309
Journal of Physical Activity and Health, 2013, 10, 309-322
© 2013 Human Kinetics, Inc.
Zhu, McAuley, and Chodzko-Zajko are with the Dept of
Kinesiology and Community Health, University of Illinois at
Urbana-Champaign. Nedovic-Budic is with the Dept of Geog-
raphy, Planning, and Environmental Policy, University College
Dublin, Belield, Ireland. Olshansky is with the Dept of Urban
and Regional Planning, University of Illinois at Urbana–Cham-
paign. Marti is with ARTIS, LLC, Salt Lake City, Utah. Gao
is with the Dept of Kinesiology, Boise State University, Boise,
ID. Park is with the Dept of Physical Education, Springield
College, Springield, MA.
Agent-Based Modeling of Physical Activity Behavior
and Environmental Correlations:
An Introduction and Illustration
Weimo Zhu, Zorica Nedovic-Budic, Robert B. Olshansky, Jed Marti, Yong Gao,
Youngsik Park, Edward McAuley, and Wojciech Chodzko-Zajko
Purpose: To introduce Agent-Based Model (ABM) to physical activity (PA) research and, using data from
a study of neighborhood walkability and walking behavior, to illustrate parameters for an ABM of walking
behavior. Method: The concept, brief history, mechanism, major components, key steps, advantages, and
limitations of ABM were irst introduced. For illustration, 10 participants (age in years: mean = 68, SD = 8)
were recruited from a walkable and a nonwalkable neighborhood. They wore AMP 331 triaxial accelerometers
and GeoLogger GPA tracking devices for 21 days. Data were analyzed using conventional statistics and high-
resolution geographic image analysis, which focused on a) path length, b) path duration, c) number of GPS
reporting points, and d) interaction between distances and time. Results: Average steps by subjects ranged
from 1810–10,453 steps per day (mean = 6899, SD = 3823). No statistical difference in walking behavior was
found between neighborhoods (Walkable = 6710 ± 2781, Nonwalkable = 7096 ± 4674). Three environment
parameters (ie, sidewalk, crosswalk, and path) were identiied for future ABM simulation. Conclusion: ABM
should provide a better understanding of PA behavior’s interaction with the environment, as illustrated using
a real-life example. PA ield should take advantage of ABM in future research.
Keywords: GPS, mapping, environment, statistical modeling
The impact of the environment, especially built
environment, on physical activity (PA) participation has
been well documented.
1–4
While measuring and tracking
individual PA participants, and their interactions with the
environment is possible using a combination of PA, global
positioning system (GPS), and Geographic Information
System (GIS) measures, modeling the environmental
factors or correlates and their impact using traditional
statistical methods is still a challenge. There are several
reasons for this:
1. It is very dificult to correlate travel-related PA with
the environment because these activities extend over
both time and space. In the end, PA must be assigned
to individual subjects, not to speciic locations. For
example, a heavily used pedestrian bridge could be
considered successful if it facilitated more walking;
it would be less successful if it simply replaced an
already heavily used crosswalk without having much
effect on overall walking activity levels. The bridge
is important only as a correlate of activity
2. It is dificult to quantitatively assess policy inter-
vention strategies based on discovered correlates.
For example, suppose that a survey reveals that
both crossing major roads and lack of sidewalks
are important inhibitors of pedestrians. In a typi-
cal urban setting with hundreds of dangerous road
crossings and miles of thoroughfares with no or
poor sidewalks, the question becomes which speciic
projects will produce the greatest yield in terms of
increased pedestrian use. Commonly used correla-
tional statistical methods are not appropriate because
of the cluster nature of the data (ie, participants from
a neighborhood are not independent of each other).
As a result, Type I errors in statistical analysis are
often heightened
5
3. The statistical methods that can take clustered
data into consideration in the data analysis (eg, the
hierarchical linear model, HLM)
5
assume that PA
participants are limited in macro units (eg, neighbor-
hood) being studied. This assumption is often not
true: a person who lives in walkable neighborhood
in the suburb may walk very little if he/she spends
most of their time in the city or a place where there
are no sidewalks or it is not safe to walk
Official Journal of ISPAH
www.JPAH-Journal.com
ORIGINAL RESEARCH