; 33(4):175–81.

    Article  PubMed  Google Scholar 

  24. Sallis JF, Bauman A, Pratt M. Environmental and policy interventions to promote physical activity. Am J Prev Med. 1998; 15(4):379–97. doi:

    Article  CAS  PubMed  Google Scholar 

  25. Ewing R, Cervero R. Travel and the built environment: a synthesis. Transp Res Rec. 2001; 1780:87–114. doi:10.3141/1780-10.

    Article  Google Scholar 

  26. Bauman AE, Sallis JF, Owen N. Environmental and policy measurement in physical research In: Welk G, editor. Physical Activity Assessments for Health-related Research. Champaign: Human Kinetics: 2002. p. 241–51.

    Google Scholar 

  27. Badland H, Schofield G. Transport, urban design, and physical activity: an evidence-based update. Transport Res Transport Environ. 2005; 10(3):177–96. doi:10.1016/j.trd.2004.12.001.

    Article  Google Scholar 

  28. Merom D, Ding D, Corpuz G, Bauman A. Walking in Sydney: trends in prevalence by geographic areas using information from transport and health surveillance systems. J Transp Health. 2015; 2(3):350–9. doi:10.1016/j.jth.2015.04.006.

    Article  Google Scholar 

  29. Forsyth A, Krizek KJ. Promoting walking and bicycling: assessing the evidence to assist planners. Built Environ. 2010; 36(4):429–46.

    Article  Google Scholar 

  30. Laine J, Kuvaja-Köllner V, Pietilä E, Koivuneva M, Valtonen H, Kankaanpää E. Cost-effectiveness of population-level physical activity interventions: a systematic review. Am J Health Promot. 2014; 29(2):71–80.

    Article  PubMed  Google Scholar 

  31. Panter JR, Jones A. Attitudes and the environment as determinants of active travel in adults: what do and don’t we know?. J Phys Act Health. 2010; 7(4):551–61.

    Article  PubMed  Google Scholar 

  32. Sato M, Du J, Inoue Y. Rate of physical activity and community health: evidence from U.S, counties. J Phys Act Health. 2016; 13(6):640–8. doi:10.1123/jpah.2015-0399.

    Article  PubMed  Google Scholar 

  33. Oftedal A, Schneider IE. Outdoor recreation availability, physical pctivity, and health outcomes: county-level analysis in Minnesota. J Park Recreat Admi. 2013; 31(1):34–56.

    Google Scholar 

  34. Kelly CM, Lian M, Struthers J, Kammrath A. Walking to work: the roles of neighborhood walkability and socioeconomic deprivation. J Phys Act Health. 2015; 12 Suppl 1:70–5. doi:10.1123/jpah.2012-0359.

    Article  Google Scholar 

  35. Rezaeian M, Dunn G, St Leger S, Appleby L. Geographical epidemiology, spatial analysis and geographical information systems: a multidisciplinary glossary. J Epidemiol Commun Health. 2007; 61(2):98–102. doi:10.1136/jech.2005.043117.

    Article  Google Scholar 

  36. Pattenden S, Casson K, Cook S, Dolk H. Geographical variation in infant mortality, stillbirth and low birth weight in Northern Ireland, 1992-2002. J Epidemiol Commun Health. 2011; 65(12):1159–65.

    Article  Google Scholar 

  37. Miranda ML, Edwards SE, Keating MH, Paul CJ. Making the environmental justice grade: the relative burden of air pollution exposure in the United States. Int J Environ Res Public Health. 2011; 8(6):1755.

    Article  PubMed  PubMed Central  Google Scholar 

  38. Huang L, Stinchcomb DG, Pickle LW, Dill J, Berrigan D. Identifying clusters of active transportation using spatial scan statistics. Am J Prev Med. 2009; 37(2):157–66. doi:10.1016/j.amepre.2009.04.021.

    Article  PubMed  PubMed Central  Google Scholar 

  39. Tamura K, Puett RC, Hart JE, Starnes HA, Laden F, Troped PJ. Spatial clustering of physical activity and obesity in relation to built environment factors among older women in three U.S, states. BMC Public Health. 2014; 14(1):1–16. doi:10.1186/1471-2458-14-1322.

    Article  Google Scholar 

  40. Australian Bureau of Statistics. Statistical Geography Volume 1: Australian Standard Geographical Classification (ASGC), July 2006 (catalogue No. 1216.0). Canberra: Commonwealth of Australia; 2006.

    Google Scholar 

  41. Australian Bureau of Statistics. CData Online. 2006. Accessed 19 Dec 2015.

  42. Australian Bureau of Statistics. Statistical Geography: Volume 2 - Census Geographic Areas, 2006 (catalogue No. 2905.0). Canberra: Commonwealth of Australia; 2007.

    Google Scholar 

  43. 45 and Up Study Collaborators. Cohort profile: the 45 and Up Study. Int J Epidemiol. 2008; 37(5):941–7. doi:10.1093/ije/dym184.

    Article  Google Scholar 

  44. Watts G. Uk biobank gets 10% response rate as it starts recruiting volunteers. BMJ. 2007; 334(7595):659–9. doi:10.1136/bmj.39167.407616.DB.

    Article  PubMed  PubMed Central  Google Scholar 

  45. Knies G, Burton J. Analysis of four studies in a comparative framework reveals: health linkage consent rates on british cohort studies higher than on uk household panel surveys. BMC Med Res Methodol. 2014; 14(1):125. doi:10.1186/1471-2288-14-125.

    Article  PubMed  PubMed Central  Google Scholar 

  46. Galea S, Tracy M. Participation rates in epidemiologic studies. Ann Epidemiol. 2007; 17(9):643–53. doi:10.1016/j.annepidem.2007.03.013.

    Article  PubMed  Google Scholar 

  47. The 45 and Up Study. The 45 and Up Study data book – April 2010 release. 2010. Accessed 16 Dec 2015.

  48. Australian Bureau of Statistics. Socio-Economic Indexes for Areas (SEIFA) - technical paper, 2006: Australian Bureau of Statistics; 2008. Accessed 26 March 2008.

  49. Australian Institute of Health and Welfare. The Active Australia Survey: a guide and manual for implementation, analysis and reporting. Canberra: AIHW; 2003.

    Google Scholar 

  50. Armstrong T, Bauman A, Davies J. Physical activity patterns of Australian adults. Results of the 1999 national physical activity survey.Canberra: Australian Institute of Health and Welfare; 2000.

    Google Scholar 

  51. Department of Health. Australia’s physical activity and sedentary behaviour guidelines: Commonwealth of Australia; 2014. Accessed 04 Dec 2015.

  52. Physical Activity Guidelines Advisory Committee. Physical Activity Guidelines Advisory Committee report, 2008: U.S. Department of Health and Human Services; 2008. Accessed 06 Jan 2016.

  53. Brown WJ, Bauman AE, Bull FC, Burton NW. Development of evidence-based physical activity recommendations for adults (18-64 years). Report prepared for the Australian Government Department of Health. 2012. Accessed 06 Jan 2016.

  54. Duncan M, Winkler E, Sugiyama T, Cerin E, duToit L, Leslie E, Owen N. Relationships of land use mix with walking for transport: do land uses and geographical scale matter?. J Urban Health. 2010; 87(5):782–95. doi:10.1007/s11524-010-9488-7.

    Article  PubMed  PubMed Central  Google Scholar 

  55. Kerr J, Sallis JF, Owen N, Bourdeaudhuij ID, Cerin E, Sugiyama T, Reis R, Sarmiento O, Frömel K, Mitáš J, Troelsen J, Christiansen LB, Macfarlane D, Salvo D, Schofield G, Badland H, Guillen-Grima F, Aguinaga-Ontoso I, Davey R, Bauman A, Saelens B, Riddoch C, Ainsworth B, Pratt M, Schmidt T, Frank L, Adams M, Conway T, Cain K, Dyck DV, Bracy N. Advancing science and policy through a coordinated international study of physical activity and built bvironments: Ipen adult methods. J Phys Act Health. 2013; 10(4):581–601.

    Article  PubMed  Google Scholar 

  56. Villanueva K, Knuiman M, Nathan A, Giles-Corti B, Christian H, Foster S, Bull F. The impact of neighborhood walkability on walking: Does it differ across adult life stage and does neighborhood buffer size matter?. Health Place. 2014; 25(0):43–6. doi:10.1016/j.healthplace.2013.10.005.

    Article  PubMed  Google Scholar 

  57. James P, Berrigan D, Hart JE, Aaron Hipp J, Hoehner CM, Kerr J, Major JM, Oka M, Laden F. Effects of buffer size and shape on associations between the built environment and energy balance. Health Place. 2014; 27:162–70. doi:10.1016/j.healthplace.2014.02.003.

    Article  PubMed  PubMed Central  Google Scholar 

  58. Sugiyama T, Salmon J, Dunstan DW, Bauman AE, Owen N. Neighborhood walkability and TV viewing time among Australian adults. Am J Prev Med. 2007; 33(6):444–9. doi:10.1016/j.amepre.2007.07.035.

    Article  PubMed  Google Scholar 

  59. Bauman AE, Reis RS, Sallis JF, Wells JC, Loos RJF, Martin BW. Correlates of physical activity: why are some people physically active and others not?. Lancet. 2012; 380(9838):258–71. doi:10.1016/S0140-6736(12)60735-1.

    Article  PubMed  Google Scholar 

  60. George ES, Jorm L, Kolt GS, Bambrick H, Lujic S. Physical activity and psychological distress in older men: findings from the New South Wales 45 and up study. J Aging Phys Act. 2012; 20(3):300–16.

    Article  PubMed  Google Scholar 

  61. Astell-Burt T, Feng X, Kolt GS. Green space is associated with walking and moderate-to-vigorous physical activity (MVPA) in middle-to-older-aged adults: findings from 203 883 Australians in the 45 and Up Study. Br J Sports Med. 2013. doi:10.1136/bjsports-2012-092006.

  62. Korda RJ, Paige E, Yiengprugsawan V, Latz I, Friel S. Income-related inequalities in chronic conditions, physical functioning and psychological distress among older people in Australia: cross-sectional findings from the 45 and Up Study. BMC Public Health. 2014; 14(1):1–10. doi:10.1186/1471-2458-14-741.

    Article  Google Scholar 

  63. Gebel K, Ding D, Chey T, Stamatakis E, Brown WJ, Bauman AE. Effect of moderate to vigorous physical activity on all-cause mortality in middle-aged and older Australians. JAMA Intern Med. 2015; 175(6):970–7. doi:10.1001/jamainternmed.2015.0541.

    Article  PubMed  Google Scholar 

  64. Stamatakis E, Rogers K, Ding D, Berrigan D, Chau J, Hamer M, Bauman A. All-cause mortality effects of replacing sedentary time with physical activity and sleeping using an isotemporal substitution model: a prospective study of 201,129 mid-aged and older adults. Int J Behav Nutr Phys Act. 2015; 12:121. doi:10.1186/s12966-015-0280-7.

    Article  PubMed  PubMed Central  Google Scholar 

  65. Hays RD, Sherbourne CD, Mazel RM. The RAND 36-item health survey 1.0. Health Econ. 1993; 2(3):217–27. doi:10.1002/hec.4730020305.

    Article  CAS  PubMed  Google Scholar 

  66. Leroux BG, Lei X, Breslow N. Estimation of disease rates in small areas: A new mixed model for spatial dependence In: Halloran ME, Berry D, editors. Statistical Models in Epidemiology, the Environment, and Clinical Trials. New York: Springer: 2000. p. 179–91. doi:10.1007/978-1-4612-1284-3.

    Google Scholar 

  67. Klassen AC, Kulldorff M, Curriero F. Geographical clustering of prostate cancer grade and stage at diagnosis, before and after adjustment for risk factors. Int J Health Geogr. 2005; 4(1):1. doi:10.1186/1476-072X-4-1.

    Article  PubMed  PubMed Central  Google Scholar 

  68. Waldhoer T, Wald M, Heinzl H. Analysis of the spatial distribution of infant mortality by cause of death in Austria in 1984 to 2006. Int J Health Geogr. 2008; 7:21. doi:10.1186/1476-072X-7-21.

    Article  PubMed  PubMed Central  Google Scholar 

  69. Besag J, York J, Mollié A. Bayesian image restoration, with two applications in spatial statistics. Ann Inst Stat Math. 1991; 43(1):1–20. doi:10.1007/BF00116466.

    Article  Google Scholar 

  70. Lawson A, Browne WJ, Vidal Rodeiro CL. Disease Mapping with WinBUGS and MLwiN. Statistics in Practice. Hoboken: Wiley; 2003.

    Book  Google Scholar 

  71. Goovaerts P, Gebreab S. How does poisson kriging compare to the popular BYM model for mapping disease risks?. Int J Health Geogr. 2008; 7(1):1–25. doi:10.1186/1476-072x-7-6.

    Article  Google Scholar 

  72. Waller L, Carlin B. Disease mapping In: Gelfand AE, Diggle PJ, Feuentes M, Guttorp P, editors. Handbook of Spatial Statistics, Chapman & Hall/CRC handbooks of modern statistical methods. Boca Raton: CRC Press: 2010. p. 217–44.

    Google Scholar 

  73. Lunn D, Jackson C, Best N, Thomas A, Spiegelhalter D. The BUGS Book: a practical introduction to Bayesian analysis. Texts in statistical science. Boca Raton: CRC Press; 2012.

    Google Scholar 

  74. Gelman A, Rubin DB. Inference from iterative simulation using multiple sequences. Stat Sci. 1992; 7(4):457–511.

    Article  Google Scholar 

  75. van der Linde A. Dic in variable selection. Statistica Neerlandica. 2005; 59(1):45–56. doi:10.1111/j.1467-9574.2005.00278.x.

    Article  Google Scholar 

  76. Spiegelhalter DJ, Best NG, Carlin BP, Van Der Linde A. Bayesian measures of model complexity and fit. J R Stat Soc Series B Stat Methodol. 2002; 64(4):583–639.

    Article  Google Scholar 

  77. Cramb SM, Mengersen KL, Baade PD. Developing the atlas of cancer in Queensland: methodological issues. Int J Health Geogr. 2011; 10:9. doi:10.1186/1476-072X-10-9.

    Article  PubMed  PubMed Central  Google Scholar 

  78. Holowaty EJ, Norwood TA, Wanigaratne S, Abellan JJ, Beale L. Feasibility and utility of mapping disease risk at the neighbourhood level within a Canadian public health unit: an ecological study. Int J Health Geogr. 2010; 9:21. doi:10.1186/1476-072X-9-21.

    Article  PubMed  PubMed Central  Google Scholar 

  79. Mealing NM, Banks E, Jorm LR, Steel DG, Clements MS, Rogers KD. Investigation of relative risk estimates from studies of the same population with contrasting response rates and designs. BMC Med Res Methodol. 2010; 10(1):1–12. doi:10.1186/1471-2288-10-26.

    Article  Google Scholar 

  80. Gelman A. Struggles with survey weighting and regression modeling. Statist Sci. 2007; 22(2):153–64. doi:10.1214/088342306000000691.

    Article  Google Scholar 

  81. Chen C, Wakefield J, Lumely T. The use of sampling weights in bayesian hierarchical models for small area estimation. Spatial Spatio-temporal Epidemiol. 2014; 11:33–43. doi:10.1016/j.sste.2014.07.002.

    Article  Google Scholar 

  82. Oleson JJ. Hierarchical bayesian modeling in dichotomous processes in the presence of nonresponse. Biometrics; 60(1):50–59. doi:10.1111/j.0006-341X.2004.00153.x.

  83. Australian Bureau of Statistics. TableBuilder Basic. 2016. Accessed 03 Feb 2016.

  84. Lee C, Moudon AV. Physical activity and environment research in the health field: implications for urban and transportation planning practice and research. J Plan Lit. 2004; 19(2):147–81. doi:10.1177/0885412204267680.

    Article  Google Scholar 

  85. Heath GW, Brownson RC, Kruger J, Miles R, Powell KE, Ramsey LT, Task Force on Community Preventive Services. The effectiveness of urban design and land use and transport policies and practices to increase physical activity: a systematic review. J Phys Act Health. 2006; 3(Suppl 1):55–76.

    Article  Google Scholar 

  86. Renalds A, Smith TH, Hale PJ. A systematic review of built environment and health. Fam Commun Health. 2010; 33(1):68–78. doi:10.1097/FCH.0b013e3181c4e2e5.

    Article  Google Scholar 

  87. Durand CP, Andalib M, Dunton GF, Wolch J, Pentz MA. A systematic review of built environment factors related to physical activity and obesity risk: implications for smart growth urban planning. Obes Rev. 2011; 12(5):173–82. doi:10.1111/j.1467-789X.2010.00826.x.

    Article  Google Scholar 

  88. Schuurman N, Bell N, Dunn JR, Oliver L. Deprivation indices, population health and geography: an evaluation of the spatial effectiveness of indices at multiple scales. J Urban Health. 2007; 84(4):591–603. doi:10.1007/s11524-007-9193-3.

    Article  PubMed  PubMed Central  Google Scholar 

  89. Sundquist K, Eriksson U, Kawakami N, Skog L, Ohlsson H, Arvidsson D. Neighborhood walkability, physical activity, and walking behavior: the Swedish Neighborhood and Physical Activity (SNAP) study. Soc Sci Med. 2011; 72(8):1266–73. doi:10.1016/j.socscimed.2011.03.004.

    Article  PubMed  Google Scholar 

  90. Owen N, Cerin E, Leslie E, duToit L, Coffee N, Frank LD, Bauman AE, Hugo G, Saelens BE, Sallis JF. Neighborhood walkability and the walking behavior of Australian adults. Am J Prev Med. 2007; 33(5):387–95. doi:10.1016/j.amepre.2007.07.025.

    Article  PubMed  Google Scholar 

  91. Hovell MF, Sallis JF, Hofstetter CR, Spry VM, Faucher P, Caspersen CJ. Identifying correlates of walking for exercise: an epidemiologic prerequisite for physical activity promotion. Prev Med. 1989; 18(6):856–66. doi:10.1016/0091-7435(89)90021-2.

    Article  CAS  PubMed  Google Scholar 

  92. Sallis JF, Hovell MF, Richard Hofstetter C. Predictors of adoption and maintenance of vigorous physical activity in men and women. Prev Med. 1992; 21(2):237–51. doi:10.1016/0091-7435(92)90022-A.

    Article  CAS  PubMed  Google Scholar 

  93. Sallis JF, Johnson MF, Calfas KJ, Caparosa S, Nichols JF. Assessing perceived physical environmental variables that may influence physical activity. Res Q Exerc Sport. 1997; 68(4):345–51.

    Article  CAS  PubMed  Google Scholar 

  94. De Bourdeaudhuij I, Sallis JF, Saelens BE. Environmental correlates of physical activity in a sample of Belgian adults. Am J Health Promot. 2003; 18(1):83–92. doi:10.4278/0890-1171-18.1.83.

    Article  PubMed  Google Scholar 

  95. Bazemore A, Phillips RL, Miyoshi T. Harnessing geographic information systems (GIS) to enable community-oriented primary care. J Am Board Fam Med. 2010; 23(1):22–31. doi:10.3122/jabfm.2010.01.090097.

    Article  PubMed  Google Scholar 

  96. Tatem AJ, Adamo S, Bharti N, Burgert CR, Castro M, Dorelien A, Fink G, Linard C, John M, Montana L, Montgomery MR, Nelson A, Noor AM, Pindolia D, Yetman G, Balk D. Mapping populations at risk: improving spatial demographic data for infectious disease modeling and metric derivation. Popul Health Metr. 2012; 10(1):1–14. doi:10.1186/1478-7954-10-8.

    Article  Google Scholar 

  97. Eriksson U, Arvidsson D, Sundquist K. Availability of exercise facilities and physical activity in 2,037 adults: cross-sectional results from the Swedish Neighborhood and Physical Activity (SNAP) study. BMC Public Health. 2012; 12(1):1–9. doi:10.1186/1471-2458-12-607.

    Article  Google Scholar 

  98. Fotheringham AS, Brunsdon C, Charlton M. Quantitative geography: perspectives on spatial data analysis. London: Sage Publications; 2000, p. 270.

    Google Scholar 

  99. Nelson NM, Wright A, Lowry RG, Mutrie N. Where is the theoretical basis for understanding and measuring the environment for physical activity?. Environ Health Insights. 2008; 2:111–6.

    CAS  PubMed  PubMed Central  Google Scholar 

  100. Langford IH, Leyland AH, Rasbash J, Goldstein H. Multilevel modelling of the geographical distributions of diseases. J R Stat Soc Ser C Appl Stat. 1999; 48(2):253–68. doi:10.1111/1467-9876.00153.

    Article  CAS  PubMed  Google Scholar 

  101. Latouche A, Guihenneuc-Jouyaux C, Girard C, Hémon D. Robustness of the BYM model in absence of spatial variation in the residuals. Int J Health Geogr. 2007; 6(1):1–8. doi:10.1186/1476-072x-6-39.

    Article  Google Scholar 

  102. Siegel PZ, Brackbill RM, Heath GW. The epidemiology of walking for exercise: implications for promoting activity among sedentary groups. Am J Public Health. 1995; 85(5):706–10.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  103. NSW Ministry of Health. NSW Healthy Eating and Active Living Strategy: Preventing Overweight and Obesity in New South Wales 2013-2018. Sydney: NSW Ministry of Health; 2013.

    Google Scholar 

  104. Best N, Richardson S, Thomson A. A comparison of Bayesian spatial models for disease mapping. Stat Methods Med Res. 2005; 14(1):35–59.

    Article  PubMed  Google Scholar 

  105. Beale L, Hodgson S, Abellan JJ, Lefevre S, Jarup L. Evaluation of spatial relationships between health and the environment: the rapid inquiry facility. Environ Health Perspect. 2010; 118(9):1306–12. doi:10.1289/ehp.0901849.

    Article  PubMed  PubMed Central  Google Scholar 

  106. Van Goor H, Verhage A. Nonresponse and recall errors in a study of absence because of illness: an analysis of their effects on distributions and relationships. Qual Quant. 1999; 33(4):411–28. doi:10.1023/A:1004732502598.

    Article  Google Scholar 

  107. Burkell J. The dilemma of survey nonresponse. Libr Inf Sci Res. 2003; 25(3):239–63. doi:10.1016/S0740-8188(03)00029-X.

    Article  Google Scholar 

  108. Criqui MH. Response bias and risk ratios in epidemiologic studies. Am J Epidemiol. 1979; 109(4):394–9.

    Article  CAS  PubMed  Google Scholar 

  109. Miettinen OS. Theoretical Epidemiology: Principles of Occurrence Research in Medicine. New York: John Wiley; 1985.

    Google Scholar 

  110. Ponsonby AL, Dwyer T, Couper D. Is this finding relevant? generalisation and epidemiology. Aust N Z J Public Health. 1996; 20(1):54–6.

    Article  CAS  PubMed  Google Scholar 

  111. Nohr EA, Frydenberg M, Henriksen TB, Olsen J. Does low participation in cohort studies induce bias?. Epidemiology. 2006; 17(4):413–8. doi:10.1097/01.ede.0000220549.14177.60.

    Article  PubMed  Google Scholar 

  112. Willett WC, Blot WJ, Colditz GA, Folsom AR, Henderson BE, Stampfer MJ. Merging and emerging cohorts: not worth the wait. Nature. 2007; 445(7125):257–8. doi:10.1038/445257a.

    Article  CAS  PubMed  Google Scholar 

  113. Kulldorff M. SaTScan™user guide for version 9.4. 2015. Accessed 01 May 2017.

  114. Jackson C, Best N, Richardson S. Hierarchical related regression for combining aggregate and individual data in studies of socio-economic disease risk factors. J R Stat Soc Series A (Stat Soc). 2008; 171(1):159–78. doi:10.1111/j.1467-985X.2007.00500.x.

    Google Scholar 

  115. Huque MH, Anderson C, Walton R, Ryan L. Individual level covariate adjusted conditional autoregressive (indicar) model for disease mapping. Int J Health Geogr. 2016; 15(1):25. doi:10.1186/s12942-016-0055-7.

    Article  PubMed  PubMed Central  Google Scholar 

  116. Greenland S. Avoiding power loss associated with categorization and ordinal scores in dose-response and trend analysis. Epidemiology. 1995; 6(4):450–4.

    Article  CAS  PubMed  Google Scholar 

  117. Rothman KJ, Greenland S, Lash T. Modern Epidemiology, vol. 3. Philadelphia: Lippincott Williams & Wilkins; 2008.

    Google Scholar 

  118. Bennette C, Vickers A. Against quantiles: categorization of continuous variables in epidemiologic research, and its discontents. BMC Med Res Methodol. 2012; 12:21. doi:10.1186/1471-2288-12-21.

    Article  PubMed  PubMed Central  Google Scholar 

  119. Learnihan VB. The physical environment as an influence of walking in the neighbourhood: objective measurement and validation: University of Western Australia; 2007. Accessed 12 March 2017.

  120. Van Dyck D, Cerin E, Conway TL, De Bourdeaudhuij I, Owen N, Kerr J, Cardon G, Frank LD, Saelens BE, Sallis JF. Perceived neighborhood environmental attributes associated with adults’ transport-related walking and cycling: Findings from the usa, australia and belgium. Int J Behav Nutr Phys Act. 2012; 9:70. doi:10.1186/1479-5868-9-70.

    Article  PubMed  PubMed Central  Google Scholar 

  121. Van Dyck D, Cerin E, Conway TL, De Bourdeaudhuij I, Owen N, Kerr J, Cardon G, Frank LD, Saelens BE, Sallis JF. Perceived neighborhood environmental attributes associated with adults’ leisure-time physical activity: findings from belgium, australia and the usa. Health Place. 2013; 19:59–68. doi:10.1016/j.healthplace.2012.09.017.

    Article  PubMed  Google Scholar 

  122. Openshaw S. The Modifiable Areal Unit Problem (CATMOG 38). Norwich: Geo Books; 1984.

    Google Scholar 

Download references


This research was completed using data collected through the 45 and Up Study ( The 45 and Up Study is managed by the Sax Institute in collaboration with major partner Cancer Council NSW; and partners: the National Heart Foundation of Australia (NSW Division); NSW Ministry of Health; NSW Government Family & Community Services – Carers, Ageing and Disability Inclusion; and the Australian Red Cross Blood Service. We thank the many thousands of people participating in the 45 and Up Study. We wish to acknowledge Associate Professor Philayrath Phongsavan for her tireless and effective coordination of this project, and Assistant Professor Casey Durand, Associate Professor Jasper Schipperijn, and Dr Nick Garrett for their constructive comments on earlier versions of our manuscript. DJM is grateful to Hevan Corrimal for their ongoing support of this research.


This study is part of the “Understanding the impact of the social, economic and environmental factors on the health of Australians in mid - later life; where are the opportunities for prevention?” study (National Health & Medical Research Council (NHMRC) Grant 402810).

Availability of data and materials

Details on accessing 45 and Up Study data are available on the The Sax Institute website (

Author information

Authors and Affiliations



DJM undertook all statistical analyses and wrote the first draft of the manuscript with contributions from GGM, AEB, and BBJ. All authors were involved in study conception, and contributed to and approved the final manuscript.

Corresponding author

Correspondence to Darren J. Mayne.

Ethics declarations

Ethics approval and consent to participate

The 45 and Up Study was approved by the University of New South Wales Human Research Ethics Committee. Our study was approved as a sub-study of the Social, Environmental and Economic Factors (SEEF) Study by The University of Sydney Human Research Ethics Committee (Project Number 2012/1725) and the 45 and Up Study Scientific Advisory Committee (Project 10009). All participants in the 45 and Up Study have given signed consent for their information to be used in approved studies, and for follow-up and linkage to routine health databases.

Consent for publication

Not applicable. This manuscript was reviewed for technical accuracy by The 45 and Up Study coordinating centre prior to its submission for peer review.

Competing interests

The authors declare that they have no competing interests.

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License(, which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver( applies to the data made available in this article, unless otherwise stated.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Mayne, D.J., Morgan, G.G., Jalaludin, B.B. et al. The contribution of area-level walkability to geographic variation in physical activity: a spatial analysis of 95,837 participants from the 45 and Up Study living in Sydney, Australia. Popul Health Metrics 15, 38 (2017).

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: