Portable Global Positioning Units to
Complement Accelerometry-Based Physical
Activity Monitors
DANIEL A. RODRÍGUEZ1, AUSTIN L. BROWN1, and PHILIP J. TROPED2
Department of City and Regional Planning, University of North Carolina, Chapel Hill, NC; and 2Department of Society,
Human Development and Health, Harvard School of Public Health, Boston, MA
1
ABSTRACT
RODRÍGUEZ, D. A., A. L. BROWN, and P. J. TROPED. Portable Global Positioning Units to Complement Accelerometry-Based
Physical Activity Monitors. Med. Sci. Sports Exerc., Vol. 37, No. 11(Suppl), pp. S572–S581, 2005. Purpose: This study examines the
usefulness of complementing accelerometry-based physical activity measurement with spatial data from portable global positioning
system (GPS) units to determine where physical activity occurs. Methods: First, using the geographic distribution of data points and
Bland–Altman plots, we examined GPS units’ validity and interunit reliability by measuring the distance to a geodetic point. We also
assessed interunit reliability by comparing GPS data collected in three built environment contexts. Second, we conducted a pilot study
in which 35 participants wore GPS units and accelerometers in free-living conditions for 3 d. Moderate and vigorous physical activity
(MVPA) bouts were matched to GPS data. We classified each bout as occurring inside or outside the participant’s home neighborhood.
Using unpaired t-tests and Fisher’s exact tests, we compared neighborhood attributes for participants having the majority of MVPA
bouts within their home neighborhood, relative to those with most bouts away from their home neighborhood. Results: Average
distance from each unit to the geodetic point was 3.02 m (SD 2.51). Average bias among units using Bland–Altman plots was 0.90 m,
ranging from ⫺0.22 to 1.86 m, within the limits of agreement. For interunit reliability in the built environment contexts, the mean
distance difference among units ranged between 10.7 m (SD 11.9) and 20.1 m (SD 21.8). For the pilot study involving participants,
GPS data were available for 59.3% of all bouts (67% of MVPA time), of which 46% were in the participants’ neighborhood.
Participants obtaining most of their MVPA in their neighborhoods tend to live in areas with higher population density, housing unit
density, street connectivity, and more public parks. Conclusion: Data recorded by portable GPS units is sufficiently precise to track
participants’ movements. Successful matching of activity monitor and locational data suggests GPS is a promising tool for complementing accelerometry-based physical activity measures. Our pilot analysis shows evidence that the relationship between environment
and activity can be clarified by examining where physical activity occurs. Key Words: PHYSICAL ACTIVITY, GLOBAL
POSITIONING SYSTEM, ACCELEROMETER, BUILT ENVIRONMENT, GEOGRAPHICAL INFORMATION SYSTEMS
E
and a lack of adequate infrastructure to support bicycling
and walking, may act as barriers or inhibitors to physical
activity (20). Indeed, built environment interventions may
be one of the most effective strategies for improving physical activity and weight status (5,21). Despite finding moderately positive relationships between environmental factors
and physical activity behavior among adults, recent reviews
also have identified a considerable number of studies showing no statistical association (8,11,15).
Several explanations for discrepancies among the findings of recent research are likely. First, physical activity is
measured and analyzed in different ways. Similarly, objective environmental measures are often disparate, limiting
comparability and the ability to replicate results across studies (11,19). Second, the environments examined tend to be
limited to the vicinity of participants’ place of residence
even though individuals may be physically active at other
locations. A third explanation is that, faced with an intervention that removes physical activity barriers, some individuals may shift the locations where their physical activity
occurs, from elsewhere to the environment with lowered
barriers. The latter may explain why recent studies evaluating trail interventions have shown increased trail use but
no changes in total physical activity (1,12).
merging theoretical models aimed at understanding
the role of neighborhood contextual factors as barriers or supporters for physical activity have shown
promise in explaining individual behavior (14,25). An expanded set of factors hypothesized to influence physical
activity behavior, such as social, social-environmental, and
physical environmental factors, are now being studied with
the aim of identifying relationships and testing potential
interventions.
Through its relevance as a social-environmental factor or
community-level factor, the built environment has emerged
as an area of interest to promote physical activity. Prevailing
development patterns, with separated residential and commercial land uses, increased reliance on automobile travel,
Address for correspondence: Daniel A. Rodrı́guez, M.S., Ph.D., Department of City and Regional Planning, University of North Carolina, CB
3140 New East Building, Chapel Hill, NC 27599-3140; E-mail:
danrod@email.unc.edu.
0195-9131/05/3711(Suppl)-S572/0
MEDICINE & SCIENCE IN SPORTS & EXERCISE®
Copyright © 2005 by the American College of Sports Medicine
DOI: 10.1249/01.mss.0000185297.72328.ce
S572
Critical to validating some of these explanations is an
understanding of where physical activity takes place. The
current study examined the usefulness of complementing
accelerometry-based physical activity measures with spatial
data from a portable global positioning system (GPS) in
order to determine where physical activity occurs. GPS is
now used in a variety of commercial and research applications, such as farming (22), geology (6), environmental
epidemiology (16), transportation planning (27,31), and exercise science (9,24). Recent technological improvements
have resulted in portable GPS units with adequate memory
to store positional data over time, thus offering opportunities
for obtaining locational information at low cost. However,
GPS technologies have limitations: they regularly fail to
record positions indoors and they can fail to record under
heavy tree canopy and in dense urban areas. In the context
of urbanized areas, the application of GPS technology to
understanding activity location requires evaluative pilottesting for reliability and validity to ensure feasibility and
lack of systematic bias. Our study contributes to the evaluation of GPS to complement objectively measured physical
activity first by examining the GPS units’ battery life and
their reliability and validity in different environmental contexts. Then we conducted a pilot study in which accelerometers and GPS units were concurrently worn by adults to
identify and examine physical activity location.
TESTING THE RELIABILITY AND VALIDITY
OF THE GPS UNITS
Methods
Units and data recorded. We tested six off-the-shelf
Foretrex 201 portable (83.8 ⫻ 43.2 ⫻ 15.2 mm) GPS units
(Garmin Ltd., Olathe, KS), each weighing 78 g. The units
are designed to be worn on the wrist, upper arm, or attached
to a belt around the waist. An internal nonvolatile memory
card provides the unit with the capacity to store 10,000
points before the data require downloading. The units were
set to record the positional coordinates of their location at
30-s intervals with the Wide Area Augmentation System
(WAAS) enabled. The WAAS consists of satellites and
ground stations that correct GPS signal errors caused by
atmospheric disturbances, timing, and satellite orbit errors,
resulting in increased positional accuracy.
The map datum used was World Geodetic Survey 1984
and the position format was latitude and longitude in degrees and minutes (HD° MM=). Recorded data, including
latitude, longitude, date, and time of day, were downloaded
to a personal computer using MapSource Trip and Waypoint
Manager (Garmin Ltd.).
After the data for each test were collected, we cleaned
each data file using a computer program developed for this
study (available on request). This procedure removed data
headers, removed data when the unit was unable to record a
position, converted coordinate information into decimal degrees, and transformed the data into wide character ASCII
GPS AND PHYSICAL ACTIVITY
format to enable further processing with geographic information system (GIS) software.
Static tests. We conducted tests of battery life and
recharge time, validity, interunit reliability, and body placement. Battery life was tested by the time it took to fully
recharge the battery of a single unit on seven trials (N ⫽ 7)
and, for the same unit, the time to exhaustion when fully
charged (N ⫽ 14). Data for the validity and interunit reliability were collected by simultaneously placing each unit
for 1 h on a geodetic point (latitude 35°54=40.12898⬙ and
longitude 79°3=2.54092⬙ NAD 83) maintained by the North
Carolina Geodetic Survey. The simultaneous placement of
the units controls experimentally for any weather and timeof-day effects. Each unit recorded 121 time-synchronized
data points (N ⫽ 726). Because the units did not move from
the known geodetic point, these tests could be considered
assessments of “static validity” and “static interunit reliability.” Static validity was determined by visually inspecting a
GIS map of recorded coordinates vis-à-vis the known coordinates of the geodetic point. Static interunit reliability
was assessed using Bland–Altman plots by comparing the
distance between each unit and the geodetic point relative to
the same distance from all other units. This resulted in six
plots, one for each unit.
Free-living tests. We collected positional data from
the six GPS units in each of three different built environment scenarios: open space with some tree canopy, clustered
development (high-density development with open space
surrounding it), and urban. Each scenario contained 25
locations, spaced an average of 15, 42, and 24 m from each
other for the open, clustered, and urban scenario, respectively. Every unit was worn by study staff three times as the
staff moved from location to location in each scenario
(yielding 18 points per location). Upon arrival at each location, study staff manually pressed a button available in the
GPS units, which automatically records the coordinates of
the location. Interunit reliability was assessed by displaying
the cumulative distribution of distances between each unit’s
recorded position at each location and the average position
indicated by the other five units at that location.
Body placement tests examined how placement of the
unit (waist or ankle relative to wrist) influenced the amount
of positional data collected by each unit. Study staff simultaneously wore units on the right wrist, waist, and on the
right ankle while moving from location to location in the
open space and clustered development scenarios described
above (25 per scenario). Unfortunately, manually pressing a
button for three units at once was cumbersome and introduced measurement error. Thus, the protocol was changed
so that when each location was reached, staff recorded the
time and remained at the location for 40 s before proceeding
to the next location. This ensured that each unit recorded the
coordinates of the location at least once while controlling
experimentally for atmospheric and time-of-day conditions
that may influence the units’ accuracy. By comparing the
staff-recorded timestamps to the timestamps collected by
the GPS units, it was possible to determine which points
were recorded at each location. We used two-sample t-tests
Medicine & Science in Sports & Exercise姞
S573
FIGURE 2—Bland–Altman plot of GPS unit 3 and average distance
from remaining units. The horizontal solid line represents the mean
bias. The dashed lines represent the 95% band around the mean bias.
The sloping solid line represents the Pearson correlation with slope r
and significance P.
FIGURE 1—Spatial distribution of observations relative to geodetic
point. The cumulative frequency of points by distance to the geodetic
point is 589 (81.1%) within a 5-m buffer, 710 (97.8%) within a 10-m
buffer, 722 (99.4%) within a 15-m buffer, and 725 (99.9%) within a
20-m buffer.
with unequal variances to compare wearing the GPS unit on
the waist and ankle relative to the wrist.
RESULTS
Battery tests showed that units operated continuously an
average of 15.97 h (SD 1.01) and recharge time averaged
2.66 h (SD 0.1). For evaluating static validity and static
reliability, Figure 1 shows the spatial distribution of 726
observed GPS points relative to the known geodetic point.
Many points overlap because the position recorded was the
same. The average distance recorded from the units to the
geodetic point was 3.02 m (SD 2.51) with 81.1% of observations within 5 m and 99.4% of observations within 15 m
of the known point.
For the six Bland–Altman plots to assess static interunit
reliability, mean bias ranged between ⫺0.22 and 1.86 m,
with an average of 0.90 m (SD 0.74). The slopes of the
scatter of points in the plots (Pearson correlation coefficients) ranged between 0.28 and 0.76, with an average of
0.53 (SD 0.18). A sample Bland–Altman plot of unit three
and the average distance recorded by all other GPS units are
provided in Figure 2. All other plots (not shown) display
similar cases of positive correlations, suggesting that error is
increasing with the magnitude of the mean.
For the 25 locations in the open space scenario, the mean
estimated distance between each unit’s recorded position
and the average position of the other five units was 10.7 m
(SD 11.9). For the clustered development scenario and the
urban scenario the mean values were 20.1 m (SD 21.8) and
18.5 m (SD 18.4), respectively. Figure 3 shows the cumulative distribution of distances by location in the clustered
development scenario. The same figures for the open space
S574
Official Journal of the American College of Sports Medicine
and urban scenarios (not shown) indicate higher accuracy in
those scenarios.
Finally, the body placement tests show that wearing the
GPS unit on the wrist yielded data points 100% of the time
in the open space scenario and 86.8% of the time in the
clustered development scenario. Compared with the wrist,
wearing the GPS unit on the ankle resulted in fewer records.
In the open space scenario, wearing the unit on the ankle
resulted in 1.6% fewer records (P ⬍ 0.05), but wearing the
unit on the belt was no different than wearing it on the wrist
(P ⫽ 0.16). In the clustered development scenario, wearing
the unit on the ankle resulted in 21% fewer records (P ⬍
0.01), and wearing the unit on the belt resulted in 5.5% more
records than wearing it on the wrist (P ⬍ 0.05).
DISCUSSION
In this study, we tested the battery life and recharge time,
static validity, static reliability, interunit reliability in simulated free-living conditions, and body placement of portable GPS units. Our results suggest that limited battery life is
a significant limitation of the portable GPS units we evaluated. Fully charged batteries lasted an average of 15.97 h,
suggesting that GPS data are likely to be missing for the
later part of a day for adult participants. However, this figure
is slightly higher than the typical amount of time in a day
participants wear accelerometers (10,29). Although other
GPS units on the market appear to have higher battery life,
we believed that the ease of portability, memory capacity,
functionality, and compactness of the GPS units used in our
study were more important than prolonged battery life. The
degree to which the limited battery life biases analyses of
the locations in which physical activity takes place is a
matter for further empirical investigation. Limited battery
life also implies that, in multiday studies, participants would
be required to recharge the unit, adding burden to their
participation, and increasing the probability of noncompliance. Another limitation is that we did not compare positional accuracy to other portable GPS units available on the
market.
http://www.acsm-msse.org
FIGURE 3—Distribution of distances between recorded positions and the estimated average position of remaining GPS units by clustered
development scenario location (18 points per location).
The multiple tests of validity and interunit reliability
suggest that the positional information recorded by the units
is sufficiently precise to track participant’s movements,
provided that a position is recorded. Figure 1 suggests that
most observations are scattered in close proximity of the
known geodetic point. The fact that 99.4% of recorded
observations are within 15 m of the geodetic point is noteworthy. The Bland–Altman plots suggest that most observations are well within the limits of agreement. For Figure 2,
only 2.5% of the observations lie beyond a 95% confidence
band (⫺0.40, 1.04) around the mean bias. Because differences within the 95% confidence interval are acceptable for
our current purposes, static interunit reliability is deemed
high. The increase in the difference as the mean increases
suggests that the limits of agreement are wider apart than
necessary for small means and narrower than they should be
for large means. However, our assessment regarding interunit reliability does not change. Kinetic interunit reliability
results show the effect of the built environment on positional
accuracy. Locations next to buildings or under dense tree
canopy showed higher interunit error than locations that
were more open. The error in the open space scenario was
also lower than in the other two scenarios.
Finally, results from testing the body placement of the GPS
units provide two findings of importance. First, the usable data
yield from the units (86.8 and 100% in each scenario tested)
appears satisfactory. It means that in the worst case scenario
GPS AND PHYSICAL ACTIVITY
13.2% of the time the units either recorded data that were not
usable or failed to record any data. Second, wearing the unit on
the ankle is not recommended.
Our examination thus far is limited in several ways. First,
we did not test for any potential interference that may exist
among GPS units even though many of our tests were
conducted wearing several GPS units at the same time to
guarantee time synchronicity. Second, the tests based on
data collected at different times may suffer from random
error related to the location of satellites. Third, we did not
examine whether disabling the WAAS could be an effective
strategy for saving battery power. Further tests may determine the tradeoff between the additional power saved and
the potential decrease in positional accuracy resulting from
disabling the units’ WAAS. In sum, this first part of our
study shows that the positional information collected with
the portable GPS units is sufficiently precise to track the
movements of individuals carrying the units.
USING GPS AND ACCELEROMETRY TO
TRACK PHYSICAL ACTIVITY
Methods
Study purpose and units used. Having established
the acceptable performance of the GPS units, the principal
aim of this second part of the study was to determine the
Medicine & Science in Sports & Exercise姞
S575
degree to which the units can complement accelerometerbased physical activity measurements in order to identify
where physical activity takes place. A secondary aim was to
show how the combination of these two data sources provides useful information for testing hypotheses regarding
the relationship between physical activity and the built environment. The same six GPS units described previously
were used for this study, but they were set to record every
minute. For the activity monitor, we used the dual mode
ActiGraph model 7164, formerly known as the Computer
Science and Applications (CSA) and Manufacturing Technology Inc. (MTI) (ActiGraph, LLC, Fort Walton Beach,
FL). Previous studies have demonstrated the ActiGraph
7164 to be reliable and valid (13,30). Each activity monitor
was calibrated and tested to ensure adequate operation and
was set to record activity in 1-min epochs. The clock in the
activity monitors and the GPS units were synchronized to
ensure the data could be matched correctly.
Subjects. Volunteer participants were recruited using
University of North Carolina–supported e-mail lists and
fliers posted in the town of Chapel Hill. A cash incentive
was provided to all participants upon their completion of the
study. All potential participants underwent a screening procedure to ensure they were at least 18 yr old, resided at a
local nonuniversity address, and had the ability to walk
continuously and unassisted for 20 min. Participants were
recruited as evenly as possible to achieve a balanced gender
distribution. The study sample (N ⫽ 35) comprised 21
(60%) females aged 23– 61 (mean 32.1, SD 10.7) and 14
(40%) males aged 20 –58 (mean 31.5, SD 10.6). Participants
were recruited on a rolling basis during a 4-wk period in
September and October 2004. The study procedures were
approved by the university’s institutional review board and
written consent was provided by all participants.
Data collection. Each participant was asked to wear a
GPS unit and activity monitor for three consecutive days.
Participants were instructed to wear the GPS and activity
monitor during all waking hours except when showering,
bathing, swimming, or during activities that would result in
submerging the units in water. Consistent with the literature,
participants were instructed to wear the ActiGraph unit on
the right hip using either a belt clip or nylon carrying pouch
(7,23,26). They were also instructed to carry the GPS units
on either their wrist or waist and always to charge the GPS
unit overnight. After the 3 d, study staff interviewed each
participant to collect information on their experience carrying the devices. GPS data from one participant were not
recorded because the participant did not adhere to the GPS
protocol. GPS data from two other participants were not
recorded because one unit malfunctioned with a participant
and the unit was mistakenly given to a second participant.
The unit was exchanged by a retailer and no further problems arose. The three participants with no GPS data were
excluded from the analysis. Eight other participants had no
GPS readings for one of their three days. Because the units
for these eight participants were determined to be working
properly, their accelerometer data (and lack of GPS data)
were included in our analysis.
S576
Official Journal of the American College of Sports Medicine
Processing of ActiGraph, GPS, and neighborhood characteristics data. GPS data were downloaded
using the same process described earlier. ActiGraph data
were downloaded using a reader interface unit and Actisoft
Analysis Software 3.2 (Fort Walton Beach, FL). A computer
program developed for this study (available on request) was
used to identify and calculate the number and duration of
moderate physical activity (MPA), vigorous physical activity (VPA), and moderate to vigorous physical activity
(MVPA) bouts per participant. The program also merged
participant’s activity monitor data with the corresponding
GPS data according to the date and time information in each
unit.
To identify intensities of physical activity and to classify
bouts as MPA or VPA, we followed recommendations in
Swartz et al. (26). Consistent with Catellier et al. (2), we
defined a bout as 10 or more continuous and consecutive
minutes of physical activity at or exceeding the defined
threshold values. Each bout had a 30% tolerance for activity
level not meeting the threshold values. For example, in a
10-min bout, 3 min could be under the threshold level. A
participant was presumed to be wearing the activity monitor
from the time when the first nonzero value was recorded and
was followed by one or more nonzero values being recorded. A participant was considered as not wearing the
monitor if 20 or more consecutive minutes had zero counts.
Missing data were not imputed. No outliers (counts exceeding 15,000 for 5 min or more) in the data were identified.
All bout information with matching positional data was
imported to ArcMap 8.3 (ESRI, Redlands, CA) for further
processing. Participants’ GPS data were overlaid with GIS
layers containing land use (residential, commercial, office,
park, or other), the road network, building footprint, and
digital orthophotographic images (1-m resolution; 1:1200
scale) provided by municipal and/or county planning offices
in the study area. A building footprint is the physical outline
of the building on the property where it stands. The address
of each participant’s primary residence was geocoded using
ESRI Streetmap in ArcCatalog (ESRI). Thirty-one addresses were successfully geocoded using default settings
and one address was manually geocoded. Addresses of the
three participants without GPS data were not geocoded.
The GIS software was used to classify bouts with GPS
data in terms of the locations where the bouts occurred
(indoors, outdoors in neighborhood, outdoors out of neighborhood). A bout was identified as occurring indoors if 33%
or more of its GPS data were located within a building
footprint. If not indoors, a bout was then classified as
outdoors in neighborhood if 33% or more of the positional
data for the bout fell within a 1.54 km (1 mi) circular buffer
around each participant’s geocoded address. Other studies
have used similar neighborhood definitions (3). All other
bouts with GPS data were classified as occurring outdoors
out of neighborhood. This classification allowed for comparisons of physical activity (bout, average bout duration,
and percent of total MVPA time) by where it occurred.
When GPS data were available for part of the bout only, we
relied on that data to identify the location where the bout
http://www.acsm-msse.org
occurred. Because this approach may introduce some misclassification for bouts that were partly outdoors but mostly
indoors, we also compared bouts with any GPS data available for 1 min or more to a stricter classification criterion
that relied on bouts having 30% or more of GPS data
available for the duration of the bout.
To perform a preliminary exploration of the relationship between neighborhood characteristics and physical
activity, we calculated measures of population density,
housing density, street density, street connectivity (threeand four-way intersection density), and accessibility to
commercial or office land uses and parks within each
participant’s neighborhood. Population and housing information were drawn from the 2000 population census’
block-level data. Data for blocks not lying completely
within the neighborhood boundary were adjusted proportionally based on the block area inside the boundary.
Street connectivity was calculated using the ArcScript
Point and Polyline Tools v.1.2 extension in ArcView 3.2
(ESRI). The presence of mixed land uses within the
neighborhood was coded using a binary variable equaling
1 if the neighborhood had at least one parcel (entirely or
partially) with a retail, institutional, or office land use
code designation, and 0 otherwise. The presence of parks
within the neighborhood was coded using a binary variable in a similar fashion to mixed land uses.
Data analysis. Comparisons of average duration of
indoors, outdoors in neighborhood, and outdoors out of
neighborhood bouts were made using unpaired t-tests assuming unequal variances. Comparisons of percentages of
bouts occurring at each location were made with ␹2 tests.
Statistical significance tests are two tailed.
Pearson’s correlation coefficients were used to examine
the association between each participant’s neighborhood
characteristics and MVPA time. We also described neighborhood characteristics stratified by where the majority of
participants’ MVPA time (with GPS data) occurred, indoors, outdoors in neighborhood or outdoors out of neighborhood. To determine whether neighborhood characteristics differ significantly between participants among the
three strata, we used an unpaired t-test assuming unequal
variances for continuous variables and Fisher’s exact test for
variables measuring percentages.
RESULTS
Table 1 summarizes participants’ physical activity.
Participants wore the activity monitors an average of 13.7
⫾ 1.8 h·d⫺1 and had an average of 142.01 min of MVPA
per day, of which 92.79 min were in a bout. The exit
interviews confirmed that eight participants reported noTABLE 1. Daily average of participants’ MVPA (N ⫽ 96).
No. of moderate or vigorous bouts
Total MVPA bout durationa (min)
Total MVPA time (min)
a
Mean
SD
Min
Max
3.97
92.79
142.01
2.29
67.45
53.94
0
0
35
10
380
296
Includes a 30% tolerance for activity level not meeting threshold values.
GPS AND PHYSICAL ACTIVITY
ticing that the GPS units ran out of battery life before the
end of their day at least once. Figure 4 displays three
bouts of MVPA for one participant. The map shows the
route taken for each bout and select attributes of the
environment in which the physical activity occurred. Table 2 summarizes the results of matching GPS positional
data with physical activity accelerometry, stratified by
the criterion used in determining GPS data availability for
a bout (1 min or 30% of bout). GPS data were available
for 59.3% of all bouts (67% of MVPA time) using the
1-min criterion and for 44.9% of all bouts (53% of MVPA
time) using the 30% of bout criterion. Comparisons between bouts with and without GPS data suggest that those
with GPS data tended to be longer than those without
GPS data (P ⱕ 0.00). Of the bouts with GPS data, there
were fewer in-neighborhood bouts than out-of-neighborhood bouts. However, the in-neighborhood bouts were
longer (P ⬍ 0.05) and contributed more to total MVPA
time than did out-of-neighborhood bouts. These differences hold regardless of the GPS data inclusion criterion
applied. No indoor bouts with GPS data were identified.
The consequences of applying the stricter GPS data inclusion criterion are shown also in Table 2. The average
duration of out-of-neighborhood bouts retained from applying the stricter criterion is not different than the average
duration of out-of-neighborhood bouts rejected due to insufficient GPS data (P ⫽ 0.914). However, applying the
stricter criterion leads to rejecting in-neighborhood bouts
that tend to be shorter than the bouts that are retained (P ⬍
0.1).
Table 3 summarizes the characteristics of the neighborhood environment where participants reside and shows correlations with total MVPA. These six attributes of the neighborhood built environment and MVPA had no statistically
significant correlations. Using the positional information
provided by the GPS units, we classified participants into
two groups according to where the majority (⬎50%) of their
physical activity occurred (in-neighborhood or out-ofneighborhood). Table 4 presents summary neighborhood
characteristics stratified by each group. Participants who got
most of their moderate and vigorous physical activity
(MVPA) in their neighborhood tended to live in environments with higher population density (P ⱕ 0.00), higher
housing unit density (P ⱕ 0.00), higher street connectivity
(P ⱕ 0.00), and higher street density (P ⬍ 0.05). When the
stricter GPS data inclusion criterion was applied, the statistical significance of these associations increased and the
variables measuring the presence of parks and mixed land
uses became statistically significant (P ⬍ 0.05).
DISCUSSION AND CONCLUSIONS
In this part of the study, we piloted the concurrent deployment of portable GPS units and accelerometers to determine where physical activity takes place. Using time and
date information, we determined that more than half of all
MVPA bouts and more than two thirds of all MVPA time
could be matched successfully with positional data from the
Medicine & Science in Sports & Exercise姞
S577
FIGURE 4 —Sample physical activity
bouts for a participant and their location
relative to streets and land use.
portable GPS units. Our results also show that the bouts with
GPS data are longer than the bouts without GPS data. This
is likely due to the fact that indoor physical activity is not
likely to yield GPS data and activity outdoors, where GPS
data are more likely to be recorded, is generally longer than
activity indoors. Further research should confirm these two
potential explanations. We also identified limitations from
the GPS units used. The data recorded did not allow for
differentiating when the GPS unit was turned off (or fully
discharged) from when the GPS unit was turned on but
unsuccessfully determining and recording its position (e.g.,
indoors or under heavy tree canopy). In either case, the GPS
TABLE 2. Descriptive statistics of the location of MVPA bouts stratified by the availability of GPS data in bouts.
Criterion 1: GPS Data Present for 1 Min of Bout
Bouts with GPS data meeting criterion
Indoors
Outdoors in neighborhood
Outdoors out of neighborhood
Bouts with GPS data not meeting criterion
Outdoors in neighborhood
Outdoors out of neighborhood
Bouts with no GPS data
Total
S578
No. of
Bouts
% of All
Bouts
Time per
Bout (min)
% of Total
MVPA
Time
226
0
104
122
—
—
—
155
381
59.3
0
27.3
32.0
26.4
0.00
30.3
23.2
—
—
—
18.9
23.4
67.0
0.0
35.0
32.0
—
—
—
33.0
100.0
40.7
100.0
Official Journal of the American College of Sports Medicine
Criterion 2: GPS Data Present for 30% of Bout
No. of
Bouts
% of All
Bouts
Time per
Bout (min)
% of Total
MVPA
Time
171
0
81
90
55
23
32
155
381
44.9
0.0
21.3
23.6
14.4
6.0
8.4
40.7
100.0
27.4
0.0
32.1
23.3
23.3
24.0
22.8
18.9
23.4
53.0
0.00
29.0
24.0
14.0
6.0
8.0
33.0
100.0
http://www.acsm-msse.org
TABLE 3. Characteristics of participants’ neighborhoods and their unadjusted correlation with MVPA minutes in a bout (N ⫽ 32).
Population density (persons per acre)
Housing density (units per acre)
Intersection density (three- and four-way intersections
per square mile)
Street density (centerline miles per square mile)
Neighborhood contains office or commercial land usesa
Neighborhood contains at least one parka
a
Mean
or %
Median
SD
Min
Max
Pearson
Correlation
P Value
5.28
2.31
110.21
4.57
1.94
108.70
3.77
1.58
65.87
0.20
0.08
2.55
11.53
4.87
204.04
0.03
0.00
⫺0.03
0.87
0.99
0.89
10.41
88%
78%
10.83
—
—
4.37
—
—
1.43
—
—
15.70
—
—
⫺0.10
0.08
0.07
0.58
0.68
0.70
Point– biserial correlation reported.
units record no information to the memory card. Although
this preserved memory space, it limits the ability to determine whether participants are adhering to the study protocol, and to identify the location of activity (indoors vs
outdoors).
Following traditional nonspatial correlational analyses of
physical activity and built environment characteristics, we
found no association between our neighborhood measures
and total MVPA (unadjusted analyses, Table 3). The limited
sample size did not allow us to adjust for other covariates.
By contrast, a second approach to examining the data highlights the usefulness of accounting for physical activity
locational information by associating neighborhood envirronment characteristics with where physical activity occurs
(Table 4). Despite the limited sample size, our preliminary evidence shows that in-neighborhood physical activity tended to
occur in environments with characteristics often hypothesized to
support physical activity (19,20). Similar results were obtained for
participants having any physical activity in their neighborhood (as
opposed to the majority) vis-à-vis those not being active in their
neighborhood (results not shown). The findings are consistent
with research suggesting that increased accessibility to opportunities for activity (8), denser areas (18), and areas with higher
connectivity (18) are related to increased physical activity. They
also echo recommendations to examine physical activity in specific settings in order to clarify its association with the built
environment (11). Further analyses can examine the characteristics of locations away from participants’ neighborhoods but in
which participants were physically active.
Our pilot study had several methodological limitations,
including the confined geographic scope and the limited
sample size. Another limitation was the potential for
misclassifying bouts as indoors, outdoors in neighborhood, and outdoors out of neighborhood. Because the
building footprint data have a horizontal accuracy of
0.518 m, misclassification of the location of activity due
to footprint error is unlikely. Thus, misclassification errors are likely to result mostly from measurement errors
with the GPS units. Bouts were classified using the only
and best locational information available to us collected
by the GPS units. However, when bouts occurred mostly
indoors but also outdoors, they would be incorrectly
classified by our method. Our comparison of criteria to
classify bouts (1 min and 30% of bout) was intended to
address this limitation, although the criteria were selected
arbitrarily. Despite the similarity between characteristics
of in-neighborhood physical activity using both criteria to
classify bouts and to minimize the potential for misclassification, we recommend that researchers use the more
stringent criterion of 30% of bout (or higher). Further
research can examine the sensitivity of the results to other
criteria.
A final shortcoming is the lack of information about
the type of physical activities (recreational or for travel)
undertaken by participants. Because the built environment is expected to influence the various types of physical activity differently (17,19,28), supplementing positional data and accelerometer data with information on
the type of physical activity could provide additional
valuable information to better understand the role of
environmental factors in explaining physical activity behavior. The application of GIS to measuring the built
environment and understanding its role in physical activity behavior deserves additional research. Although our
measures of the built environment parallel similar measures tested elsewhere (4), the high resolution of the
positional data enable the calculation of finer grain built
environment measures than before. Last, the use of GIS to
TABLE 4. Characteristics (mean (SD) or percentage) of participants’ neighborhood of residence stratified by where the majority of physical activity takes place.
Criterion 1: GPS Data Present for 1 Min of Bout
Population density (persons per acre)
Housing density (units per acre)
Intersection density (three- and four-way intersections
per square mile)
Street density (centerline miles per square mile)
Neighborhood contains office or commercial land uses
Neighborhood contains at least one park
Majority of
MVPA Time in
Neighborhood
(N ⴝ 15)
Majority of
MVPA Time out of
Neighborhood
(Nⴝ 16)
7.50 (3.91)
3.19 (1.59)
145.05 (63.57)
3.02 (2.04)
1.38 (0.97)
73.07 (46.55)
12.31 (4.13)
100% —
93% —
8.38 (3.80)
75% —
63% —
Criterion 2: GPS Data Present for 30% of Bout
P
Majority of
MVPA Time in
Neighborhood
(N ⴝ 16)
Majority of
MVPA Time out of
Neighborhood
(N ⴝ 14)
P
0.00
0.00
0.00
6.92 (4.32)
3.04 (1.8)
151.20 (71.65)
2.77 (2.04)
1.27 (0.97)
66.19 (45.46)
0.00
0.00
0.00
0.02
0.10
0.08
27.24 (4.7)
100% —
94% —
7.52 (3.45)
71% —
57% —
0.00
0.04
0.03
One participant was excluded from both strata summaries because the location of the physical activity bouts could not be determined due to lack of GPS data. A second participant
was excluded from the criterion 2 strata because no bout had GPS data for more than 30% of the duration of the bout.
GPS AND PHYSICAL ACTIVITY
Medicine & Science in Sports & Exercise姞
S579
classify the location of bouts as indoors, outdoors in
neighborhood, and outdoors out of neighborhood needs
further exploration. For example, land use information
can be combined with the last GPS coordinate recorded
before a bout and the next GPS coordinate available after
a bout to make inferences about the location of the bout.
Additional research regarding how to complement accelerometry with GPS data is necessary. The study focused on
a self-selected population with high levels of physical activity. The preliminary relationships identified here must be
tested in more diverse populations and in various urban
contexts. A larger and more diverse sample of participants
may allow for stratifying the analysis by physical activity
level to uncover potential behavioral differences by levels of
activity.
This pilot study complements accelerometer data by contributing the richness of spatial behavior contained in GPS
data. Inherent in the positional data for each participant is
information about preferred and nonpreferred outdoor environments for physical activity. As shown by our analysis,
the relative comparison between environments promises to
clarify relationships and contribute to understanding of
physical activity behavior. Coupled with GIS databases that
show, for example, the location of green areas, buildings,
population activity, and streets, spatial physical activity data
provide distinct advantages compared with relying on nonspatial physical activity data. First, they expand the usefulness of objective environmental data captured through GIS
databases. Second, our measures were aggregated to a level
that made them consistent and comparable with other studies, but given the accuracy of the GPS data, using more
disaggregate measures such as street width, presence of
crosswalks, and sidewalk width shows increased promise.
Finally, the availability of GPS data opens the possibility of
validating physical activity diaries.
We are grateful to Tracy Hadden and Jeffrey Grim for their research assistance.
Financial support for this study was provided in part by a grant
from the University of North Carolina’s Research Council and a
Junior Faculty Development Award.
The results of the present study do not constitute endorsement
by the authors or ACSM of the products described in this paper.
REFERENCES
1. BROWNSON, R. C., E. A. BAKER, R. L. BOYD, et al. A communitybasedapproach to promoting walking in rural areas. Am. J. Prev.
Med. 27:28–33, 2004.
2. CATELLIER, D. J., P. J. HANNAN, and D. M. MURRAY, et al.
Imputation of missing data when measuring physical activity
by accelerometry. Med. Sci. Sports Exerc. 37:S555–S562,
2005.
3. CERVERO, R., and M. DUNCAN. Walking, bicycling, and urban
landscapes: evidence from the San Francisco bay area. Am. J.
Public Health 93:1478–1483, 2003.
4. FRANK, L. D., P. O. ENGELKE, and T. L. SCHMID. Health and
Community Design: The Impact of the Built Environment on
Physical Activity. Washington, DC: Island Press, 2003, xiii,
253.
5. FRENCH, S. A., M. STORY, and R. W. JEFFERY. Environmental
influences on eating and physical activity. Annu. Rev. Public
Health 22:309–335, 2001.
6. GILI, J. A., J. COROMINAS, and J. RIUS. Using Global Positioning
System techniques in landslide monitoring. Eng. Geol. 55:167–
192, 2000.
7. HENDELMAN, D., K. MILLER, C. BAGGETT, E. DEBOLD, and P. FREEDSON. Validity of accelerometry for the assessment of moderate
intensity physical activity in the field. Med. Sci. Sports Exerc.
32:S442–449, 2000.
8. HUMPEL, N., N. OWEN, and E. LESLIE. Environmental factors associated with adults’ participation in physical activity. Am. J.
Prev. Med. 22:188–199, 2002.
9. LARSSON, P. Global positioning system and sport-specific testing.
Med. Sci. Sports Exerc. 33:1093–1101, 2003.
10. MATTHEWS, C. E., B. E. AINSWORTH, R. W. THOMPSON, D. R.
BASSETT, JR. Sources of variance in daily physical activity levels as
measured by an accelerometer. Med. Sci. Sports Exerc. 34:1376–
1381, 2002.
11. MCCORMACK, G., B. GILES-CORTI, A. LANGE, T. SMITH, K. MARTIN, and T.
PIKORA. An update of recent evidence of the relationship between objective and self-report measures of the physical environment and physical
activity behaviours. J. Sci. Med. Sport. 7:81–92, 2004.
12. MEROM, D., A. BAUMAN, P. VITA, and G. CLOSE. An environmental
intervention to promote walking and cycling the impact of a newly
constructed Rail Trail in Western Sydney. Prev. Med. 36:235–242,
2003.
13. METCALF, B. S., J. S. CURNOW, C. EVANS, L. D. VOSS, and T. J.
S580
Official Journal of the American College of Sports Medicine
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
WILKIN. Technical reliability of the CSA activity monitor: The
Early Bird Study. Med. Sci. Sports Exerc. 34:1533–1537, 2002.
NORTHRIDGE, M. E., E. SCLAR, and P. BISWAS. Sorting out the
connections between the built environment and health: a conceptual framework for navigating pathways and planning healthy
cities. J. Urban Health 80:556–590, 2003.
OWEN, N., N. HUMPEL, E. LESLIE, A. BAUMAN, and J. F. SALLIS.
Understanding environmental influences on walking; review and
research agenda. Am. J. Prev. Med. 27:67–76, 2004.
PHILLIPS, M. L., T. A. HALL, N. A. ESMEN, R. LYNCH, and D. L.
JOHNSON. Use of global positioning system technology to track
subject’s location during environmental exposure sampling.
J. Expo. Anal. Environ. Epidemiol. 11:207–215, 2001.
PIKORA, T., B. GILES-CORTI, F. BULL, K. JAMROZIK, and R. DONOVAN. Developing a framework for assessment of the environmental
determinants of walking and cycling. Soc. Sci. Med. 56:1693–
1703, 2003.
RODRIGUEZ, D. A., A. J. KHATTAK, and K. R. EVENSON. Can community design increase physical activity: evidence from a new
urbanist and a conventional community. J. Am. Plann. Assoc. 72:
In press, 2006.
SAELENS, B. E., J. F. SALLIS, and L. D. FRANK. Environmental
correlates of walking and cycling: findings from the transportation, urban design, and planning literatures. Ann. Behav. Med.
25:80–91, 2003.
SALLIS, J., M. JOHNSON, K. CALFAS, S. CAPAROSA, and J. NICHOLS.
Assessing perceived physical environmental variables that may
influence physical activity. Res. Q. Exerc. Sport. 68:345–351,
1997.
SALLIS, J. F., A. BAUMAN, and M. PRATT. Environmental and policy
interventions to promote physical activity. Am. J. Prev. Med.
15:379–397, 1998.
SCHLECHT, E., C. HULSEBUSCH, F. MAHLER, and K. BECKER. The use
of differentially corrected global positioning system to monitor
activities of cattle at pasture. Appl. Anim. Behav. Sci. 85:185–202,
2004.
SCHMIDT, M. D., P. S. FREEDSON, and L. CHASAN-TABER. Estimating physical activity using the CSA accelerometer and a
physical activity log. Med. Sci. Sports Exerc. 35:1605–1611,
2003.
SCHUTZ, Y., and R. HERREN. Assessment of speed of human locomotion using a differential satellite global positioning system.
Med. Sci. Sports Exerc. 32:642–646, 2000.
http://www.acsm-msse.org
25. STOKOLS, D. Establishing and maintaining healthy environments: toward a social ecology of health promotion. Am. Psychol. 47:6–22, 1992.
26. SWARTZ, A. M., S. J. STRATH, D. R. BASSETT, JR., W. L. O’BRIEN,
KING, and G. A. AINSWORTH. B. E. Estimation of energy expenditure using CSA accelerometers at hip and wrist sites. Med. Sci.
Sports Exerc. 32:S450–S456, 2000.
27. TAYLOR, M. A. P., J. E. WOOLLEY, and R. ZITO. Integration of the
global positioning system and geographical information systems
for traffic congestion studies. Transp. Res. C Emerg. Technol.
8:257–285, 2000.
28. TROPED, P. J., R. P. SAUNDERS, R. R. PATE, B. REININGER, and ADDY. C. L.
GPS AND PHYSICAL ACTIVITY
Correlates of recreational and transportation physical activity among
adults in a New England community. Prev. Med. 37:304–310, 2003.
29. TUDOR-LOCKE, C., B. E. AINSWORTH, R. W. THOMPSON, and C. E. MATTHEWS. Comparison of pedometer and accelerometer measures of freeliving physical activity. Med. Sci. Sports Exerc. 34:2045–2051, 2002.
30. WELK, G. J., J. A. SCHABEN, and J. R. MORROW, JR. Reliability of
accelerometry-based activity monitors: a generalizability study.
Med. Sci. Sports Exerc. 36:1637–1645, 2004.
31. WOLF, J., M. OLIVEIRA, and M. THOMPSON. Impact of underreporting mileage and travel time estimates - Results from Global
Positioning System-enhanced Household Travel Survey.
Transp. Res. Record 1854:189–198, 2003.
Medicine & Science in Sports & Exercise姞
S581