Abstract
Background
Studies of the built environment and physical activity have implicitly assumed that a substantial amount of activity occurs near home, but in fact the location is unknown.
Purpose
Examine associations between built environment variables within home and work buffers and moderate-vigorous physical activity occurring within these locations.
Methods
Adults (n= 148) from Massachusetts wore an accelerometer and GPS unit for up to four days. Moderate and vigorous intensity activity was quantified within 50 m and 1 km home and work buffers. Multiple regression models were used to examine associations between five objective built environment variables within 1 km home and work buffers (intersection density, land use mix, population and housing unit density, vegetation index) and moderate-vigorous physical activity within those areas.
Results
The mean daily minutes of moderate-vigorous physical activity accumulated in all locations = 61.1 ± 32.8, while duration within 1 km home buffers = 14.0 ± 16.4 min. Intersection density, land use mix, and population and housing unit density within 1 km home buffers were positively associated with moderate-vigorous physical activity in the buffer, while a vegetation index showed an inverse relationship (all p< 0.05). None of these variables showed associations with total moderate-vigorous activity. Within 1 km of work, only population and housing unit density were significantly associated with moderate-vigorous physical activity within the buffer.
Conclusions
Findings are consistent with studies showing that certain attributes of the built environment around homes are positively related to physical activity, but in this case only when the outcome was “location-based”. Simultaneous accelerometer-GPS monitoring shows promise as a method to improve understanding of how the built environment influences physical activity behaviors by allowing activity to be quantified in a range of physical contexts and thereby provide a more explicit link between physical activity outcomes and built environment exposures.
Introduction
Physical inactivity continues to be a significant public health problem among the U.S. population and in numerous other countries. This is true despite a shift in behavior change paradigms from one dominated by a focus on psychological factors and individual responsibility to one recognizing that environmental factors are important in shaping healthy behaviors of populations, such as physical activity.1–3 A growing body of evidence demonstrates positive associations between characteristics of neighborhood built environments, including higher levels of land use mix, population and residential density, and street connectivity, and participation in recreational and utilitarian physical activity.4–6 Despite a rapid growth in this research and improved methods of measuring both the built environment (e.g., via geographic information systems (GIS)) and physical activity outcomes (e.g., via accelerometers)—the current evidence base is still emerging. One area needing more work is use of an activity monitoring approach that provides a more precise spatial match between built environment exposures and physical activity outcomes—an approach that spatially contextualizes physical activity behaviors.7
Current evidence has not yet revealed the dynamic interactions between individuals, their environment, and their physical activity behaviors over time and space. Recent studies have examined relationships between objective built environment variables and objectively measured physical activity using an analytic approach that assumes activity occurs within a designated area surrounding a residential address8 or within another defined neighborhood area.9 The lack of specificity with respect to where physical activity occurred in these studies may contribute to dilution of the observed associations, resulting in an inability to observe true associations or an underestimation of the strength of real associations.
To address this limitation, a sample of free-living adults was monitored with accelerometer and global positioning system (GPS) devices. This study builds on previous research by Rodriguez and colleagues7 by examining a larger sample over a longer monitoring period and using different methodological approaches. The aims of this cross-sectional study were to: 1) quantify moderate-vigorous physical activity occurring within buffers around home and work locations using accelerometer and GPS data; and 2) examine associations between built environment characteristics and moderate-vigorous physical activity within the buffers (i.e., “location-based” physical activity).
Methods
Participants and Recruitment Procedures
Participants were adults (19–78 years of age) who completed brief (5 min) intercept surveys at one of five trails in eastern Massachusetts, were either walking, jogging/running, bicycling or in-line skating, reported using the trail at least four times during the previous 4 weeks, and agreed to wear an accelerometer and GPS unit for four days. The sampling frame was limited to trail users because this project was a follow-up to a previous study examining physical characteristics of trails.10
The specific trail location, time of day, and day of the week for intercepts were systematically varied. Surveys were conducted on a minimum of two weekdays and two weekend days during the fall of 2004 and spring and summer of 2005. Among 1194 trail users surveyed, 294 individuals (24.6%) initially agreed to participate in activity monitoring and provided contact information. Equipment was deployed to 178 individuals (14.9% of those surveyed). Among 116 individuals who provided contact information, but did not participate, the primary reasons were schedule conflicts, loss of interest, or inability to re-contact them. Methods for this study were approved by the Human Subjects Committee at the Harvard School of Public Health. Participants in the activity monitoring signed an informed consent form.
Equipment
The Actigraph™ accelerometer (model 7164) is a small, lightweight (42.6g) uniaxial activity monitor that captures vertical acceleration and stores acceleration data as dimensionless units11 referred to as activity counts (cts). Previous studies in both the laboratory and field have established the validity of the Actigraph for measuring the volume, intensity and temporal patterns of activities such as walking and running.12–14 Accelerometers were initialized to collect data in one-minute epochs. After each deployment, accelerometers were checked using the manufacturer’s calibration device and recalibrated as necessary. Participants were instructed to wear the monitor on their right hip using an adjustable nylon belt.
The GeoStats (Atlanta, GA) Wearable GeoLogger™ is a GPS device designed for collection of detailed travel data. Components included a data logger that recorded position and speed, a rechargeable battery and a patch antenna (total weight ~ 0.45 kg). Small backpacks were purchased to house the units and the antenna was fastened to one of the shoulder straps (Figure 1). The GPS units were initialized to spatial coordinates within the metropolitan Boston area and were initially programmed to collect data at five-second intervals. Subsequently, when a longer-lasting battery became available, this interval was changed to one second.
Figure 1.
GPS unit with receiver on shoulder strap.
Data Collection
Research staff met participants at a public location (e.g., library, town hall, coffee shop) 1–3 days before activity monitoring began. Participants completed an informed consent form, were instructed on how to wear the equipment, and were given a daily log sheet for both devices. Participants were instructed to wear the accelerometer at all times, except when sleeping, bathing, or swimming, for four consecutive days (2 weekend days and 2 weekdays). They were also told to wear the GPS unit anytime they were outdoors, regardless of whether they were being physically active (e.g., walking) or traveling in a car, train, or bus. After the monitoring period, research staff met participants within 1–3 days to collect equipment, review log sheets, and have participants complete the International Physical Activity Questionnaire.15
Data Processing
Accelerometer and GPS data were downloaded using Actigraph and GeoStats software, respectively. For each participant, an analyst manually reviewed animations of the raw GPS data over the 4-day monitoring period and identified outlying points and discontinuities. Outliers resulting from poor GPS signals and multiple points clouding around stops were removed from the database. After this step, a processor was applied to aggregate the GPS points into one-minute intervals that temporally aligned with the accelerometer data. GPS and accelerometer data were merged using their respective time stamps and processed into a database with one record for each minute of activity. The precision of the clock systems and synchronicity of the GPS and accelerometer data that allowed for merging were described previously.16 GPS coordinates for each monitoring minute (starting and ending latitude and longitude) were imputed prospectively in cases where a missing GPS reading followed a monitoring minute with “actual” (non-imputed) coordinates. This last known GPS point position was maintained for each minute until a new reading was obtained.
Physical activity summary variables were created for monitor wearing time (min/d), average counts per minute per day (ct/min/d) and for time (min) spent in moderate (1952–5724 ct/min) and vigorous activity (≥ 5725 ct/min)12. Valid days of monitoring were based on both accelerometer and GPS criteria. A valid accelerometer day was defined as ≥ 600 minutes of wear time, determined using the algorithm developed by NHANES.17,18 To establish criteria for a valid GPS day, the distribution of non-imputed GPS readings for 642 participant-days of monitoring that had valid GPS readings were examined. GPS monitoring time ranged from 2 to 601 minutes per day, with mean and median values of 130.7 ± 90.1 and 113.0 minutes, respectively. A minimum value equivalent to 1 standard deviation below the mean (i.e., ≥ 40 minutes) was selected as the criterion for a valid GPS day. Out of 178 participants, 151 met the accelerometer and GPS criteria. Three additional individuals were excluded because their home or work address was missing or was not considered valid, resulting in a final sample = 148.
GIS Processing: Location-Based Physical Activity Variables
A GIS road layer (StreetMap 2005, Environmental Systems Research Institute, Redlands, CA) was used for geocoding addresses and to create network buffers. In total, 174 home addresses and 87 work addresses were successfully located using standard geocoding procedures (out of 174 and 96 participants who provided these addresses, respectively). Participants who provided work addresses reported only one work site; thereby obviating the need to geocode and quantify activity within multiple work buffers. Areas within a 1 km road network distance of home and work locations were delineated using network polygon buffers. This buffer size is consistent with recent studies.8,19,20 Turning restrictions applicable to vehicular traffic were not used in the network buffer algorithm because the intention was to model pedestrian movement. Interstates, major highways, and off-ramps were excluded from the street network.
GIS procedures were used to create dichotomous (yes/no) location variables that characterized where a participant was during each monitoring minute. The last GPS point recorded during each minute was used to represent the spatial location associated with the corresponding accelerometer reading. Minutes occurring within 50 m of home and work addresses were identified based on straight-line distance from geocoded addresses to the GPS coordinates (point) for a given minute. These variables were used to estimate activity occurring indoors at home and work or within close proximity to these locations. Spatial queries were also applied to create location variables that identified whether GPS coordinates occurred within a participant’s home or work 1 km buffer. Location variables for the 50 m and 1 km buffers were mutually exclusive. If a given minute was identified as falling within the 50 m buffer, it was coded as not in the 1 km buffer. Location-based physical activity variables were then created using the GPS and accelerometer data for each monitoring minute. Summary activity variables were created for each of the four buffers examined (50 m home and work, 1 km home and work). An illustration of location-based physical activity occurring near one participant’s home and work is shown in Figure 2.
Figure 2.
Location-based physical activity for participant over 4 days (A) with inset showing activity around home (B)
GIS Processing: Built Environment Variables
Within the 1 km home and work buffers, five built environment variables that have been used in previous studies were created4,6,21–25: intersection density (connectivity), land use mix, residential population density, housing unit density, and a vegetation index (see Table 1). Intersection density was defined as the number of intersections within the network buffer divided by the total street segment length within that buffer (intersections per km). Land use mix was estimated using 1999 land use data provided by the Massachusetts Office of Geographic and Environmental Information (MassGIS http://www.mass.gov/mgis/). Four categories of land use (residential, commercial, recreational, urban public) were used in an entropy formula developed by Frank and colleagues.19 Population and housing unit density within network buffers were estimated from Census 2000 data at the block group level using methods modified from Forsyth.26 Greenness within network buffers was estimated using a satellite image of the study region captured on 27 September 2000 by the Landsat Enhanced Thematic Mapper Plus sensor. The normalized difference vegetation index (NDVI) is commonly used in environmental remote sensing applications and has been shown to be a significant predictor of plant health, percent vegetated ground cover, and photosynthetic green biomass.27 NDVI is a unitless metric that compares reflectance values in satellite remote sensing measurements. NDVI values range between +/− 1, with higher values indicative of healthy green vegetation and low values characteristic of non-vegetated land cover. Greenness was estimated as the mean NDVI within network buffers following methods used in previous studies that found this variable was a significant predictor of smaller increases in child BMI24 and increased pedestrian trail traffic.23
Table 1.
Operational definitions for built environment variables within 1 km home and work network buffers
| Built environment domain |
Specific variable | Operational definition and data source(s) |
|---|---|---|
| Street connectivity | Intersection density (intersections per km) | Number of intersections within network buffer divided by total street segment length (km) within buffer. Intersections and street segment lengths obtained from ESRI Street Map 2005 Data. |
| Land use mix | Land use mix | Land use data from MA GIS 1999 aerial photography (36 categories collapsed into 5 categories) – measure of evenness of uses across 5 categories |
| Density | Residential population density | Number of persons estimated within network buffer from 2000 Census divided by area of residential land use within buffer (sq km) |
| Density | Housing unit density | Number of housing units estimated within network buffer from 2000 Census divided by area of residential land use within buffer (sq km) |
| Greenness | Vegetation index | Average Normalized Difference Vegetation Index (NDVI) within buffer using Landsat satellite image from 9/27/2000 |
Statistical Analysis
Descriptive statistics (means and standard deviations) were used to summarize total moderate-vigorous physical activity and activity accumulated within buffers (i.e., location-based physical activity). The percentage of total time within each buffer classified as moderate and vigorous was also calculated. Five separate multiple linear regression models (one for each built environment variable) were used to estimate associations between the built environment variables within the home buffer and mean minutes per day of moderate-vigorous physical activity, controlling for potential confounding by age, gender and race. Associations with total and location-based moderate-vigorous physical activity were examined. Both outcomes had non-normal distributions; therefore, square root transformations were used. Location-based physical activity around work was also non-normally distributed. Due to the large number of observations with ‘0’ values, square root and log transformations did not provide adequate correction. Therefore, associations were estimated using a generalized linear model (Poisson regression). All data analyses were completed during 2009.
Results
Participant and Monitoring Characteristics
Participants’ average age was 44.0 ± 13.0 years with no significant differences by gender. Twenty-seven percent of the sample was non-white and about twice as many women as men were African-American or black (Table 2). Overall, participants were well educated. Women tended to live in areas with higher intersection density, land use mix, and density, as compared to men (Table 2). Alternatively, the vegetation index was slightly higher for men than women. None of these gender differences were statistically significant.
Table 2.
Demographic and built environment characteristics for study participants (n = 148a)
| All % (n) |
Men % (n) |
Women % (n) |
|
|---|---|---|---|
| Race | |||
| White | 73.5 (108) | 78.6 (55) | 68.8 (53) |
| African-American/Black | 19.7 (29) | 14.3 (10) | 24.7 (19) |
| Other | 6.8 (10) | 7.4 (5) | 6.5 (5) |
| Hispanic or Latino | |||
| Yes | 1.4 (2) | 0.0 (0) | 2.6 (2) |
| No | 98.6 (144) | 100.0 (70) | 97.4 (74) |
| Educational level | |||
| Some college or less | 19.1 (28) | 21.4 (15) | 16.9 (13) |
| Undergraduate degree | 38.1 (56) | 34.3 (24) | 41.6 (32) |
| Some graduate school or above | 42.9 (63) | 44.3 (31) | 42.6 (32) |
| Built environment |
Mean (SD), Median |
Mean (SD), Median |
Mean (SD), Median |
| Intersection density | 5.25 (1.50), 5.48 | 5.08 (1.52), 5.34 | 5.40 (1.47), 5.55 |
| Land use mix | 0.52 (0.23), 0.54 | 0.49 (0.22), 0.52 | 0.55 (0.24), 0.59 |
| Residential population density | 9205.0 (9134.6), 7800.6 | 7836.8 (7222.6), 5998.3 | 10463.8 (10486.1), 9024.9 |
| Residential housing unit density | 4156.6 (4764.7), 3251.8 | 3452.1 (3784.7), 2437.9 | 4804.7 (5460.8), 3640.0 |
| Vegetation index | 0.25 (0.14), 0.23 | 0.27 (0.14), 0.25 | 0.23 (0.14), 0.21 |
Total sample size; sample sizes for each variable vary slightly due to missing data
Forty-nine percent of 148 participants (n= 72) had 4 valid monitoring days, 22 % (n= 32) had 3 days, 17 % (n= 25) had 2 days, and 13 % (n= 19) had 1 valid day. The average number of monitoring days was 3.1 ± 1.1; average monitor wearing time was 14.4 ± 1.6 hrs/d, and mean ct/min/d was 492.3 ± 198.1. The average number of daily monitoring minutes within the 1 km and 50 m home buffers was 201.8 ± 198.4 and 194.7 ± 199.1 min, respectively. Among a subset of participants with work addresses (n= 80), mean minutes in the 1 km and 50 m work buffers were 134.9 ± 130.9 and 52.9 ± 103.2, respectively. Mean monitoring minutes in the 1 km and 50 m work buffers were lower on weekend days (28.1 ± 84.3 and 9.4 ± 70.1, respectively) and substantially higher on weekdays (268.5 ± 234.0 and 108.0 ± 172.9, respectively).
Moderate and Vigorous Physical Activity by Location
The median distance from home to locations where minutes of moderate-vigorous physical activity took place was 1855 m (mean = 7413.5 ± 17438.8 m). Table 3 displays results for moderate and vigorous intensity physical activity at all locations and within home and work buffers. Moderate-vigorous intensity physical activity averaged 61 minutes per day. On average participants accumulated 6 minutes of moderate-vigorous intensity physical activity per day within their 50 m home buffer and 14 minutes within the 1 km home buffer. The corresponding proportion of daily monitoring time within the 1 km home and work buffers of moderate to vigorous intensity was 11.6% and 8.2%, respectively. There were no significant differences between men and women in either mean minutes of activity or the proportion of time that was moderate or vigorous intensity in both home and work buffers.
Table 3.
Moderate and vigorous physical activity at all locations and within home and work buffers
| All locations |
Home buffer 50 m |
Home buffer 1 km |
Work buffer 50 m |
Work Buffer 1 km |
|||||
|---|---|---|---|---|---|---|---|---|---|
| Mean minutes (SD) |
% timea |
Mean minutes (SD) |
% time |
Mean minutes (SD) |
% time |
Mean minutes (SD) |
% time |
Mean minutes (SD) |
|
| Moderate | 53.8 | 9.7 | 5.7 | 10.4 | 12.9 | 7.8 | 1.5 | 7.5 | 7.6 |
| All | (30.5) | (23.2) | (7.9) | (14.6) | (15.5) | (22.9) | (4.0) | (9.5) | (8.8) |
| Men | 52.8 | 8.9 | 6.1 | 10.9 | 12.0 | 7.2 | 1.8 | 8.4 | 7.5 |
| (30.3) | (21.3) | (7.9) | (14.4) | (15.5) | (22.1) | (4.2) | (11.5) | (7.7) | |
| Women | 54.8 | 10.4 | 5.3 | 9.9 | 13.7 | 8.5 | 1.3 | 6.5 | 7.7 |
| (30.8) | (24.9) | (8.0) | (14.8) | (15.5) | (24.6) | (3.8) | (6.6) | (9.9) | |
| Vigorous | 7.0 | 0.1 | 0.3 | 1.2 | 1.1 | 2.8 | 0.1 | 0.7 | 0.9 |
| All | (13.3) | (0.3) | (1.1) | (6.4) | (3.3) | (16.7) | (0.6) | (1.8) | (2.6) |
| Men | 8.3 | 0.1 | 0.4 | 1.9 | 1.5 | 5.0 | 0.2 | 0.9 | 1.3 |
| (15.4) | (0.4) | (1.4) | (9.1) | (4.3) | (22.4) | (0.9) | (1.8) | (3.2) | |
| Women | 5.8 | 0.1 | 0.2 | 0.5 | 0.9 | 0 | 0 | 0.4 | 0.5 |
| (11.1) | (0.2) | (0.7) | (1.3) | (2.1) | (0) | (0) | (1.7) | (1.6) | |
% time represents proportion of time within 50 m or 1 km buffer that is moderate or vigorous intensity.
Associations between Built Environment and Location-based Physical Activity
Associations between the five built environment variables within the 1 km home buffers and total moderate-vigorous physical activity (i.e., physical activity irrespective of location) were non-significant (Table 4). However, higher levels of intersection density, land use mix, residential population density, and residential housing unit density were associated with higher levels of moderate-vigorous physical activity within the 1 km home buffer. The vegetation index was inversely associated with moderate-vigorous physical activity within this buffer. The amount of variance in location-based moderate-vigorous physical activity explained by the five separate adjusted regression models ranged between 11.4% and 13.6%. The variance explained by the built environment variables alone (i.e., unadjusted models) ranged between 4.7% and 8.2%. Interactions between gender and race/ethnicity and each built environment variable were examined (models not shown). Significant interactions were found between gender and intersection density and between race/ethnicity and residential population density. In stratified models, intersection density was positively related to moderate-vigorous activity within women’s home buffers, but not for men. Additionally, residential population density was positively associated with location-based activity for whites, but not for non-whites.
Table 4.
Associations between built environment variables and location-based moderate-vigorous physical activity within 1 km home buffersa,b
| Built environment variables |
Total moderate- vigorous PA, all locations |
Location-based moderate-vigorous PA within 1 km home buffer |
|
|---|---|---|---|
| Intersection density (connectivity) | Estimate | 0.01788 | 0.43372 |
| SE | 0.03969 | 0.11866 | |
| Pr > | t | | 0.6531 | 0.0004 | |
| Adj. R2 | 0.0126 | 0.1136 | |
| Land use mix | Estimate | 0.03436 | 3.01284 |
| SE | 0.24916 | 0.76308 | |
| Pr > | t | | 0.8905 | 0.0001 | |
| Adj. R2 | −0.0041 | 0.1293 | |
| Residential population density | Estimate | 0.00000139 | 0.00006987 |
| SE | 0.00000623 | 0.00001925 | |
| Pr > | t | | 0.8241 | 0.0004 | |
| Adj. R2 | −0.0039 | 0.1152 | |
| Residential housing unit density | Estimate | 0.00000606 | 0.00014096 |
| SE | 0.00001168 | 0.00003583 | |
| Pr > | t | | 0.6046 | 0.0001 | |
| Adj. R2 | −0.0023 | 0.1287 | |
| Vegetation index | Estimate | −0.24911 | −5.37040 |
| SE | 0.43639 | 1.28832 | |
| Pr > | t | | 0.5690 | <0.0001 | |
| Adj. R2 | 0.0135 | 0.1363 |
All models adjusted for age, gender, race/ethnicity, and education.
Sample size for each model ranged from 142 to 146. Adj. = adjusted
Unadjusted models predicting activity within work buffers indicated intersection density was inversely associated, while residential population and housing unit density were positively associated with moderate-vigorous physical activity (Table 5). In adjusted models, only residential population density and housing unit density remained statistically significant.
Table 5.
Associations between built environment variables and location-based moderate-vigorous physical activity within 1 km work buffersa
| Built environment variables | Unadjusted models |
Adjusted modelsb |
|
|---|---|---|---|
| Intersection density | Estimate | −0.0205 | −0.0140 |
| SE | 0.0215 | 0.0221 | |
| Pr > | t | | <0.0001 | 0.5270 | |
| Land use mix | Estimate | 0.00004032 | 0.00003167 |
| SE | 0.00004609 | 0.00004659 | |
| Pr > | t | | 0.3817 | 0.4966 | |
| Residential population density | Estimate | 0.00000236 | 0.00000395 |
| SE | 0.00000106 | 0.00000112 | |
| Pr > | t | | 0.0264 | 0.0004 | |
| Residential housing unit density | Estimate | 0.00000488 | 0.00000783 |
| SE | 0.0000212 | 0.00000227 | |
| Pr > | t | | 0.0215 | 0.0006 | |
| Vegetation index | Estimate | 0.0185 | −0.2969 |
| SE | 0.2533 | 0.2637 | |
| Pr > | t | | 0.9417 | 0.2601 |
Sample size for each model ranged from 75 to 77.
Adjusted for age, gender, race/ethnicity, and education.
Discussion
Using a novel method of combining accelerometer and GPS data, the authors found associations between built environment variables within 1 km home buffers and location-based moderate-vigorous physical activity, but no associations between these variables and total moderate-vigorous physical activity. In adjusted analyses, only residential population density and housing unit density within a 1 km work buffer were related to moderate-vigorous physical activity occurring within that area.
This is one of the first studies to simultaneously monitor free-living adults with accelerometers and GPS devices, objectively quantify what is referred to as “location-based physical activity,” and examine relationships between objective measures of the built environment and location-based activity outcomes. Also, though recent studies have examined relationships between the built environment and walking to work,28 there do not appear to be any previous attempts to characterize the built environment within a buffer around work and examine effects on physical activity within that buffer.
Previous studies examining associations between GIS-derived built environment measures and physical activity measured with accelerometers provide the most appropriate basis of comparison for the present study. Only one other published study has simultaneously monitored free-living adults with accelerometers and GPS devices and examined these data in relation to the built environment.7 Although investigators were limited by a small sample size (~30), they also found no associations between objective built environment variables and total moderate-vigorous physical activity. Also, consistent with the current study, Rodriguez and colleagues found that compared to adults who performed a majority of their physical activity outside their neighborhood, adults performing most of their activity within their neighborhood lived in areas with higher density, street connectivity, and land use mix.7 In a study of adults from metropolitan Atlanta, investigators found that an objective measure of walkability, comprised of land use mix, net residential density, and intersection density, was positively associated with moderate physical activity assessed with accelerometers.8 This is consistent with findings in this study for location-based physical activity. However, contrary to the Atlanta study, there was no relationship between built environment variables and moderate-vigorous physical activity accumulated at all locations. Also, each built environment variable examined in the current study appears to have explained more variance in location-based physical activity than the previous study found for total moderate activity. Indirectly this supports the proposition that researchers may see stronger effects of built environment variables when physical activity outcomes are spatially linked to specific physical contexts.
The results from this study generally support the notion that higher levels of land use mix, street connectivity, and density are related to higher levels of physical activity. Alternatively, the finding that landscape greenness within 1 km home buffers was inversely associated with physical activity appears inconsistent with recent studies in adults and youth,24,25,29,30 which have found positive effects of greenness on outcomes such as trail traffic and body mass index. The presumed link is both perceptual and pragmatic; green landscapes may be visually appealing and greenness can be indicative of spaces conducive to outdoor physical activity (e.g., parks, greenway trails, and sports fields). In the present study the vegetation index within home buffers showed strong negative correlations with the other four built environment variables, ranging from −0.70 to −0.83. Therefore, in buffers with higher levels of street connectivity, land use mix, and density, greenness was lower. The relative influence of vegetation index on activity may vary across different contexts and may take a subordinate role to connectivity and density in more developed areas. Further research is needed to determine the interactive effects of greenness and other built environment variables.
In addition to providing a more explicit link between physical activity outcomes and built environment exposures, another potential contribution of this study is the quantification of moderate-vigorous physical activity in different physical contexts that included home and work buffers. For example, it was found that on average, more moderate-vigorous physical activity occurred outside of home and work buffers than occurred within these areas (66.7 ± 25.2% and 81.9 ± 24.0%, respectively). This finding suggests the need for more careful consideration of spatial context in future studies of built environment and physical activity, including the need for more dynamic models of spatial context beyond home and work environments.
Several limitations should be addressed in future research. The GPS data processing decisions were challenging and need further testing and refinement. For example, it is not clear whether missing GPS data represented someone not wearing the device (i.e., non-compliant) or someone inside a building where signals were lost. Missing GPS data were prospectively imputed using the last available coordinates; the validity of this approach needs further assessment. An integrated GPS-accelerometer device could help to identify the causes for missing GPS data. For example, missing GPS data combined with the presence of accelerometer activity counts could indicate physical activity occuring indoors or in a location with poor reception. A clear determination of time indoors could not be made in this study. Therefore, 50 m buffers were created as a proxy measure for time indoors. Improved GPS instrumentation, geographic data layers, and processing procedures may allow for a more definitive determination of time indoors. The possibility of a Hawthorne effect from wearing the GPS unit cannot be ruled out, since the units weighed about a pound and were stored in a small backpack. However, no participants indicated that wearing the devices affected their activity patterns. A key limitation of this study was the fact that the sample was selected from highly educated, regular trail users in one state, who had physical activity levels that were higher than the general U.S. adult population.18 Therefore, the results are likely not generalizable to the population at-large. Finally, the cross-sectional design precludes making causal inferences about effects of the built environment on location-based physical activity.
Simultaneous accelerometer and GPS monitoring has the potential to yield new insights into the dynamic nature of the relationships between the built environment and physical activity behaviors. This monitoring approach allows activity to be objectively quantified in different physical contexts and thus provides a more explicit link between physical activity outcomes and built environment exposures. Eventually, evidence obtained from studies employing combined accelerometer/GPS assessments may be used to inform the development of more effective physical activity interventions that utilize environmental and policy-level strategies and target specific behavior settings.32
Acknowledgments
This study was supported with funding from the Active Living Research Program, The Robert Wood Johnson Foundation, and additional funding from the College of Liberal Arts at Purdue University. We thank Marcelo Oliveira and Jean Wolf from GeoStats, Inc., Atlanta, GA, for their assistance with GPS data collection and processing. In addition, we wish to thank Dr. Bruce Craig (Department of Statistics, Purdue University) for his recommendations on statistical modeling approaches, Andreana Robertson for SAS programming, and Kosuke Tamura for manuscript preparation.
References
- 1.King AC. Community and public health approaches to the promotion of physical activity. Med Sci Sports Exerc. 1994;26(11):1405–1412. [PubMed] [Google Scholar]
- 2.King AC, Jeffery RW, Fridinger F, et al. Environmental and policy approaches to cardiovascular disease prevention through physical activity: issues and opportunities. Health Educ Q. 1995;22(4):499–511. doi: 10.1177/109019819502200407. [DOI] [PubMed] [Google Scholar]
- 3.Schmid TL, Pratt M, Howze E. Policy as intervention: environmental and policy approaches to the prevention of cardiovascular disease. Am J Public Health. 1995;85(9):1207–1211. doi: 10.2105/ajph.85.9.1207. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Ewing R, Cervero R. Travel and the built environment: a synthesis. Transportation Research Record. 2001;1780:87–113. [Google Scholar]
- 5.Humpel N, Owen N, Leslie E. Environmental factors associated with adults' participation in physical activity: a review. Am J Prev Med. 2002;22(3):188–199. doi: 10.1016/s0749-3797(01)00426-3. [DOI] [PubMed] [Google Scholar]
- 6.Saelens BE, Sallis JF, Frank LD. Environmental correlates of walking and cycling: findings from the transportation, urban design, and planning literatures. Ann Behav Med. 2003;25(2):80–91. doi: 10.1207/S15324796ABM2502_03. [DOI] [PubMed] [Google Scholar]
- 7.Rodriguez DA, Brown AL, Troped PJ. Portable global positioning units to complement accelerometry-based physical activity monitors. Med Sci Sports Exerc. 2005;37(11 Suppl):S572–S581. doi: 10.1249/01.mss.0000185297.72328.ce. [DOI] [PubMed] [Google Scholar]
- 8.Frank LD, Schmid TL, Sallis JF, Chapman J, Saelens BE. Linking objectively measured physical activity with objectively measured urban form: findings from SMARTRAQ. Am J Prev Med. 2005;28(2 Suppl 2):117–125. doi: 10.1016/j.amepre.2004.11.001. [DOI] [PubMed] [Google Scholar]
- 9.Oakes JM, Forsyth A, Schmitz KH. The effects of neighborhood density and street connectivity on walking behavior: the Twin Cities walking study. Epidemiol Perspect Innov. 2007;4:16. doi: 10.1186/1742-5573-4-16. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Troped PJ, Cromley EK, Fragala MS, et al. Development and Reliability and Validity Testing of an Audit Tool for Trail/Path Characteristics: The Path Environment Audit Tool (PEAT) J Phys Act Health. 2006;3(1 Suppl):S158–S175. doi: 10.1123/jpah.3.s1.s158. [DOI] [PubMed] [Google Scholar]
- 11.Troiano RP. Translating accelerometer counts into energy expenditure: advancing the quest. J Appl Physiol. 2006;100(4):1107–1108. doi: 10.1152/japplphysiol.01577.2005. [DOI] [PubMed] [Google Scholar]
- 12.Freedson PS, Melanson E, Sirard J. Calibration of the Computer Science and Applications, Inc. accelerometer. Med Sci Sports Exerc. 1998;30(5):777–781. doi: 10.1097/00005768-199805000-00021. [DOI] [PubMed] [Google Scholar]
- 13.Melanson EL, Jr., Freedson PS. Validity of the Computer Science and Applications, Inc. (CSA) activity monitor. Med Sci Sports Exerc. 1995;27(6):934–940. [PubMed] [Google Scholar]
- 14.Nichols JF, Morgan CG, Chabot LE, Sallis JF, Calfas KJ. Assessment of physical activity with the Computer Science and Applications, Inc., accelerometer: laboratory versus field validation. Res Q Exerc Sport. 2000;71:36–43. doi: 10.1080/02701367.2000.10608878. [DOI] [PubMed] [Google Scholar]
- 15.Craig CL, Marshall AL, Sjostrom M, Bauman AE, Booth ML, Ainsworth BE, et al. International physical activity questionnaire: 12-country reliability and validity. Med Sci Sports Exerc. 2003;35(8):1381–1395. doi: 10.1249/01.MSS.0000078924.61453.FB. [DOI] [PubMed] [Google Scholar]
- 16.Troped PJ, Oliveira MS, Matthews CE, Cromley EK, Melly SJ, Craig BA. Prediction of activity mode with global positioning system and accelerometer data. Med Sci Sports Exerc. 2008;40(5):972–978. doi: 10.1249/MSS.0b013e318164c407. [DOI] [PubMed] [Google Scholar]
- 17.Matthews CE, Chen KY, Freedson PS, Buchowski MS, Beech BM, Pate RR, et al. Amount of time spent in sedentary behaviors in the United States, 2003–2004. Am J Epidemiol. 2008;167(7):875–881. doi: 10.1093/aje/kwm390. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.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(1):181–188. doi: 10.1249/mss.0b013e31815a51b3. [DOI] [PubMed] [Google Scholar]
- 19.Frank LD, Andresen MA, Schmid TL. Obesity relationships with community design, physical activity, and time spent in cars. Am J Prev Med. 2004;27(2):87–96. doi: 10.1016/j.amepre.2004.04.011. [DOI] [PubMed] [Google Scholar]
- 20.Oliver LN, Schuurman N, Hall AW. Comparing circular and network buffers to examine the influence of land use on walking for leisure and errands. International Journal of Health Geographics. 2007;6(41):1–11. doi: 10.1186/1476-072X-6-41. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Frank L, Pivo G. Impacts of mixed use and density on utilization of three modes of travel: single-occupant vehicle, transit, and walking. Transportation Research Record. 1995;1466:44–52. [Google Scholar]
- 22.Handy S. Critical assessment of the literature on the relationships among transportation, land use, and physical activity: Transportation Research Board and the Institute of Medicine Committee on Physical Activity, Health, Transportation, and Land Use. 2004 [Google Scholar]
- 23.Lindsey G, Wilson J, Yang JA, Alexa C. Urban Greenways, Trail Characteristics and Trail Use: Implications for Design. Journal of Urban Design. 2008;13:53–79. [Google Scholar]
- 24.Liu GC, Wilson JS, Qi R, Ying J. Green neighborhoods, food retail and childhood overweight: differences by population density. Am J Health Promot. 2007;21(4 Suppl):317–325. doi: 10.4278/0890-1171-21.4s.317. [DOI] [PubMed] [Google Scholar]
- 25.Wilson JS, Lindsey GH, Liu GC. Viewshed characteristics of urban pedestrian trails: Indianapolis, Indiana, USA. Journal of Maps. 2008:108–118. [Google Scholar]
- 26.Forsyth A. Environment and Physical Activity: GIS Protocols, Version 4.1. Minneapolis, MN: University of Minnesota; 2007. [Google Scholar]
- 27.Jensen JR. Introductory Digital Image Processing: A Remote Sensing Perspective. 3rd ed. New York: Prentice-Hall; 2005. [Google Scholar]
- 28.Craig CL, Brownson RC, Cragg SE, Dunn AL. Exploring the effect of the environment on physical activity: a study examining walking to work. Am J Prev Med. 2002;23(2 Suppl):36–43. doi: 10.1016/s0749-3797(02)00472-5. [DOI] [PubMed] [Google Scholar]
- 29.Lindsey GH, Wilson JS, Yang JA, Alexa C. Urban form, trail characteristics, and trail use: Implications for urban design. Journal of Urban Design. 2008;13(1):53–79. [Google Scholar]
- 30.Liu GC, Colbert JT, Wilson JS, Yamada I, Hoch SC. Examining urban environment correlates of childhood physical activity and walkability perception with GIS and remote sensing. In: Jensen RR, Gatrell JD, McLean DD, editors. Geo-Spatial Technologies in Urban Environments: Policy, Practice, and Pixels. 2nd ed. Berlin: Springer-Verlag; 2007. pp. 121–140. [Google Scholar]
- 31.Owen N, Leslie E, Salmon J, Fotheringham MJ. Environmental determinants of physical activity and sedentary behavior. Exerc Sport Sci Rev. 2000;28(4):153–158. [PubMed] [Google Scholar]