Journal of Physical Activity and Health, 2015, 12, 569 -578 http://dx.doi.org/10.1123/jpah.2013-0126 © 2015 Human Kinetics, Inc.
ORIGINAL RESEARCH
Relationship Between Built Environment, Physical Activity, Adiposity, and Health in Adults Aged 46–80 in Shanghai, China Zhang Ying, Liu Dong Ning, and Liu Xin Background: Seldom studies are about the relationship between built environment and physical activity, weight, and health outcome in meso- and microscales. Methods: 1100 residents aged 46 to 80 were recruited from 80 neighborhoods of 13 selected communities of Shanghai, China. An analysis of the relationship between dependent variables (physical activity, Body Mass Index [BMI], overweight/obesity, weight, and health outcomes) and independent variables (involved a geographic-informationsystem-derived measure of built environment) was conducted with hierarchical linear models. Results: Street connectivity was positively associated with physical activity (P < .01). River proximity was inversely related with overweight/obesity (P = .0220). Parkland and square proximity have a significant relationship with physical activity (P = .0270, .0010), BMI (P = .0260, .0130), and overweight/obesity (P = .0020, .0470). Land-use mix was positively associated with physical activity (P < .01) and inversely associated with BMI (P = .0240) and overweight/obesity (P = .0440). Green and open spaces were positively related with BMI (P < .01) and health status (P < .01). For residential style, residents living in a village were more likely to have a lower BMI and overweight/obesity than those living in an urban old or newer residential building. The direct effect of square proximity is much stronger than the indirect effect on BMI through physical activity. Conclusions: The findings can help planners build more pedestrian-friendly communities. They are also useful for creating interventions that are sensitive to possible environmental barriers to physical activity in older adults. Keywords: BMI, obesity, health status The link between physical activity and health problems has been proved such as cognitive functioning,1 psychological barriers,2 depression,3 cardiovascular fitness,4 dyspnea,5 metabolic syndrome,6 hip fractures,7–11 and arthritic pain.12,13 Despite the health benefits of physical activity, recent data from the China Health and Nutrition Surveys found average weekly physical activity among adults in China fell by 32% between 1991–2006.2,14 So it is an essential and critical issue to investigate the relationship between influential factors, physical activity, and related outcomes. There is growing interest in how physical activity, adiposity, and health problems are affected by environmental factors. The horizons have expanded from individual models of behavior to more inclusive ecologic models that recognize the importance of both physical and social environments as determinants of obesity and health15–19. The physical activity, weight, and health are interacted outcomes affected by an array of built environment attributes that belong to urban planning, traffic environment, and residential environment. All these factors would be adopted in the study to make more comprehensive conclusions. Among all the relative studies, the environments can be divided into perceived and objectively measured built environments. The former study is more concentrated on the perceived environment survey methods,20,21 and the latter is more associated with built environments measured objectively with physical activity or health, such as obesity and body mass index (BMI).22 While the relative studies are well documented about Europe and America22–26 with an increasing trend, none have been found in China, especially for middle-aged and elderly Ying ([email protected]) and Ning are with the Dept of Leisure Sports and Arts, Shanghai University of Sport, China. Xin is with the Institute of Sports Science, Shanghai, China.
adults because of less realization. Despite the increase in research on obesity and physical inactivity, there is a paucity of information on built environment factors and their associations with health and physical activity in a population of middle-aged and older adults, which will become the major demographic related to healthcare utilization in the next 20 years in China. According to the report of Shanghai Civil Affairs Bureau,1 by 2020, people aged ≥50 will constitute 30% of the total Chinese population (compared with 10.97% currently, Shanghai is the highest 18.48% of China), and the numbers of those aged ≥60 will more than double from current levels (24.5%). Now, there is no report about the physical activity, weight, or health outcome about middle-aged and older adults in China. The environment and health research scope are mainly focused on physical and chemical properties, environmental pollutants in toxicology and distribution, human health damage mechanism, chronic health risk assessment of the cumulative effect, and so on.27 While with a high-speed urbanization and aging society, more health problems are strongly associated with being overweight and obese28 and are now rising in developing countries at a faster rate than experienced by developed countries.29 With China as the most populous country in the world, declining rates of physical activity among this population will have striking consequences. This paper focuses on the study of Chinese adults aged 46 to 80 in Shanghai and used a cross-sectional, ecologic, and multilevel design. Based on the planning and public health literatures combined with the Chinese features, objectively measured urban planning, traffic environment, and residential environment are included in the built environment attributes in the study. It was postulated that the neighborhood built environment factors specified in this study would independently account for neighborhood-level variation in residents’ levels of being overweight/obese and physically active. On the basis of prior research, the working hypotheses were as follows: 569
570 Ying, Ning, and Xin
1. For physical activity, street connectivity, green, and open spaces have positive relation with it. Availability of facilities, road areas, net residential density, and land-use mix are negatively related with it. The physical activity level of villagers is higher than those people living in urban old building and newer homes. The same relationship is suitable for health status 2. The relationships between environment and BMI and overweight/obesity would both be inverse to physical activity. This article focuses on China as a case study and measured built environment at the community and neighborhood levels. It aims to explore the relationship between environment, adiposity, and health in Chinese middle-aged and older adults.
instructed to wear the pedometer at all times and any places, except when bathing or swimming. Objective anthropometric measures of body weight (in pounds) and height (in inches) were obtained from the study participants. After that, 500 pedometers were employed among 1100 participants on the sequence of 2 batches. Meanwhile, the questionnaires about sociodemographics, health report, and address were delivered to participants in April 2010 to October 2010, and the analysis was conducted in January 2011 to June 2011. After the monitoring period, participants sent the pedometers and questionnaires to the community activity center, from where the research staff would take them. The research protocol was approved by the Sport Scientific Research Institute of Shanghai.
Measures
Downloaded by Washington Univ In St Louis on 09/18/16, Volume 12, Article Number 4
Methods The Study’s Geographic Area Shanghai is located in the estuary of Yangtze River of China. It is the largest industrial and commercial city in China, with 18 administrative districts. Covering an area of 5800 km2, Shanghai approximately has a population of 18.7 million, in which 2 million are floating population. Adults aged >45 accounted for approximately 30% of the population.30
Study Design and Sampling The cross-sectional and ecologic study used data from the Strategies of Environment for Physical Activity of Shanghai, hosted by Shanghai Sports Science Institute under the 2010 World Export background. Three administrative areas were stratified by the downtown, subcivic center, and suburb, which is defined by the China Office of Management and Budget as State 1 to ensure the representativeness of samples. In Stage 2, 13 samples were stratified based on different community types, which are classified as senior neighborhoods (eg, Shikumen houses), public houses, common commodity apartments, expensive commodity apartments, and others. In Stage 3, 80 census tracts were used to stratify the neighborhood population to guarantee a balance in geographic distribution of the study samples. In Stage 4, households were selected randomly within each census tract. The number of households selected was proportional to the percentage of the neighborhood’s total population in that tract. At the last stage, respondents (N = 1100; aged 46 to 80 years) were selected from a stratified random residents sample of households with 5 to 20 per tract, with 40 neighborhoods having samples of 18 or more (median = 8; 25th to 75th interquartile range = 5 to 10) according to the World Health Organization’s middle-age and older adults age division. The target resident population consisted of adults aged 46 to 80 who were not members of any sports team and were independently ambulatory (including cane users) and cognitively intact. Ninety-four percent of the respondents’ home addresses were mapped using a geographic information system (GIS) because 1034 had known places of residence for which built environment indices were available. Hence, it was possible to link built environment indices directly to health-related data for all respondents.
Procedure Selected household residents were initially sent an invitation to participate and were provided with some study details. At the appointed date, research staff met participants at a community activity center. After completing an informed consent form, the participants were
The built environment factors were related to levels of physical activity, obesity, BMI, and health status for a self-designing questionnaire. Hierarchical linear and nonlinear modeling (HLM) methods were used to control for covariates, such as age, gender, and education at the individual level, while examining the effects of environment factors at the population level.
Dependent Measures Physical Activity. Total steps of walking were measured as a physical activity level for its popularity and availability. Total physical activity level in April–October 2010, rather than only 7 consecutive days’ step counts (5 weekdays and 2 weekends), was measured objectively with the Omron HJ-720ITC Pedometer (OMRON Inc., China), which can store step counts and aerobic movement for 41 days.31 Incentives ranging up to 20RMB were paid upon receipt of the equipment with valid data. The monitor has a unique dual-sensor technology and can be carried in a pocket or a bag for more accurate counts, instead of the traditional pedometer, which is only attached to the belt.32 BMI and Overweight/Obesity. Beside the physical activity, 2 weight-related measures were included as outcome variables: BMI and overweight/obesity. Objective anthropometric measures of participants’ stature and body mass were assessed to calculate BMI. Nonoverweight and overweight were defined using China age- and gender-referenced cutoff points. BMI (kg/m2) was calculated from measurements of stature (to the nearest 0.1 cm) and body mass (to the nearest 0.1 kg). Overweight/obesity was assigned to tw2o categories: 1 = overweight or obese (BMI ≥ 24); 0 = otherwise (BMI < 24). All demonstrations were video recorded and subsequently examined by 1 trained assessor for the presence or absence of process characteristics for each skill and scored appropriately. Health Status. A health status variable was also modeled for its known relationship with inactivity and obesity. Health status was self-reported according to both of identified by hospital or selffeeling; therefore, it has potential for measurement bias. However, it has been shown that health status was highly correlated with body and self-dissatisfaction for males and females.33 Health status employed a 5-point response option by asking, “How do you feel about your health status?” with the responses being 1 (worst), 2 (poorer), 3 (fine), 4 (better), and 5 (best).
Sociodemographics Characteristics Data obtained from individual participants included age, gender, employment status, and education. Unless otherwise noted, the sociodemographics were included in models as individual-level covariates.
JPAH Vol. 12, No. 4, 2015
Environment and Physical Activity-Related Outcome 571
Age is a continuous variable; gender is a dichotomous variable that is assigned to 2 categories: male (1 score) and female (0 score). Education and employment are both categorical variables. The former is assigned to 4 categories: 1 = less than primary school, 2 = high school graduate, 3 = senior high school, and 4 = college graduate. The latter is assigned to 3 categories: 1 = working (either part time or full time), 2 = retired, and 3 = unemployed.
Independent Measures
Downloaded by Washington Univ In St Louis on 09/18/16, Volume 12, Article Number 4
Buffer Definition. An earlier study34 found that older women
rarely walked to destinations more than a 20-minute walk away. It is about 1500 m, according to the common adult gait speed at 1.2 m per second.35,36 But in consideration of the somatotype difference between Western and Oriental people, it was hypothesized that a closer distance from one’s home might be more influential on an individual’s physical activity choice.37 Meanwhile, urban infrastructure in Shanghai usually was constructed in 500-m network distance from the participant’s residence. So 500 m was defined as the residents’ activity buffer radius. Figure 1 conveys the 500-m network buffer size around a household within a living environment.
Data Source. Regional land-use data (land-use type, residential type, facility proximity), street network data (road type, width, length), and census data were provided by Roads and Traffic
Authority, Land and Resources Bureau, and Public Security Bureau, respectively. Census-tract-level population data were obtained from the Neighborhood Committee of every community. All these data were spatially integrated within a GIS, using ArcView software (10.0) to characterize the built environment of the sampled study areas. Variable Construction. With the frequency analysis method
combining with the variable data obtainment possibility of this study, the top highest frequently used variables that had been related to physical activity, weight, or health outcome in some previous study were included; they are land-use mix, net residential density, and street connectivity.22,38–40 Beside these, other environment variables such as proximity of river, parkland, and square were also included instead of other facilities, like fast-food outlets16 or physical activity recreation facilities.39 Because the river, parkland, and square are the most common characteristics around Chinese residential buildings and much population of China with few physical activity recreation facility resources is another reason. Road areas in this study were included in the index instead of those that had been adopted in previous study such as road type,39 density of public transit stations,22 traffic counts, and speed limits,41 because of difficult obtainment of the data in China. To construct the index, these 9 neighborhood variables from the smaller set were rated in 3 dimensions, namely traffic environment,
Figure 1 — Neighborhood environments within 500-m buffer. JPAH Vol. 12, No. 4, 2015
Downloaded by Washington Univ In St Louis on 09/18/16, Volume 12, Article Number 4
572 Ying, Ning, and Xin
urban planning, and residential environment. As for each dimension, a composite factor was extracted from several observed variables via principal components analysis (PCA). The 9 variables all are modeled; thus, the 3 composite factors only were used to describe the variables conveniently in this study. More details about the derivation of these indices are described in the “Data Analysis” section. The 1 on which the observed variables loaded most heavily (Table 1) used the developed PCA method, which has been used by Wenlin Zhang.42
Block groups allow census data to be used for population estimates and are a far more accurate data source than data available at the 1-km buffer level, which is expected to have an inverse impact on obesity.44–46 The second variable is residential style. According to the land-use database, 3 residential styles were identified: natural villages, urban old residential buildings built before 1980 (a proxy for older neighborhoods), and newer residential buildings were assigned 1, 2, and 3 scores, respectively.
Traffic Environment. Three variables made up this factor. The first
Data Analysis
is street connectivity, defined as the number of street intersections, including those with traffic lights and those without, divided by area in km2 within the 500-m buffer.43 The second variable is availability of facilities, which refers to the river, parkland, and square distance (m) from the household in the 500-m buffer. The 3 places were seldom taken as independent variables in previous studies, but they are the most common characteristics of the Chinese building environment. The final variable is road areas, defined as the total road areas (m2) divided by population in the surveyed census tract. Urban Planning. Two variables made up this factor. The first
variable is green and open spaces. The index has been adopted by Fuzhong Li22 and is the total m2 of green and open spaces areas (including public parklands areas) in a 500-m buffer for recreation divided by population in the surveyed census tract. The second variable, land-use mix, follows this equation: LUM = − ∑
n p ln pi / ln n i =1 i
, where pi is the proportion of estimated square footage attributed to land use i, and n is the number of land uses. Land-use mix was operationalized with the method provided by Frank et al,26 who provide a measure of the evenness of distribution of several land-use types (ie, residential, commercial, office, and institutional) within the study’s geographic area. Values near 0 reflect single-use environments, while values near 1 reflect maximal mixed usage. Here, according to land-use types set by the Land and Resources Bureau of China, it was divided into 10 land-use types, namely industrial, traffic, public building, residential, storehouse, harbor, green-space, river, square, and unused land. Residential Environment. Two variables made up this factor. The
first is net residential density, which equals the number of persons per residential hectare within the household’s census block group.
HLM uses the method of maximum likelihood to optimally combine information from different samples. In this study, neighborhoods with small samples contributed less information to the estimation of parameters than neighborhoods with large samples.47 Because maximum likelihood took into account the information from each neighborhood, and because the number of neighborhoods in this study was larger, neighborhoods with small samples were not problematic from a statistical standpoint.48,49 Multicollinearity problems should be considered before statistical analysis, which was seldom considered in previous studies.16 The problem also has been usually dealt with by 3 methods in some studies: 1. The multicollinearity problem was not considered22 2. Independent variables were eligible to enter and were retained in the final model if the associated t test in Phase 1 of the analysis was significant at the 0.05 level.41 The result could be biased from statistics point. 3. Factors were adjusted to find the greatest explanatory power of the variation of optimal model to weight the variable through the linear regression model, which cannot measure the individual- and neighborhood-level variables simultaneously to avoid the problem. Then the independent variables were weighted equally and combined into an overall index44,47,50 by exploratory factor analysis. PCA was employed to weight the independent variables to avoid a multicollinearity problem in this study. The central idea of PCA is to reduce the dimensionality of a data set consisting of a large number of interrelated variables, while retaining as much as possible of the variation present in the data set. This is achieved by transforming to a new set of variables, the principal components, which are uncorrelated and ordered so that the first few retain
Table 1 Environment Index Variables and Factor Loadings Array
Observed variable
Traffic environment
Street connectivity Availability of facilities (m) Road areas (m2)
0.217 –0.157 0.246
Urban planning
Green and open spaces (m2) Land-use mix
0.200 0.072
Residential environment
Net residential density (person/hectare) Residential style
0.077 –0.08
JPAH Vol. 12, No. 4, 2015
Factor loading
Environment and Physical Activity-Related Outcome 573
Downloaded by Washington Univ In St Louis on 09/18/16, Volume 12, Article Number 4
most of the variation present in all of the original variables.50,51 The Kaiser–Meyer–Olkin (KMO) Measure of Sampling Adequacy index (valued from 0 to 1) is an important standard to judge if the variables are suitable to use the PCA method. The closer the value is to 1, the more correlation there is between variables and the more appropriate it is to use PCA.51 The KMO metric was defined by Kaiser52: >0.9 = the data are very appropriate for adopting PCA method, 0.8–0.9 = more appropriate, 0.7–0.8 = appropriate, 0.6–0.7 = not very appropriate, and
ORIGINAL RESEARCH
Relationship Between Built Environment, Physical Activity, Adiposity, and Health in Adults Aged 46–80 in Shanghai, China Zhang Ying, Liu Dong Ning, and Liu Xin Background: Seldom studies are about the relationship between built environment and physical activity, weight, and health outcome in meso- and microscales. Methods: 1100 residents aged 46 to 80 were recruited from 80 neighborhoods of 13 selected communities of Shanghai, China. An analysis of the relationship between dependent variables (physical activity, Body Mass Index [BMI], overweight/obesity, weight, and health outcomes) and independent variables (involved a geographic-informationsystem-derived measure of built environment) was conducted with hierarchical linear models. Results: Street connectivity was positively associated with physical activity (P < .01). River proximity was inversely related with overweight/obesity (P = .0220). Parkland and square proximity have a significant relationship with physical activity (P = .0270, .0010), BMI (P = .0260, .0130), and overweight/obesity (P = .0020, .0470). Land-use mix was positively associated with physical activity (P < .01) and inversely associated with BMI (P = .0240) and overweight/obesity (P = .0440). Green and open spaces were positively related with BMI (P < .01) and health status (P < .01). For residential style, residents living in a village were more likely to have a lower BMI and overweight/obesity than those living in an urban old or newer residential building. The direct effect of square proximity is much stronger than the indirect effect on BMI through physical activity. Conclusions: The findings can help planners build more pedestrian-friendly communities. They are also useful for creating interventions that are sensitive to possible environmental barriers to physical activity in older adults. Keywords: BMI, obesity, health status The link between physical activity and health problems has been proved such as cognitive functioning,1 psychological barriers,2 depression,3 cardiovascular fitness,4 dyspnea,5 metabolic syndrome,6 hip fractures,7–11 and arthritic pain.12,13 Despite the health benefits of physical activity, recent data from the China Health and Nutrition Surveys found average weekly physical activity among adults in China fell by 32% between 1991–2006.2,14 So it is an essential and critical issue to investigate the relationship between influential factors, physical activity, and related outcomes. There is growing interest in how physical activity, adiposity, and health problems are affected by environmental factors. The horizons have expanded from individual models of behavior to more inclusive ecologic models that recognize the importance of both physical and social environments as determinants of obesity and health15–19. The physical activity, weight, and health are interacted outcomes affected by an array of built environment attributes that belong to urban planning, traffic environment, and residential environment. All these factors would be adopted in the study to make more comprehensive conclusions. Among all the relative studies, the environments can be divided into perceived and objectively measured built environments. The former study is more concentrated on the perceived environment survey methods,20,21 and the latter is more associated with built environments measured objectively with physical activity or health, such as obesity and body mass index (BMI).22 While the relative studies are well documented about Europe and America22–26 with an increasing trend, none have been found in China, especially for middle-aged and elderly Ying ([email protected]) and Ning are with the Dept of Leisure Sports and Arts, Shanghai University of Sport, China. Xin is with the Institute of Sports Science, Shanghai, China.
adults because of less realization. Despite the increase in research on obesity and physical inactivity, there is a paucity of information on built environment factors and their associations with health and physical activity in a population of middle-aged and older adults, which will become the major demographic related to healthcare utilization in the next 20 years in China. According to the report of Shanghai Civil Affairs Bureau,1 by 2020, people aged ≥50 will constitute 30% of the total Chinese population (compared with 10.97% currently, Shanghai is the highest 18.48% of China), and the numbers of those aged ≥60 will more than double from current levels (24.5%). Now, there is no report about the physical activity, weight, or health outcome about middle-aged and older adults in China. The environment and health research scope are mainly focused on physical and chemical properties, environmental pollutants in toxicology and distribution, human health damage mechanism, chronic health risk assessment of the cumulative effect, and so on.27 While with a high-speed urbanization and aging society, more health problems are strongly associated with being overweight and obese28 and are now rising in developing countries at a faster rate than experienced by developed countries.29 With China as the most populous country in the world, declining rates of physical activity among this population will have striking consequences. This paper focuses on the study of Chinese adults aged 46 to 80 in Shanghai and used a cross-sectional, ecologic, and multilevel design. Based on the planning and public health literatures combined with the Chinese features, objectively measured urban planning, traffic environment, and residential environment are included in the built environment attributes in the study. It was postulated that the neighborhood built environment factors specified in this study would independently account for neighborhood-level variation in residents’ levels of being overweight/obese and physically active. On the basis of prior research, the working hypotheses were as follows: 569
570 Ying, Ning, and Xin
1. For physical activity, street connectivity, green, and open spaces have positive relation with it. Availability of facilities, road areas, net residential density, and land-use mix are negatively related with it. The physical activity level of villagers is higher than those people living in urban old building and newer homes. The same relationship is suitable for health status 2. The relationships between environment and BMI and overweight/obesity would both be inverse to physical activity. This article focuses on China as a case study and measured built environment at the community and neighborhood levels. It aims to explore the relationship between environment, adiposity, and health in Chinese middle-aged and older adults.
instructed to wear the pedometer at all times and any places, except when bathing or swimming. Objective anthropometric measures of body weight (in pounds) and height (in inches) were obtained from the study participants. After that, 500 pedometers were employed among 1100 participants on the sequence of 2 batches. Meanwhile, the questionnaires about sociodemographics, health report, and address were delivered to participants in April 2010 to October 2010, and the analysis was conducted in January 2011 to June 2011. After the monitoring period, participants sent the pedometers and questionnaires to the community activity center, from where the research staff would take them. The research protocol was approved by the Sport Scientific Research Institute of Shanghai.
Measures
Downloaded by Washington Univ In St Louis on 09/18/16, Volume 12, Article Number 4
Methods The Study’s Geographic Area Shanghai is located in the estuary of Yangtze River of China. It is the largest industrial and commercial city in China, with 18 administrative districts. Covering an area of 5800 km2, Shanghai approximately has a population of 18.7 million, in which 2 million are floating population. Adults aged >45 accounted for approximately 30% of the population.30
Study Design and Sampling The cross-sectional and ecologic study used data from the Strategies of Environment for Physical Activity of Shanghai, hosted by Shanghai Sports Science Institute under the 2010 World Export background. Three administrative areas were stratified by the downtown, subcivic center, and suburb, which is defined by the China Office of Management and Budget as State 1 to ensure the representativeness of samples. In Stage 2, 13 samples were stratified based on different community types, which are classified as senior neighborhoods (eg, Shikumen houses), public houses, common commodity apartments, expensive commodity apartments, and others. In Stage 3, 80 census tracts were used to stratify the neighborhood population to guarantee a balance in geographic distribution of the study samples. In Stage 4, households were selected randomly within each census tract. The number of households selected was proportional to the percentage of the neighborhood’s total population in that tract. At the last stage, respondents (N = 1100; aged 46 to 80 years) were selected from a stratified random residents sample of households with 5 to 20 per tract, with 40 neighborhoods having samples of 18 or more (median = 8; 25th to 75th interquartile range = 5 to 10) according to the World Health Organization’s middle-age and older adults age division. The target resident population consisted of adults aged 46 to 80 who were not members of any sports team and were independently ambulatory (including cane users) and cognitively intact. Ninety-four percent of the respondents’ home addresses were mapped using a geographic information system (GIS) because 1034 had known places of residence for which built environment indices were available. Hence, it was possible to link built environment indices directly to health-related data for all respondents.
Procedure Selected household residents were initially sent an invitation to participate and were provided with some study details. At the appointed date, research staff met participants at a community activity center. After completing an informed consent form, the participants were
The built environment factors were related to levels of physical activity, obesity, BMI, and health status for a self-designing questionnaire. Hierarchical linear and nonlinear modeling (HLM) methods were used to control for covariates, such as age, gender, and education at the individual level, while examining the effects of environment factors at the population level.
Dependent Measures Physical Activity. Total steps of walking were measured as a physical activity level for its popularity and availability. Total physical activity level in April–October 2010, rather than only 7 consecutive days’ step counts (5 weekdays and 2 weekends), was measured objectively with the Omron HJ-720ITC Pedometer (OMRON Inc., China), which can store step counts and aerobic movement for 41 days.31 Incentives ranging up to 20RMB were paid upon receipt of the equipment with valid data. The monitor has a unique dual-sensor technology and can be carried in a pocket or a bag for more accurate counts, instead of the traditional pedometer, which is only attached to the belt.32 BMI and Overweight/Obesity. Beside the physical activity, 2 weight-related measures were included as outcome variables: BMI and overweight/obesity. Objective anthropometric measures of participants’ stature and body mass were assessed to calculate BMI. Nonoverweight and overweight were defined using China age- and gender-referenced cutoff points. BMI (kg/m2) was calculated from measurements of stature (to the nearest 0.1 cm) and body mass (to the nearest 0.1 kg). Overweight/obesity was assigned to tw2o categories: 1 = overweight or obese (BMI ≥ 24); 0 = otherwise (BMI < 24). All demonstrations were video recorded and subsequently examined by 1 trained assessor for the presence or absence of process characteristics for each skill and scored appropriately. Health Status. A health status variable was also modeled for its known relationship with inactivity and obesity. Health status was self-reported according to both of identified by hospital or selffeeling; therefore, it has potential for measurement bias. However, it has been shown that health status was highly correlated with body and self-dissatisfaction for males and females.33 Health status employed a 5-point response option by asking, “How do you feel about your health status?” with the responses being 1 (worst), 2 (poorer), 3 (fine), 4 (better), and 5 (best).
Sociodemographics Characteristics Data obtained from individual participants included age, gender, employment status, and education. Unless otherwise noted, the sociodemographics were included in models as individual-level covariates.
JPAH Vol. 12, No. 4, 2015
Environment and Physical Activity-Related Outcome 571
Age is a continuous variable; gender is a dichotomous variable that is assigned to 2 categories: male (1 score) and female (0 score). Education and employment are both categorical variables. The former is assigned to 4 categories: 1 = less than primary school, 2 = high school graduate, 3 = senior high school, and 4 = college graduate. The latter is assigned to 3 categories: 1 = working (either part time or full time), 2 = retired, and 3 = unemployed.
Independent Measures
Downloaded by Washington Univ In St Louis on 09/18/16, Volume 12, Article Number 4
Buffer Definition. An earlier study34 found that older women
rarely walked to destinations more than a 20-minute walk away. It is about 1500 m, according to the common adult gait speed at 1.2 m per second.35,36 But in consideration of the somatotype difference between Western and Oriental people, it was hypothesized that a closer distance from one’s home might be more influential on an individual’s physical activity choice.37 Meanwhile, urban infrastructure in Shanghai usually was constructed in 500-m network distance from the participant’s residence. So 500 m was defined as the residents’ activity buffer radius. Figure 1 conveys the 500-m network buffer size around a household within a living environment.
Data Source. Regional land-use data (land-use type, residential type, facility proximity), street network data (road type, width, length), and census data were provided by Roads and Traffic
Authority, Land and Resources Bureau, and Public Security Bureau, respectively. Census-tract-level population data were obtained from the Neighborhood Committee of every community. All these data were spatially integrated within a GIS, using ArcView software (10.0) to characterize the built environment of the sampled study areas. Variable Construction. With the frequency analysis method
combining with the variable data obtainment possibility of this study, the top highest frequently used variables that had been related to physical activity, weight, or health outcome in some previous study were included; they are land-use mix, net residential density, and street connectivity.22,38–40 Beside these, other environment variables such as proximity of river, parkland, and square were also included instead of other facilities, like fast-food outlets16 or physical activity recreation facilities.39 Because the river, parkland, and square are the most common characteristics around Chinese residential buildings and much population of China with few physical activity recreation facility resources is another reason. Road areas in this study were included in the index instead of those that had been adopted in previous study such as road type,39 density of public transit stations,22 traffic counts, and speed limits,41 because of difficult obtainment of the data in China. To construct the index, these 9 neighborhood variables from the smaller set were rated in 3 dimensions, namely traffic environment,
Figure 1 — Neighborhood environments within 500-m buffer. JPAH Vol. 12, No. 4, 2015
Downloaded by Washington Univ In St Louis on 09/18/16, Volume 12, Article Number 4
572 Ying, Ning, and Xin
urban planning, and residential environment. As for each dimension, a composite factor was extracted from several observed variables via principal components analysis (PCA). The 9 variables all are modeled; thus, the 3 composite factors only were used to describe the variables conveniently in this study. More details about the derivation of these indices are described in the “Data Analysis” section. The 1 on which the observed variables loaded most heavily (Table 1) used the developed PCA method, which has been used by Wenlin Zhang.42
Block groups allow census data to be used for population estimates and are a far more accurate data source than data available at the 1-km buffer level, which is expected to have an inverse impact on obesity.44–46 The second variable is residential style. According to the land-use database, 3 residential styles were identified: natural villages, urban old residential buildings built before 1980 (a proxy for older neighborhoods), and newer residential buildings were assigned 1, 2, and 3 scores, respectively.
Traffic Environment. Three variables made up this factor. The first
Data Analysis
is street connectivity, defined as the number of street intersections, including those with traffic lights and those without, divided by area in km2 within the 500-m buffer.43 The second variable is availability of facilities, which refers to the river, parkland, and square distance (m) from the household in the 500-m buffer. The 3 places were seldom taken as independent variables in previous studies, but they are the most common characteristics of the Chinese building environment. The final variable is road areas, defined as the total road areas (m2) divided by population in the surveyed census tract. Urban Planning. Two variables made up this factor. The first
variable is green and open spaces. The index has been adopted by Fuzhong Li22 and is the total m2 of green and open spaces areas (including public parklands areas) in a 500-m buffer for recreation divided by population in the surveyed census tract. The second variable, land-use mix, follows this equation: LUM = − ∑
n p ln pi / ln n i =1 i
, where pi is the proportion of estimated square footage attributed to land use i, and n is the number of land uses. Land-use mix was operationalized with the method provided by Frank et al,26 who provide a measure of the evenness of distribution of several land-use types (ie, residential, commercial, office, and institutional) within the study’s geographic area. Values near 0 reflect single-use environments, while values near 1 reflect maximal mixed usage. Here, according to land-use types set by the Land and Resources Bureau of China, it was divided into 10 land-use types, namely industrial, traffic, public building, residential, storehouse, harbor, green-space, river, square, and unused land. Residential Environment. Two variables made up this factor. The
first is net residential density, which equals the number of persons per residential hectare within the household’s census block group.
HLM uses the method of maximum likelihood to optimally combine information from different samples. In this study, neighborhoods with small samples contributed less information to the estimation of parameters than neighborhoods with large samples.47 Because maximum likelihood took into account the information from each neighborhood, and because the number of neighborhoods in this study was larger, neighborhoods with small samples were not problematic from a statistical standpoint.48,49 Multicollinearity problems should be considered before statistical analysis, which was seldom considered in previous studies.16 The problem also has been usually dealt with by 3 methods in some studies: 1. The multicollinearity problem was not considered22 2. Independent variables were eligible to enter and were retained in the final model if the associated t test in Phase 1 of the analysis was significant at the 0.05 level.41 The result could be biased from statistics point. 3. Factors were adjusted to find the greatest explanatory power of the variation of optimal model to weight the variable through the linear regression model, which cannot measure the individual- and neighborhood-level variables simultaneously to avoid the problem. Then the independent variables were weighted equally and combined into an overall index44,47,50 by exploratory factor analysis. PCA was employed to weight the independent variables to avoid a multicollinearity problem in this study. The central idea of PCA is to reduce the dimensionality of a data set consisting of a large number of interrelated variables, while retaining as much as possible of the variation present in the data set. This is achieved by transforming to a new set of variables, the principal components, which are uncorrelated and ordered so that the first few retain
Table 1 Environment Index Variables and Factor Loadings Array
Observed variable
Traffic environment
Street connectivity Availability of facilities (m) Road areas (m2)
0.217 –0.157 0.246
Urban planning
Green and open spaces (m2) Land-use mix
0.200 0.072
Residential environment
Net residential density (person/hectare) Residential style
0.077 –0.08
JPAH Vol. 12, No. 4, 2015
Factor loading
Environment and Physical Activity-Related Outcome 573
Downloaded by Washington Univ In St Louis on 09/18/16, Volume 12, Article Number 4
most of the variation present in all of the original variables.50,51 The Kaiser–Meyer–Olkin (KMO) Measure of Sampling Adequacy index (valued from 0 to 1) is an important standard to judge if the variables are suitable to use the PCA method. The closer the value is to 1, the more correlation there is between variables and the more appropriate it is to use PCA.51 The KMO metric was defined by Kaiser52: >0.9 = the data are very appropriate for adopting PCA method, 0.8–0.9 = more appropriate, 0.7–0.8 = appropriate, 0.6–0.7 = not very appropriate, and
