Summary
Background
Physical inactivity is a global pandemic responsible for over 5 million deaths annually through its effects on multiple non-communicable diseases. We aimed to document how objectively measured attributes of the urban environment are related to objectively measured physical activity, in an international sample of adults.
Methods
We based our analyses on the International Physical activity and Environment Network (IPEN) adult study, which was a coordinated, international, cross-sectional study. Participants were sampled from neighbourhoods with varied levels of walkability and socioeconomic status. The present analyses of data from the IPEN adult study included 6822 adults aged 18–66 years from 14 cities in ten countries on five continents. Indicators of walkability, public transport access, and park access were assessed in 1.0 km and 0.5 km street network buffers around each participant’s residential address with geographic information systems. Mean daily minutes of moderate-to-vigorous-intensity physical activity were measured with 4–7 days of accelerometer monitoring. Associations between environmental attributes and physical activity were estimated using generalised additive mixed models with gamma variance and logarithmic link functions.
Results
Four of six environmental attributes were significantly, positively, and linearly related to physical activity in the single variable models: net residential density (exp[b] 1.006 [95% CI 1.003–1.009]; p=0.001), intersection density (1.069 [1.011–1.130]; p=0.019), public transport density (1.037 [1.018–1.056]; p=0.0007), and number of parks (1.146 [1.033–1.272]; p=0.010). Mixed land use and distance to nearest public transport point were not related to physical activity. The difference in physical activity between participants living in the most and least activity-friendly neighbourhoods ranged from 68 min/week to 89 min/week, which represents 45–59% of the 150 min/week recommended by guidelines.
Interpretation
Design of urban environments has the potential to contribute substantially to physical activity. Similarity of findings across cities suggests the promise of engaging urban planning, transportation, and parks sectors in efforts to reduce the health burden of the global physical inactivity pandemic.
Funding
Funding for coordination of the IPEN adult study, including the present analysis, was provided by the National Cancer Institute of National Institutes of Health (CA127296) with studies in each country funded by different sources.
Introduction
Physical inactivity is a global pandemic, responsible for more than 5 million deaths per year and is one of the UN’s primary targets to reduce non-communicable diseases.1–3 Improvements to urban environments to facilitate physical activity for transportation and recreation is a recommended strategy.4,5
People who live in walkable neighbourhoods that are densely populated, have interconnected streets, and are close to shops, services, restaurants, public transport, and parks, tend to be more physically active than residents of less walkable areas.6,7 Studies of built environments and physical activity have been criticised for being done in only a few countries,6,8,9 not capturing all types of urban environment, and relying on self-reported environmental measures. International studies are needed to represent the full range of environmental variability. If findings are generally applicable across countries, then built environment interventions are likely to be viewed as relevant to non-communicable disease policies inter nationally.
The purpose of this 14 city and ten country study was to document the strength, shape, and generalisability of associations between neighbourhood environment attributes and total moderate to vigorous intensity physical activity (MVPA). Objective measures of built environments and physical activity enhance precision and credibility of the findings.
Methods
Study design and neighbourhood selection
The International Physical Activity and Environment Network (IPEN) adult study was a multicountry cross-sectional epidemiological study with the same design and similar methods, described in detail elsewhere.10 The study included participants from 17 cities in 12 countries: Australia (Adelaide), Belgium (Ghent), Brazil (Curitiba), Colombia (Bogota), Czech Republic (Olomouc and Hradec Kralove), Denmark (Aarhus), China (Hong Kong), Mexico (Cuernavaca), New Zealand (North Shore, Waitakere, Wellington, and Christchurch), Spain (Pamplona), the UK (Stoke-on-Trent), and the USA (Seattle, WA; and Washington, DC and Baltimore, MD). The IPEN adult study was designed to maximise variation in neighbourhood walkability and socioeconomic status (SES) by identifying similar numbers of neighbourhoods stratified as having higher walkability and higher SES,11 higher walkability and lower SES, lower walkability and higher SES, and lower walkability and lower SES. Neighbourhood walkability index scores were created for small geographical areas in each city (termed administrative units, equivalent to US Census block groups) with geographic information systems (GIS),11 with some differences by country.10 Net residential density, intersection density, and mixed land use variables were standardised, and the mean of the three z scores was computed as the index.11 The SES indicator was usually area-level income, but sometimes it was education or a government-created composite.10 Indicators were chosen based on the data available. Neighbourhoods that met the criteria for the four stratification groups were selected and participants were recruited from those neighbourhoods.
Participant recruitment
Households in selected neighbourhoods were identified with databases from commercial and government sources with various methods used to obtain representative samples in each neighbourhood, including recruitment by mail or telephone and personal visits.10 In each selected household an adult was invited to complete a survey and wear an accelerometer to objectively measure physical activity. Study dates ranged from 2002 to 2011 across countries, with each country typically recruiting over a full year. Each country obtained ethics approval from their local institutional review boards and all participants provided written informed consent.
Participants
The IPEN adult study included 14 222 adults aged 18–66 years. The present study included 10 008 participants also aged 18–66 years from 14 of the 17 cities from ten countries where objective measures were available. Three cities were excluded because either no accelerometer data were collected (Adelaide, Australia) or no GIS data were available (Pamplona, Spain, and Hradec Kralove, Czech Republic). About a quarter (n=2739) of participants did not wear an accelerometer, either because they did not consent or the investigators could not afford to collect accelerometer data for all participants. For cities able to collect accelerometer data for all participants, 87–100% provided complete data. Characteristics of the 6822 participants with 4 or more days of valid accelerometer data by city are shown in table 1. Of these participants, 1740 (26%) were in the higher walkability and higher SES quadrant, 1736 (25%) were in the higher walkability and lower SES quadrant, 1845 (27%) were in the lower walkability and higher SES quandrant, and 1501 (22%) were in the lower walkability and lower SES quadrant.
Table 1:
Descriptive statistics of sample sociodemographic characteristics and accelerometer based moderate to vigorous physical activity
| All cities | Ghent, Belgium* | Curitiba, Brazil† | Bogota, Colombia† | Olomouc, Czech Republic* | Aarhus, Denmark† | Hong Kong, China† | Cuernavaca, Mexico* | North Shore, New Zealand* | Waitakere, New Zealand* | Wellington, New Zealand* | Christchurch, New Zealand* | Stoke-on-Trent, UK† | Seattle, WA, USA* | Baltimore, MD, USA* | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Sociodemographic characteristics | |||||||||||||||
| Number with ≥4 days valid physical activity data | 6822 (68%) | 1050 (90%) | 330 (47%) | 223 (23%) | 258 (78%) | 272 (42%) | 269 (56%) | 656 (97%) | 373 (73%) | 399 (78%) | 416 (84%) | 373 (75%) | 135 (16%) | 1198 (93%) | 870 (95%) |
| Age (years) | 43 (12) | 43 (13) | 42 (13) | 46 (12) | 39 (14) | 40 (14) | 42 (13) | 42 (13) | 43 (12) | 42 (11) | 40 (12) | 43 (12) | 44 (13) | 44 (11) | 47 (11) |
| Male sex | 3161 (46·3%) | 508 (48·4%) | 160 (48·5%) | 71 (31·8%) | 93 (36·0%) | 106 (39·0%) | 109 (40·5%) | 300 (45·7%) | 139 (37·3%) | 161 (40·4%) | 198 (47·6%) | 170 (45·6%) | 63 (46·7%) | 659 (55·0%) | 424 (48·7%) |
| Education level | |||||||||||||||
| Less than high school | 837 (12·3%) | 45 (4·3%) | 92 (27·9%) | 104 (46·6%) | 53 (20·5%) | 20 (7·4%) | 98 (36·4%) | 286 (43·6%) | 9 (2·4%) | 15 (3·8%) | 2 (0·5%) | 32 (8·6%) | 52 (38·5%) | 13 (1·1%) | 16 (1·8%) |
| High school graduate | 2597 (38·1%) | 340 (32·4%) | 103 (31·2%) | 81 (36·3%) | 100 (38·8%) | 115 (42·3%) | 62 (23·0%) | 188 (28·7%) | 215 (57·6%) | 258 (64·7%) | 187 (45·0%) | 212 (56·8%) | 62 (45·9%) | 417 (34·8%) | 257 (29·5%) |
| College or more | 3335 (48·9%) | 654 (62·3%) | 135 (40·9%) | 38 (17·0%) | 77 (29·8%) | 137 (504%) | 109 (40·5%) | 178 (27·1%) | 145 (38·9%) | 126 (31·6%) | 227 (54·6%) | 128 (34·3%) | 20 (14·8%) | 766 (63·9%) | 595 (68·4%) |
| Missing data | 53 (0·8%) | 11 (0·1%) | 0 (0·0%) | 0 (0·0%) | 28 (10·9%) | 0 (0·0%) | 0 (0·0%) | 4 (0·6%) | 4 (1·1%) | 0 (0·0%) | 0 (0·0%) | 1 (0·3%) | 1 (0·7%) | 2 (0·2%) | 2 (0·2%) |
| Employed | 5370 (78·7%) | 843 (80·3%) | 262 (79·4%) | 135 (60·5%) | 201 (77·9%) | 205 (75·4%) | 163 (60·6%) | 466 (71·0%) | 285 (76·4%) | 344 (86·2%) | 364 (87·5%) | 319 (85·5%) | 87 (64·4%) | 974 (81·3%) | 722 (83·0%) |
| Married or living with partner | 4389 (64·3%) | 766 (73·0%) | 199 (60·3%) | 137 (61·4%) | 155 (60·1%) | 188 (69·1%) | 151 (56·1%) | 425 (64·8%) | 265 (71·0%) | 303 (75·9%) | 249 (59·9%) | 213 (57·1%) | 61 (45·2%) | 768 (64·1%) | 529 (60·8%) |
| Accelerometer variables | |||||||||||||||
| Valid days of accelerometer wear time | 6·5 (1·1) | 6·7 (1·1) | 6·7 (1·0) | 6·6 (1·0) | 6·2 (1·2) | 7·0 (0·8) | 5·9 (1·0) | 5·7 (1·0) | 6·4 (1·3) | 6·4 (1·3) | 6·7 (1·3) | 6·5 (1·3) | 6·6 (1·0) | 6·7 (0·8) | 6·7 (1·2) |
| Accelerometer wear time (h/day) | 14·4 (1·3) | 14·7 (1·3) | 14·0 (1·3) | 13·9 (1·2) | 13·9 (1·4) | 14·9 (1·1) | 14·4 (1·4) | 14·0 (1·4) | 14·2 (1·2) | 14·1 (1·3) | 14·0 (1·2) | 14·0 (1·2) | 14·6 (1·2) | 14·7 (1·3) | 14·8 (1·4) |
| MVPA (min/day)‡ | 37·3 (26·5) | 35·5 (23·5) | 31·5 (24·6) | 37·0 (26·4) | 47·1 (27·7) | 39·7 (23·2) | 44·9 (25·3) | 31·2 (25·2) | 45·7 (28·4) | 37·2 (29·2) | 50·1 (31·0) | 44·0 (32·5) | 36·7 (27·3) | 36·3 (24·9) | 29·2 (22·0) |
Data are mean (SD) or n (%) unless otherwise indicated. Totals might not be exactly 100% because of rounding. MVPA=moderate to vigorous physical activity.
Study city aimed to collect accelerometer data in the total sample.
Study city aimed to collect accelerometer data in a fixed proportion of the total sample.
Average for valid days.
Outcomes
Physical activity was measured objectively with accelerometers, a reliable, valid, and accepted method.12–14 Participants were instructed to wear accelerometers for 7 days around the waist, except during sleep, swimming, and showering. Except for New Zealand, which used Actical devices (Philips Respironics, Murrysville, PA, USA), all countries used varying models of ActiGraph monitors (Pensacola, FL, USA). Only vertical axis data were included in the scoring, expressed as counts per min (cpm). For Actical data, we developed moderate (730–3399 cpm) and vigorous (≥3400 cpm) intensity cutpoints to enable comparison with the ActiGraph estimates.15 60 s periods were used in data collection and non-wear time was defined as 60 consecutive min or more with zero cpm. Valid days had 10 h or more of wear time. Participants with 4 or more valid days were included in analyses. These methods were consistent with recommendations and common practices.12,16 Data were scored with MeterPlus 4.3 software, with Freedson’s cutpoint of 1952 cpm for moderate intensity to derive the outcome variable, mean minutes of MVPA per valid day.17
Variables related to built environment were created with GIS software. Areas known as buffers within 0.5 km and 1.0 km of the participants’ homes, reachable by the street network, were defined to estimate accessible neighbourhood features. Templates were developed to guide international teams on constructing comparable GIS variables.18 The templates were also used to document protocol adherence, which allowed for comparability evaluations. A description of GIS methods and variables, examples of data sources for each country, comparability evaluations, and descriptive results of variation in GIS-based environmental variables within and across cities has been published.18 The following variables were adequately comparable across cities and were used in analyses: net residential density, street intersection density, retail and civic land use ratio to buffer area (access to common destinations), public transport density, public park density, and distance to nearest transport. The panel provides definitions of variables and key terms. Table 2 presents descriptive findings for environmental variables overall and by city.
Table 2:
Descriptive statistics of objectively-assessed environmental attributes
| All cities | Ghent, Belgium | Curitiba, Brazil | Bogota, Colombia | Olomouc, Czech Republic | Aarhus, Denmark | Hong Kong, China | Cuernavaca, Mexico | North Shore, New Zealand | Waitakere, New Zealand | Wellington, New Zealand | Christchurch, New Zealand | Stoke-on-Trent, UK | Seattle, WA, USA | Baltimore, MD, USA | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Net residential density (per km2)—10 km buffer | |||||||||||||||
| Mean | 6682 (12 522) | 7853 (6795) | 5993 (3950) | 9273 (3248) | 18 086 (9808) | 7115 (5649) | 57 322 (25 592) | 2237 (933) | 1764 (739) | 2029 (738) | 4203 (4973) | 1658 (360) | 4579 (1447) | 3015 (3574) | 2498 (2330) |
| Median | 2493 (1529–2493) | 5214 (1240–12 665) | 4652 (3851–6832) | 8995 (6677–11 726) | 18 810 (9424–24 512) | 4635 (2205–11 750) | 65456 (35 629–78 709) | 2202 (1597–2730) | 1735 (1619–1921) | 2126 (1519–2484) | 1909 (1506–5643) | 1544 (1438–1925) | 4161 (3766–4916) | 2183 (1292–3180) | 1585 (791–3934) |
| Net residential density (per km2)—0·5 km buffer | |||||||||||||||
| Mean | 7025 (13355) | 7246 (6894) | 6338 (5262) | 12 997 (5147) | 19219 (15579) | 8398 (7633) | 57276 (30728) | 2619 (1964) | 1748 (382) | 2665 (1476) | 3559 (3014) | 1669 (447) | 4471 (1674) | 3328 (4471) | 3424 (4505) |
| Median | 2729 (1530–6214) | 5190 (1561–9894) | 4776 (3542–6936) | 12475 (9120–15154) | 14880 (8070–25457) | 4661 (1916–16792) | 60912 (29896–86372) | 2309 (1467–3267) | 1790 (1570–2015) | 2462 (1891–2959) | 1767 (1471–4930) | 1552 (1339–1982) | 3979 (3501–4923) | 2244 (1214–3215) | 1774 (786–4294) |
| Intersection density (per km2)—10 km buffer | |||||||||||||||
| Mean | 76 (57) | 84 (62) | 76 (16) | 227 (91) | 67 (20) | 83 (22) | 128 (58) | 146 (47) | 27 (7) | 28 (10) | 42 (15) | 35 (6) | 93 (29) | 71 (22) | 55 (28) |
| Median | 65 (38–94) | 73 (35–92) | 72 (64–83) | 234 (153–301) | 67 (54–80) | 86 (72–98) | 129 (85–172) | 135 (111–168) | 27 (23–32) | 28 (24–31) | 43 (34–52) | 36 (32–40) | 87 (75–114) | 71 (56–85) | 53 (38–67) |
| Intersection density (per km2)—0·5 km buffer | |||||||||||||||
| Mean | 87 (69) | 86 (61) | 84 (23) | 249 (111) | 75 (25) | 105 (31) | 174 (79) | 174 (72) | 31 (12) | 35 (18) | 42 (21) | 37 (9) | 113 (37) | 76 (28) | 64 (40) |
| Median | 71 (40–109) | 74 (36–109) | 80 (68–94) | 222 (163–342) | 75 (60–88) | 105 (85–125) | 162 (122–216) | 166 (119–234) | 31 (21–40) | 33 (24–40) | 40 (26–54) | 36 (30–43) | 116 (81–133) | 77 (57–95) | 59 (40–78) |
| Proportion of retail and civic land area to total buffer area—10 km buffer* | |||||||||||||||
| Mean | 0·17 (0·24) | 0·14 (0·13) | 0·16 (0·10) | 0·12 (0·09) | 0·06 (0·06) | 0·52 (0·42) | 0·54 (0·24) | 0·17 (0·12) | 0·12 (0·27) | 0·10 (0·13) | 0·31 (0·52) | 0·18 (0·31) | 0·04 (0·04) | 0·08 (0·06) | 0·12 (0·13) |
| Median | 0·09 (0·04–0·20) | 0·09 (0·03–0·25) | 0·13 (0·10–0·20) | 0·08 (0·04–0·18) | 0·04 (0·02–0·07) | 0·44 (0·16–0·80) | 0·58 (0·47–0·68) | 0·15 (0·07–0·27) | 0·08 (0·04–0·14) | 0·07 (0·03–0·11) | 0·11 (0·05–0·30) | 0·06 (0·03–0·13) | 0·02 (0·01–0·05) | 0·06 (0·04–0·10) | 0·08 (0·04–0·17) |
| Ratio retail and civic land area to total buffer area—0·5 km buffer* | |||||||||||||||
| Mean | 0·17 (0·33) | 0·16 (0·21) | 0·14 (0·11) | 0·10 (0·08) | 0·07 (0·09) | 0·74 (0·76) | 0·73 (0·44) | 0·17 (0·16) | 0·14 (0·31) | 0·08 (0·15) | 0·28 (0·60) | 0·08 (0·30) | 0·05 (0·07) | 0·07 (0·09) | 0·11 (0·21) |
| Median | 0·07 (0·01–0·18) | 0·09 (0·01–0·24) | 0·13 (0·06–0·19) | 0·07 (0·04–0·14) | 0·04 (0·01–0·06) | 0·49 (0·15–1·07) | 0·65 (0·51–0·96) | 0·14 (0·01–0·27) | 0·06 (0·01–0·17) | 0·01 (0·00–0·09) | 0·07 (0·00–0·21) | 0·02 (0·00–0·08) | 0·01 (0·00–0·06) | 0·04 (0·00–0·10) | 0·05 (0·00–0·13) |
| Public transport density (per km2)—10 km buffer† | |||||||||||||||
| Mean | 15·8 (12·9) | 9·4 (6·3) | 25·8 (7·3) | 2·2 (2·6) | 13·6 (5·4) | 9·4 (4·6) | 12·0 (8·2) | 29·1 (24·4) | 19·0 (7·3) | 9·0 (7·1) | 16·6 (8·6) | 16·0 (9·0) | 25·3 (7·9) | 15·9 (9·7) | 16·9 (13·6) |
| Median | 14·3 (6·6–21·9) | 7·7 (1·5–15·4) | 25·0 (20·4–29·8) | 1·2 (0·0–3·3) | 14·5 (9·1–17·8) | 9·1 (5·9–12·7) | 13·0 (4·0–16·9) | 26·0 (11·4–39·4) | 20·1 (14·4–24·8) | 6·9 (4·5–12·7) | 14·9 (11·8–22·0) | 16·2 (9·4–20·0) | 24·0 (19·1–31·5) | 15·7 (9·0–22·8) | 15·7 (7·4–24·5) |
| Public transport density (per km2)—0·5 km buffer† | |||||||||||||||
| Mean | 17·0 (17·0) | 10·4 (9·8) | 24·0 (11·5) | 2·4 (4·5) | 15·0 (9·1) | 10·9 (7·1) | 13·0 (13·2) | 33·3 (35·6) | 20·1 (12·2) | 8·4 (7·5) | 19·4 (12·2) | 16·8 (14·7) | 28·2 (13·7) | 16·8 (13·1) | 18·0 (17·7) |
| Median | 14·1 (4·2–24·2) | 8·6 (3·3–14·3) | 23·5 (16·2–31·2) | 0·0 (0·0–2·7) | 14·4 (9·0–21·0) | 10·8 (5·9–16·4) | 11·2 (0·0–20·0) | 25·2 (0·0–49·0) | 20·0 (12·8–27·7) | 7·6 (2·5–12·5) | 20·7 (12·8–26·1) | 15·8 (5·1–22·5) | 26·4 (19·1–36·0) | 16·8 (5·0–26·9) | 15·5 (0·0–26·7) |
| Street network distance to nearest transport stop or station (m) | |||||||||||||||
| Mean | 421 (638) | 317 (284) | 178 (111) | 1863 (1525) | 265 (173) | 303 (230) | 426 (350) | 501 (659) | 245 (216) | 343 (266) | 222 (284) | 300 (240) | 212 (136) | 382 (439) | 639 (1017) |
| Median | 242 (124–429) | 258 (150–360) | 161 (83–249) | 1193 (557–3385) | 232 (159–350) | 235 (151–384) | 353 (205–506) | 239 (102–591) | 186 (93–338) | 297 (153–468) | 147 (60–301) | 242 (118–415) | 189 (100–299) | 227 (128–449) | 238 (117–550) |
| Number of parks contained or intersected by buffer of 1·0 km | |||||||||||||||
| Mean | 5·5 (6·5) | 3·8 (3·6) | 6·0 (4·6) | 25·4 (13·3) | 3·7 (4·4) | 4·4 (3·4) | 13·5 (10·0) | 1·5 (2·1) | 12·3 (6·5) | 8·7 (3·9) | 4·6 (2·5) | 5·6 (2·5) | 2·8 (1·5) | 3·8 (2·9) | 2·4 (2·1) |
| Median | 4·0 (1·0–7·0) | 3·0 (1·0–6·0) | 5·0 (2·0–8·0) | 24·0 (14·0–34·0) | 2·0 (1·0–5·0) | 4·0 (2·0–6·0) | 11·0 (5·0–19·0) | 1·0 (0·0–2·0) | 11·0 (8·0–16·0) | 8·0 (6·0–11·0) | 4·0 (3·0–7·0) | 6·0 (3·0–7·0) | 2·0 (2·0–4·0) | 3·0 (1·0–6·0) | 2·0 (1·0–3·0) |
| Number of parks contained or intersected by buffer of 0·5 km | |||||||||||||||
| Mean | 1·8 (2·3) | 1·2 (1·3) | 2·0 (2·3) | 7·4 (4·6) | 1·1 (1·6) | 1·3 (1·4) | 4·0 (3·5) | 0·6 (0·9) | 4·1 (2·7) | 3·5 (2·6) | 1·4 (1·0) | 1·6 (1·2) | 1·3 (0·9) | 1·2 (1·2) | 0·9 (1·0) |
| Median | 1·0 (0·0–2·0) | 1·0 (0·0–2·0) | 1·0 (0·0–3·0) | 7·0 (4·0–10·0) | 1·0 (0·0–2·0) | 1·0 (0·0–2·0) | 3·0 (1·0–5·0) | 0·0 (0·0–1·0) | 4·0 (2·0–6·0) | 3·0 (2·0–5·0) | 1·0 (1·0–2·0) | 1·0 (1·0–2·0) | 1·0 (1·0–2·0) | 1·0 (0·0–2·0) | 1·0 (0·0–1·0) |
Data are mean (SD) and median (first quartile to third quartile).
Descriptive statistics were computed after assigning a value of 3.1 to all participants with values greater than 3.1 (one value truncated for 1.0 km buffer measure and 22 for 0.5 km buffer measure).
Measured by number of transport options.
Covariates included age, sex, education (<12 years or high school level, high school graduation, and university degree or higher), marital status (married or living with partner vs other), employment status (unemployed vs employed), city, accelerometer wear time, and SES of administrative unit (low vs high).
Statistical analysis
Associations between environmental variables and physical activity (min/day) were estimated with generalised additive mixed models (GAMMs) with gamma variance and logarithmic link functions, appropriate for the sampling strategy and distributional properties of the outcome variable.19,20 These models also allowed for the simultaneous estimation of the amount of variability in participants’ individual MVPA attributable to factors at city level, administrative unit level (within city), and individual level. Covariate-adjusted single environmental variable (SEV) and multiple environmental variable (MEV) GAMMs were estimated. The MEV GAMM included only statistically significant (p<0.05) buffer-specific environmental correlates for each buffer size. Environmental variables were entered simultaneously in the MEV GAMMs as collinearity was not problematic. Curvilinearity of relations was assessed with thin-plate spline smooth terms.20 Separate GAMMs were run to estimate environmental features by study city interaction effects to assess whether the associations of environmental features with MVPA differed across cities. The significance of interactions was assessed by comparison of Akaike information criterion values of models with and without an interaction term (≥10 difference indicated significance).21 To quantify effect sizes of significant environmental correlates of MVPA, covariate-adjusted differences in weekly minutes of MVPA were estimated between participants living in buffers with the lowest 5% and highest 5% values of environmental correlates and between participants living in areas with values of environmental correlates corresponding to the lowest and highest average city-level values. We also expressed these differences in activity in percentages of the amount needed to comply with the WHO physical activity guidelines (ie, percentages of 150 min/week of MVPA).22
To assess built environment contributions to differences in physical activity at the city level, administrative unit level, and person level, three-level GAMMs with random intercepts at the city and administrative unit levels adjusted and unadjusted for environmental features were estimated, and the percentage reductions in residual variances were computed. As only 220 (2.2%) of 10 008 cases had missing data, analyses were only done for complete cases. All analyses used R.
Role of the funding source
Funding for coordination of the IPEN adult study was provided by the National Cancer Institute of National Institutes of Health (CA127296), with studies in each country funded by different sources. Funders were not involved in planning or executing the study and they were not involved in preparing the manuscript. JFS had full access to all of the data in the study and had the final responsibility for the decision to submit for publication.
Results
On average, participants accumulated about 37 min/day of MVPA. Baltimore (USA) had the lowest average value (29.2 min) and Wellington (New Zealand) had the highest average value of MVPA (50.1 min; table 1).The standard deviation of MVPA at the city level was 6.3 min/day, at the administrative unit level was 4.6 min/day, and at the person level was 24.4 min/day. Higher variability at the person level was expected. Four of six environmental variables were significantly associated (p<0.05) with MVPA in the SEV models (table 3). These four variables explained 0–11% of MVPA variability at the city level and explained 7–11% at the administrative unit levels, but virtually no variance at the person (within administrative unit) level. Net residential density, intersection density, public transport density, and number of parks within participants’ buffers were linearly and positively related with MVPA. Both buffer sizes were tested and with the exception of number of parks, stronger relations were noted for variables calculated for 1.0 km than for 0.5 km buffers. Relations for variables calculated for the most significant buffers (1.0 km or 0.5 km) are reported (table 3).
Table 3:
Pooled associations of environmental attributes with daily minutes of moderate to vigorous physical activity by model (n=6679)
| Buffer size (km) | exp(b) | exp (95% CI) | p value | |
|---|---|---|---|---|
| Net residential density (1000 dwellings/km2) | ||||
| SEV | 1·0 | 1·006 | 1·003–1·009 | 0·001 |
| MEV | 1·0 | 1·004 | 1·001–1·007 | 0·006 |
| Intersection density (100 intersections/km2) | ||||
| SEV | 1·0 | 1·069 | 1·011–1·130 | 0·019 |
| MEV | 1·0 | ·· | ·· | ·· |
| Proportion of retail combined and civic land area to total buffer area | ||||
| SEV | 1·0 | 1·056 | 0·964–1·157 | 0·238 |
| MEV | 1·0 | ·· | ·· | ·· |
| Public transport density (10 transport points/km2) | ||||
| SEV | 1·0 | 1·037 | 1·018–1·056 | 0·0007 |
| MEV | 1·0 | 1·030 | 1·011–1·049 | 0·006 |
| Number of parks contained or intersected by buffer (10 parks/km2)* | ||||
| SEV | 0·5 | 1·146 | 1·033–1·272 | 0·010 |
| MEV | 0·5 | 1·111 | 1·000–1·233 | 0·046 |
| Street network distance to nearest transport stop (1000 m) | ||||
| SEV | 1·0 | 1·033 | 0·996–1·071 | 0·078 |
| MEV | 1·0 | ·· | ·· | ·· |
All regression cofficients adjusted for respondents’ age, sex, marital status, educational attainment, employment status, administrative-unit socioeconomic status, accelerometer wear time, and study city. Units of measurement shown after variable name in parentheses. exp(b) is the proportional increase in physical activity associated with a 1 unit of measurement increase in the predictor (eg, 1000 dwellings/km2 is 1 unit of measurement for net residential density). Only results for the buffer size (1.0 km or 0.5 km) showing the strongest relationships with physical activity are reported. SEV=single environmental variable; MEV=multiple environmental variable (only significant environmental correlates included); exp(b)=antilogarithm of regression coefficient; exp (95% CI)=antilogarithm of confidence intervals.
Adjusted for net residential density, intersection density, and transport density.
After we adjusted for other environmental variables in the multiple environmental variable (MEV) models, net residential density and public transport density remained significant, positive, and linear correlates of MVPA for both buffer sizes. Additionally, number of parks significantly contributed to explaining MVPA in the model based on 0.5 km buffers (table 3). The MEV models explained 11–12% of the total MVPA variance.
Based on the absence of significant interactions between environmental and city features, we can conclude that associations were generalisable across study cities, with the exception of number of parks in 0.5 km buffers. Specifically, positive associations between parks within 0.5 km buffers and physical activity in the SEV model were reported only in Ghent, Belgium (exp[b]=1.772; 95% CI 1.177–2.669; p=0.006) and Seattle, USA (exp[b]=2.064; 95% CI 1.399–3.045; p<0.001). After we adjusted for other environmental variables, the park counts by city interaction was no longer significant and a significant positive association of park counts with MVPA was reported across all cities (table 3). Thus, we noted evidence of similar relations of urban environment variables and physical activity across diverse cities. Analyses examining the shape of associations reported no sufficient evidence for curvilinearity of effects. Therefore, we concluded that environment associations with physical activity were linear.
Table 4 reports the estimated differences in minutes per week of MVPA between participants living in areas at the lowest and highest 5% of the sample values for specific significant environmental correlates, including areas with values of environmental correlates equal to those of the cities with the lowest and highest average values. The differences in MVPA between residents living in areas at the lowest and highest 5% for specific single environmental features ranged from 21 to 32 min/week. The differences in MVPA between participants living in areas with values of single environmental correlates equal to those of study cities with the lowest and highest average values ranged from 24 min/week to 89 min/week. This finding corresponded to meeting between 16% and 59% of the recommended 150 min/week of physical activity. The estimated differences in minutes per week of MVPA between participants living in areas with all significant environmental correlates at the lowest and highest average city values was 68 min/week when comparing the lowest 5% versus highest 5% neighbourhood buffer values. A difference of 89 min/week was found when comparing the lowest versus highest average city values. These differences are equivalent to meeting 45–59% of the 150 min/week physical activity guidelines.
Table 4:
Differences in estimated MVPA between participants with low and high values for significant environmental correlates
| 5% lowest values of environmental features | 5% highest values of environmental features | Differences in weekly minutes of MVPA between lowest 5% and highest 5% values of environmental correlate (95% CI) | Lowest average study-city value for environmental features | Highest average study-city value for environmental features | Differences in weekly minutes of MVPA between lowest and highest average study-city values of environmental features (95% CI) | |
|---|---|---|---|---|---|---|
| SEV | ||||||
| Net residential density—1·0 km buffer | 710 | 21 078 | 29 (12–46) 19% of PAG |
1658·0 | 57 322·0 | 89 (38–147) 59% of PAG |
| Intersection density—1·0 km buffer | 16 | 198 | 31 (5–60) 21% of PAG |
27·0 | 227·0 | 34 (5–68) 23% of PAG |
| Public transport density—1·0 km buffer | 0 | 35 | 32 (17–52) 21% of PAG |
2·2 | 29·1 | 24 (12–36) 16% of PAG |
| Number of parks contained or intersected by 0·5 km buffer | 0 | 6 | 21 (5–37) 14% of PAG |
0·6 | 7·4 | 24 (5–43) 16% of PAG |
| MEV | ||||||
| Net residential density—1·0 km buffer | 710 | 21 078 | 49 (15–86)* | 1658·0 | 57 322·0 | 89 (29–161)* |
| Public transport density—1·0 km buffer | 0 | 35 | 33% of PAG | 2·2 | 29·1 | 59% of PAG |
| Net residential density—0·5 km buffer | 652 | 28 917 | 48 (6–78)† | 1669·0 | 57 276·0 | 68 (11–144)† |
| Public transport density—0·5 km buffer | 0 | 46 | 32% of PAG | 2·4 | 33·3 | 45% of PAG |
| Number of parks contained or intersected by 0·5 km buffer | 0 | 6 | ·· | 0·6 | 7·4 | ·· |
The residual variability in MVPA at a specific level is expressed in standard deviations (after adjusting for sociodemographics and accelerometer-wear time). Only results for the buffer size (1.0 km or 0.5 km) showing the strongest relationships with physical activity are reported. MVPA=moderate to vigorous physical activity. SEV=single environmental variable. MEV=multiple environmental variable (only significant environmental correlates included). PAG=physical activity guidelines (total recommended amount of 150 min/week of MVPA).22
Combined effect of net residential density and public transport density.
Combined effect of net residential density, public transport density, and number of parks contained or intersected by 0.5 km buffer.
Discussion
This multicountry study identified urban environmental attributes that accounted for large differences in adults’ physical activity. Combinations of environmental features generally explained more variation in physical activity than single variables, suggesting that a comprehensive approach is needed to design activity supportive neighbourhoods. When we compared participants living in the 5% most with the 5% least activity supportive neighbourhoods, SEV models accounted for a smaller number of minutes of MVPA compared with models including all significant environmental variables. When we compared participants living in areas similar to the cities with the most versus the least activity supportive environments, single variables accounted for a difference of 24–89 weekly minutes of physical activity, compared with 68–89 min for combined variable models. Living in the most activity-friendly environments could help the average resident to achieve 45–59% of the 150 min/week of physical activity recommended guidelines.22 These observed effect sizes suggest that designing urban envìronments to be activity-supportive could have large effects on physical activity and those effects can be expected to generally apply to adults living in the neighbourhoods. Such widespread and long-term effects are in contrast to programmes that target individuals and tend to reach small numbers of people and produce short-term effects.23
Three environmental attributes had significant independent associations with total MVPA in the MEV and SEV models: net residential density, public transport density, and park density. Net residential density’s strong associations were consistent with those shown in many other studies.24 High residential density is generally deemed to be necessary for other components of walkability because local patronage is needed to support nearby shops and services and enough riders to support frequent transport service.25 Density of public transport stops was independently related to total activity. Public transport density was notably a significant correlate of MVPA but distance to nearest transport stop was not significant. One interpretation is that having various options for transport lines makes residents more likely to walk to a transport facility that meets their needs. Public transport access has been studied less often in relation to physical activity.6,24 Good transport access is a requirement for living a less car-dependent lifestyle.26 Particularly in the middle-income cities in the sample, car ownership was low and in these settings, active transport, such as walking and cycling could represent necessity and not choice. Thus, research into the role of public transport access in car owners and non-owners would be useful. The third significant variable in the final model was number of parks in the 0.5 km buffer. Park density is a relatively consistent correlate of adult physical activity.6,24 Although parks are usually thought of as supporting recreational activities through facilities and aesthetics, nearby parks can also be a destination for active transportation. Thus, the most well supported environmental variables were probably related to total physical activity through their effects on both recreational and transportation activities.
All reported associations were linear so we did not note a threshold or a point of diminishing returns for environmental attributes. Present findings, with probably the widest range of environmental variables yet reported, support a recommendation that higher levels of residential density, public transport access, and local parks should be recommended when designing physical activity supportive environments.
The measure of mixed land use was not related to physical activity in our study, although this factor is one of the more consistent correlates of physical activity.6,24 Proximal (eg, within 1 km buffers) retail shops and services provide frequently used destinations that stimulate regular walking. Because of the large variation in the proportion of retail and civic land use to buffer area within and between countries, the non-significant results were surprising. One possible explanation is the limitations of the GIS-based measure. Because most countries only had data for the land area devoted to each use, as opposed to building floor area, we were unable to tell whether each use was operating on part of the parcel or on several floors of a building covering the entire parcel. A related limitation was that the data were based on number of parcels, not on number of shops or offices, which might be more strongly related to frequency of use and thus to physical activity. In middle-income cities with a high prevalence of walking for transport, many shops were not registered, including those in permanent buildings as well as informal markets and street vendors. These data limitations could have reduced the power to detect an association.
Intersection density is an indicator of street connectivity that provides direct pathways for pedestrians and vehicles. This variable was significant in SEV models, but not in the full models, suggesting a confounding effect with other variables, such as residential density or public transport density.25
An important finding was the strong support for the similarity or generalisability of associations between built environment and physical activity across countries diverse in income, culture, and activity supportiveness. The diversity of the study cities in climate, demographics,10 and built environments has been documented in previous publications.18 Present results suggest systematic principles of environments that support physical activity apply on a global scale.19 Generalisable associations with physical activity were also reported in analyses of self-reported environment measures in the same study.19
Study strengths included the use of objective measures of both urban environments and physical activity, comparable variables across diverse countries, assessment of two buffer sizes, and analyses that tested for curvilinear effects and generalisability of associations across cities. Limitations included a small number of environmental variables that could be assessed through common environmental measures, likely variations in the quality of those measures across countries, scarce representation of low-income countries, a modest sample size in some cities that reduced power to detect differences in associations across cities, and cross-sectional design. Another limitation is that covariates, such as sex and education, might have different meanings and functions across countries. Other patterns of association might be noted with other age groups and built environment correlates are expected to differ by physical activity outcome. Absence of adjustment for self-selection into neighbourhoods is a frequent criticism of built environment studies but not all countries in the present study included measures that assessed reasons for neighbourhood selection.27
Our recommendations for future research are to expand the number of countries, especially low-income countries, in which associations between urban environment and physical activity are assessed; to develop objective measures for other environmental attributes relevant to physical activity, such as sidewalks, pedestrian zones, bicycle facilities, and factors affecting intersection quality (eg, crosswalks, pedestrian signals, and traffic calming); and to implement prospective studies and quasi-experimental evaluations of improvements in urban environments.
A recommendation for practice is to make the creation of activity supportive environments a regular function of public health agencies globally through work with sectors outside of public health. Regular assessment and reporting (ie, surveillance) of the quality of activity supportive environments is a vital component of efforts to foster creation of these environments. Health department staff should seek training, develop collaborations, and become advocates for improved policies in city planning, transportation, and parks agencies.
Design of urban environments has the potential to contribute nearly 90 min/week of physical activity, which is 60% of the 150 min/week recommended in physical activity guidelines. These potentially large effects of built environments were reported to apply similarly across ten diverse countries, indicating that urban design should be a globally relevant public health priority. Building, retrofitting, and maintenance of physical activity supportive features in cities worldwide to increase residential density, provide good transport service, and ensure access to parks would be expected to substantially increase physical activity in the population on a permanent basis and contribute to meeting the UN’s goals to reduce non-communicable diseases.2,3 Our study findings provide an impetus for public health proponents to collaborate with other sectors, including environmental sustainability groups, to promote physical activity supportive develo pment as a means to reduce energy consumption, greenhouse gas emissions, and air pollution,11,28 while achieving health and economic benefits.29
Research in context.
Evidence before this study
Evidence is growing that the design of urban environments has a role in the pandemic of physical inactivity, which is contributing to several non-communicable diseases. Numerous reviews have reported evidence that adults tend to be more physically active when they live in higher density, mixed-use neighbourhoods with destinations such as shops and parks within walking distance. However, findings have been inconsistent, perhaps due in part to assessments of only individual geographical sites with little environmental variability, infrequent use of standardised measures, and over-reliance on self-reported measures. Improvements in the evidence about built environments and physical activity are important because environments are constantly changing in ways that could have positive or negative effects on whole populations over many years.
Added value of this study
This analysis of data from a coordinated international study was designed to improve the quality of evidence by assessing a broad range of built environments across 14 cities in ten middle-income and high-income countries on five continents. The quality of measures was enhanced by using comparable objective measures of built environments (geographic information systems) and physical activity (electronic accelerometers that recorded motion every minute). Four of six environmental attributes were significantly, positively, independently, and linearly related to physical activity in the single variable models: residential density, intersection density, number of public transport stops, and number of parks within walking distance. The study provided novel information about the important role of access to public transport. In models adjusting for all the significant built environment variables, adults who lived in the most activity-friendly neighbourhoods did 68–89 min more of physical activity per week than those in the least activity-friendly neighbourhoods. This difference is larger than reported in most other studies. The relation of built environments to physical activity was generally similar across diverse cities, suggesting changing built environments is a solution that could be applied internationally.
Implications of all the available evidence
This study adds strength to previous calls for policy changes in the urban planning, transport, and parks and recreation sectors. Communities with high residential density also tend to have connected streets, shops, and services within walking distance. Access to public transport encourages physical activity because people walk to and from buses and trains. Public parks provide places for recreational physical activity. These activity-friendly characteristics can be deemed to be design principles that apply across countries. Because the associations were linear, every environmental improvement can be expected to contribute to increased physical activity, irrespective of whether the residents of the city are starting at a low or high level. The large differences in physical activity between participants living in the most and least activity-friendly neighbourhoods provide strong justification for public health agencies to work with other agencies to create healthier cities. Making cities more activity-friendly than at present could be a partial but substantial long-term solution to international pandemics of physical inactivity and non-communicable diseases.
Panel: Definitions and built environment variables.
Administrative units
Geographical areas with government-defined spatial boundaries in which population or socioeconomic status (SES) data are available. IPEN investigators selected the administrative unit in each country that roughly represented a small neighborhood-level geographic scale (about 600–1500 people) with available and best-quality SES and geographic information systems data. Examples include New Zealand meshblocks, US census block groups, and Hong Kong tertiary planning units. Administrative units were used for a-priori identification of study neighborhoods representing high and low walkability and high and low SES in each city.
Parcel
A division of immovable land created for taxation purposes and defined by its ownership, size, shape (boundaries), and functional land use.
Participant buffers
An irregular shaped polygon around a participant’s home address (geocoded). Buffer polygons were created for two distances (eg, 0.5 km and 1.0 km) in Esri’s ArcGIS software (Redlands, CA, USA) by tracing through unique street network in all directions to approximate accessible areas. The detailed no trim setting was used. The total area of the buffer was used as the denominator for density variables (except for residential land density). These buffer sizes were used to define attributes within walking distances of participants’ homes.
Net residential density
Number of residential dwellings (houses and apartments) divided by the residential land area (derived from residential parcels only) within participants’ buffers.
Intersection density
Number of pedestrian-accessible street intersections divided by the area within participants’ buffers. Intersections on limited access roads (eg, limited-access highways and on–ramps) were excluded.
Retail and civic land use ratio
Ratio of retail (including food and entertainment) and civic (public buildings) parcel land areas to participants’ buffer areas. These land uses are destinations participants could reach by walking. A value of zero indicates the absence of retail or civic destinations within participants’ buffers, which is typical in predominantly residential neighbourhoods, and a value of 1.0 indicates that retail and civic land uses dominate participants’ buffers.
Public transit density
Number of bus, rail, or ferry stops and stations divided by the land area within participants’ buffers. The complexity was shown by a variety of modes (ie, bus, rail, and ferry) and mode types (eg, regular bus vs bus rapid transit and light vs heavy rail) within and across cities.
Distance to nearest transit
Distance in meters by the street network from participant homes to the nearest stop or station.
Public park density
Number of public parks of any size contained in or intersected by the buffer, divided by the land area within participants’ buffers. A public park was defined as a government designated park of any size that was free of cost and open to the public and maintained by a government agency. Parks included improved or landscaped areas and unimproved or natural areas.
Acknowledgments
All authors declare financial support for this study from the National Cancer Institute (NCI) of the National Institutes of Health (NIH). Data collection in Hong Kong was supported by the HK Research Grants Council GRF grants (HKU740907H and 747807H) and the University of Hong Kong University Research Council Strategic Research Theme (Public Health). US data collection and coordinating centre processing was supported by the NIH grants R01 HL67350 (National Heart, Lung, and Blood Institute) and R01 CA127296 (NCI). The study in Bogota was funded by Colciencias grant 519_2010, Fogarty, and CeiBA. EC was supported by an Australian Research Council Future Fellowship (FT3 140100085). NO was supported by National Health and Medical Research Council Program Grant 569940, NHMRC Senior Principal Research Fellowship 1003960, and by the Victorian Government’s Operational Infrastructure Support Program. The Danish study was partly funded by the Municipality of Aarhus. Data collection in the Czech Republic was supported by the grant Ministry of Education, Youth and Sports (MSM 6198959221). Data collection in New Zealand was supported by the Health Research Council of New Zealand (grant 07/356). Data collection in Mexico was supported by the Centers for Disease Control and Prevention (CDC) Foundation, which received an unrestricted grant from The Coca-Cola Company. The UK study was funded by the Medical Research Council under the National Preventive Research Initiative. DS was supported by a training grant from the CDC Foundation. Kavita A Gavand at University of California, San Diego, CA, USA, supported the preparation and submission of this manuscript.
Footnotes
Declaration of interests
JFS has received grants and personal fees from the Robert Wood Johnson Foundation outside of the present study, grants and non-financial support from Nike outside of this Article, and is a consultant and receiver of royalties from SPARK Programs of School Specialty. LDF is principal owner of Urban Design 4 Health. All other authors declare no competing interests.
References
- 1.Lee IM, Shiroma EJ, Lobelo F, Puska P, Blair SN, Katzmarzyk PT. Effect of physical inactivity on major non-communicable diseases worldwide: and analysis of burden of disease and life expectancy. Lancet 2012; 380: 219–99. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Beaglehole R, Bonita R, Horton R, et al. Priority actions for the non-communicable disease crisis. Lancet 2011; 377: 1438–47. [DOI] [PubMed] [Google Scholar]
- 3.WHO. Draft comprehensive global monitoring framework and targets for the prevention and control of noncommunicable diseases. Sixty-sixth World Health Assembly. March 15, 2013. Geneva: World Health Organization, 2013. [Google Scholar]
- 4.Kohl HW 3rd, Craig CL, Lambert EV, et al. , for The Lancet Physical Activity Series Working Group. The pandemic of physical inactivity: global action for public health. Lancet 2012; 380: 294–305. [DOI] [PubMed] [Google Scholar]
- 5.WHO. Global strategy on diet, physical activity and health. Geneva: World Health Organization. May 2004. http://www.who.int/dietphysicalactivity/en (accessed June 25, 2015). [Google Scholar]
- 6.Bauman AE, Reis RS, Sallis JF, Wells JC, Loos RJF, Martin BW, on behalf of The Lancet Physical Activity Series Working Group. Correlates of physical activity: why are some people physically active and others not? Lancet 2012; 380: 258–71. [DOI] [PubMed] [Google Scholar]
- 7.Heath G, Brownson R, Kruger J, et al. 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: S55–71. [DOI] [PubMed] [Google Scholar]
- 8.Ding D, Gebel K. Built environment, physical activity, and obesity: what have we learned from reviewing the literature? Health Place 2012; 18: 100–05. [DOI] [PubMed] [Google Scholar]
- 9.van Holle V, Deforche B, van Cauwenberg J, et al. Relationship between the physical environment and different domains of physical activity in European adults: a systematic review. BMC Public Health 2012; 12: 807. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Kerr J, Sallis JF, Owen N, et al. Advancing science and policy through a coordinated international study of physical activity and built environments: IPEN methods. J Phys Act Health 2013; 10: 581–601. [DOI] [PubMed] [Google Scholar]
- 11.Frank LD, Sallis JF, Saelens BE, et al. The development of a walkability index: application to the Neighborhood Quality of Life Study. Br J Sports Med 2010; 44: 924–33. [DOI] [PubMed] [Google Scholar]
- 12.Freedson PS, Miller K. Objective monitoring of physical activity using motion sensors and heart rate. Res Q Exerc Sport 2000; 71: S21–29. [PubMed] [Google Scholar]
- 13.Troiano RP, Berrigan D, Dodd KW, Masse LC, Tilert T, McDowell M. Physical activity in the United States measured by accelerometer. Med Sci Sports Exerc 2008; 40: 181–88. [DOI] [PubMed] [Google Scholar]
- 14.Welk GJ. Use of accelerometry-based activity monitors to assess physical activity. In: Welk GJ, ed. Physical activity assessments for health-related research. Champaign: Human Kinetics, 2002: 125–42. [Google Scholar]
- 15.Cain K Accelerometer scoring protocol for the IPEN adult study. University of California; San Diego, CA. 2013. http://www.ipenproject.org/documents/methods_docs/IPEN_Protocol.pdf. (accessed July 1, 2015). [Google Scholar]
- 16.Winkler EA, Paul A, Healy GN, et al. Distinguishing true sedentary from accelerometer non-wearing time: accuracy of two automated wear-time estimations. Med Sci Sports Exerc 2009; 41: 171–72. [Google Scholar]
- 17.Freedson PS, Melanson E, Sirard J. Calibration of the computer science and applications, Inc. accelerometer. Med Sci Sports Exerc 1998; 30: 777–81. [DOI] [PubMed] [Google Scholar]
- 18.Adams MA, Frank LD, Schipperijn J, et al. International variation in neighborhood walkability, transit, and recreation environments using geographic information systems: the IPEN adult study. Int J Health Geogr 2014; 13: 43. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Cerin E, Cain K, Conway TL, et al. Neighborhood environments and objectively measured physical activity in 11 countries. Med Sci Sports Exerc 2014; 46: 2253–64. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Wood SN. Generalized additive models: an introduction with R. Boca Raton: Chapman and Hall, 2006. [Google Scholar]
- 21.Burnham KP, Anderson DR. Model selection and multimodel inference: a practical information-theoretic approach, 2nd edn. New York: Springer, 2002. [Google Scholar]
- 22.WHO. Global recommendations on physical activity for health. Geneva: World Health Organization, 2010. [PubMed] [Google Scholar]
- 23.Sallis JF, Cervero RB, Ascher W, Henderson KA, Kraft MK, Kerr J. An ecological approach to creating more physically active communities. Annu Rev Public Health 2006; 27: 297–322. [DOI] [PubMed] [Google Scholar]
- 24.Gebel K, Bauman AE, Petticrew M. The physical environment and physical activity: a critical appraisal of review articles. Am J Prev Med, 2007. 32: 361–69. [DOI] [PubMed] [Google Scholar]
- 25.Frank LD. Land use and transportation interaction: implications on public health and quality of life. J Plan Educ Res 2000; 20: 6–22. [Google Scholar]
- 26.Shoham DA, Dugas LR, Bovet P, et al. Association of car ownership and physical activity across the spectrum of human development: Modeling the Epidemiologic Transition Study (METS). BMC Public Health 2015; 15: 173. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Cao X, Mokhtarian P, Handy S. Examining the impacts of residential self-selection on travel behavior: a focus on empirical findings. Transp Rev 2009; 29: 359–95. [Google Scholar]
- 28.Ewing R, Bartholomew K, Winkelman S, Walters J, Chen D. Growing cooler: the evidence on urban development and climate change. Washington: Urban Land Institute, 2008. [Google Scholar]
- 29.Sallis JF, Spoon C, Cavill N, et al. Co-benefits of designing communities for active living: an exploration of literature. Int J Behav Nutr Phys Act 2015; 12: 30. [DOI] [PMC free article] [PubMed] [Google Scholar]