Journal Logo

New Methods for Assessing and Modeling Walking Behavior

Multilevel Modeling of Walking Behavior

Advances in Understanding the Interactions of People, Place, and Time

KING, ABBY C.1; SATARIANO, WILLIAM A.2; MARTI, JED3; ZHU, WEIMO4

Author Information
Medicine & Science in Sports & Exercise: July 2008 - Volume 40 - Issue 7 - p S584-S593
doi: 10.1249/MSS.0b013e31817c66b7
  • Free

Abstract

With the identification of brisk walking and other forms of regular moderate-intensity physical activity as protective for a range of chronic diseases and conditions (32,46), increased efforts have been focused on strategies for increasing such forms of physical activity across the population (16). Much of the intervention research to date has focused on personal-level approaches to increasing physical activity (20). Drawing from psychosocial and behavioral theory (e.g., social cognitive theory, the transtheoretical model, theory of reasoned action or planned behavior), the primary goals of such approaches have been to help individuals change structured, leisure-time physical activity levels through the application of empirically supported self-regulatory skills (e.g., goal-setting, self-monitoring, attainment of ongoing feedback on progress, relapse prevention training, prompts) and staff- or group-delivered social support (20). Over two decades of research have yielded a set of empirically supported individually adapted interventions that have been strongly recommended for dissemination by the US Community Preventive Services Taskforce on physical activity (43).

Despite the progress made with respect to personal-level approaches, it has become increasingly clear that such efforts alone will be inadequate for significantly curtailing the epidemic of physical inactivity observed in industrialized nations (20). Ecological approaches recognize that walking and other forms of physical activity occur within a broader social, cultural, and environmental context (38). Such perspectives have also stimulated an interest in a more diverse array of walking behaviors that include, in addition to structured, leisure-time walking, incidental walking occurring as part of household and other routine activities as well as for transportation purposes (31).

ECOLOGICAL MODELS

Ecological models reflect a multilevel perspective that considers walking and other forms of physical activity to be a product of personal, behavioral, social, cultural, and environmental factors (38). These factors, shown in Figure 1, are interactive and transactional, meaning that they can influence as well as be influenced by walking (i.e., regular walkers can influence their environment by demanding changes in their local neighborhoods, such as the addition of crosswalks, that can lead to safer walking routes). In addition to the typical ecological domains of influence shown in Figure 1, we have added the domain of time, which is discussed below.

FIGURE 1
FIGURE 1:
Ecological model, with time as an additional factor.

THE IMPACT OF TIME IN UNDERSTANDING WALKING BEHAVIORS

Time is a key component of the ecological model because the interplay of people and places can only occur over time. It is fair to conclude, however, that time has received relatively little attention in research on walking. This is not surprising given that work in this area is both conceptually and logistically challenging. There is no one definition of time, and the effects of time may differ depending upon the person and the place. There are also practical challenges. For example, the complexity and the resources that are required to conduct longitudinal studies of walking are significantly greater than those needed to undertake cross-sectional investigations. Although the study of time may be daunting, a failure to consider it will likely result in an incomplete understanding of the causes and consequences of walking behavior. Indeed, one of the most persistent concerns associated with studies of people and places is whether a specific outcome, such as walking, is affected by characteristics of the place itself or by the possibility that certain types of individuals decide to reside in that place (in effect, "selecting" themselves to live there) because it affords them an opportunity to walk (30). The likelihood of this "selection bias" can only be resolved by examining the intersection of people and places over time. In the following section, we consider different definitions of time as well as some of the challenges associated with the measurement of this important variable.

Definitions of time.

There is no one definition of time. Following from the ecological model, time intersects at the level of the individual and the environment. For example, time can be defined in terms of the life course of individuals. As people age, the timing, the frequency, the intensity, and the circumstances of walking are affected. Of note, walking itself is often used as an indicator of human maturation. In the early years, for instance, the timing and the ease of walking represent important behavioral markers of childhood development. Later in life, decrements in walking behavior are used as key indicators of loss in functional capacity, balance, and lower-body strength associated with aging. An important research question concerns how patterns of walking behavior over time are affected by the level and variability in biological, behavioral, social, demographic, and environmental characteristics (40). Along these lines, there is some evidence to suggest that levels of physical activity in the early years (i.e., childhood and adolescence) may be associated with the likelihood of physical activity later in life (44). Although to our knowledge this question has not been investigated for walking specifically, it is reasonable to hypothesize that the memories of past walking behavior as well as the memories of the circumstances and locations of those activities may conceivably affect walking later in life. In addition, the environment itself may impose time-related barriers for specific population subgroups based on their physical functioning levels. For example, in many areas of the country, the timing of traffic lights is based on the walking speed of an average adult and does not take into account the fact that many older adults typically walk at a slower gait than younger adults (as do parents walking with small children) and thus may have difficulty crossing those intersections safely (25).

Time also can be defined in terms of the time of the day, week, or season of the year. The frequency of walking to destinations, for example, school, work, or shopping, is affected by the time of the day when activities at those locations take place. Likewise, walking for leisure or exercise is likely to occur at particular times of the day and week. Because poor weather is one of the leading reported barriers to walking (33,46), walking is more likely to occur during particular seasons, especially in those geographic areas that are subject to extremes in temperature, humidity, and/or precipitation (33,39). In seasons of poor weather, walking may occur indoors through a system of enclosed walkways in urban areas such as downtown Minneapolis or through opportunities for early morning walks in enclosed shopping malls or schools.

Time can further be defined in a cultural context. For example, the timing and the circumstances of walking may be affected by the cultural expectations of residents of different geographic areas, the expectations of members of specific racial and ethnic groups, and finally the expectations of residents of the broader society. For example, the timing and the frequency of walking may differ for men and women depending upon perceptions of safety, public self-consciousness, and similar constructs.

Characteristics of place additionally may affect the timing of walking. Housing density, the location of outlets for goods and services and the proximity of those outlets to residences, the presence and quality of sidewalks and streetlights, the street patterns, the presence of parks and walking trails, and the perceived safety are all associated with walking and other forms of physical activity (12,13). As noted previously, the location and the timing of activities at specific destinations, such as school or work, affect the timing of walking to and from those destinations. If other facilities, such as restaurants and theaters, are present, then walking for some groups of individuals also may take place later in the day and into the evening hours, with the type, purpose, and circumstances of walking differing at those times. The land use and the architecture of specific places also may be designed with a specific volume and speed of walking in mind. For example, the architecture of shops and services that front sidewalks with ready access and with prominent window displays encourages a more leisurely speed of walking, which reflects another dimension of time.

Finally, time can be defined at the level of the population. In demography, temporal characteristics of the population are defined in terms of age, period, and cohort effects (47). Age, in this case, is defined as the age distribution of the population in a particular area. A period effect refers to a specific event in time. This can include a historical event or period, for example, the economic depression of the 1930s, which may affect people, albeit perhaps in different ways, across the age spectrum. Finally, a cohort effect is characteristic of a generation of people, for example, people born at a particular period, such as members of the so-called "Baby Boom Generation" who were born between 1946 and 1964. The specific characteristics of a cohort can have a particular effect on both members of that generation and others in society, such as the anticipated increase in the costs of health services as the large Baby Boom Generation begins to reach the age of 65 in 2011. This demographically based definition of time is important for investigations of walking behavior within individuals as well as across the larger population. If people perceive an area as safe, for example, they are more likely to walk in that area (28,35). It is reasonable to hypothesize that if older people feel safer among their same-age peers, they are more likely to walk in areas frequented by other older adults. Some places also may be more accommodating than others for people of particular ages. For example, older walkers may be more likely to be attracted to a shaded walkway with park benches and easy access to safe and clean public restrooms than other age groups. It is also possible that the expenditure of public funds for walking trails and other places for recreational walking (a period effect) encourages walking among people of different ages. Finally, people of a particular generation may share a common experience that could conceivably influence their levels of walking and other forms of physical activity throughout life. For example, generations that attended kindergarten through 12th grade schools at a time during which physical education classes were required may be more active over their lives than generations in which such classes were not required. (Conversely, if the manner in which mandatory physical education classes were conducted was perceived as unpleasant by a generation of individuals, then such experiences could diminish subsequent enthusiasm for engaging in physical activity later on.) Generations of women with greater access to extracurricular sports programs, following policies to enhance gender equity in sports (e.g., Title 9 funding), may be more likely to engage in walking and other forms of physical activity as they age than women who did not have such access. Age, period, and cohort or generational effects thus represent ways of defining time to better understand the behavior of populations and the decisions of individuals within the context of those populations.

Measurement of time.

There are many challenges associated with the measurement of time. Although there is growing recognition of the importance of a life-course approach in the population sciences (9,24), it is challenging to obtain longitudinal data that cover extended periods of time in population studies, as opposed to utilizing the historical cohort designs or cross-sectional time-series surveys that have been reported in other population sciences as well as in exercise science (9,24). One of the challenges for the future will be the provision of scientific and financial investment in cohort studies that will reflect the geographic and demographic diversity of the United States.

Most longitudinal studies are designed to enroll a sample of individuals and then follow those people for a specific period. In most cases, the study participants are assessed several times over that time interval. The timing and the intervals of assessment are often fairly standard, for example, at 6- or 12-month intervals. In addition, the timing and the intervals of assessment are specified by the investigators and often do not necessarily reflect or are designed to capture important "transition points" in the outcome, in this case, walking. More important, the standard longitudinal design does not afford an opportunity to assess changes that occur between these standard points of assessment. Although study participants may be asked to recall events during the time since the last assessment, retrospective assessments of that kind may not fully or accurately capture all of the relevant information. Of course, in large population studies, it is not reasonable to propose that many participants be assessed on a daily or weekly basis. It may be possible, however, to incorporate strategies that have been applied currently in smaller populations. There are techniques that have been used in psychology, admittedly with relatively small samples, to assess behavioral and health outcomes on a fairly frequent basis over a specific period. Ecological momentary assessment (EMA), for example, is a technique that allows participants either to report events as they occur or to report events on a more frequent episodic basis (41,42). Participants typically use electronic devices or paper-and-pencil diaries to record their responses to specific questions associated with the timing of specific events or activities (1,2). To our knowledge, EMA has not been used extensively in longitudinal, epidemiological studies.

Another measurement challenge concerns the integration of time across the levels of both people and place. Although multilevel modeling has focused on the integration of biologic, behavioral, social, and environmental factors, less attention has been given to the systematic evaluation and coordination of time at each of these levels of analysis. In most studies, assessments are made across these different levels at the same time intervals. However, given the different definitions and "metrics" of time across the different levels (i.e., behavioral, environmental), it is unknown whether this is appropriate. This issue needs to be assessed in greater detail.

It is reported that a person's perceptions of the environment can affect his or her decision to walk in a specific locale (3,31,35). As noted earlier, however, it is unknown whether a person's past experiences in that setting or place may affect these perceptions and subsequently his or her walking behavior.

Time, then, is an important component of the ecological model that has received minimal systematic attention to date. Although the variable of "time" has been implicit in most ecological models, the theoretical and methodological implications of this key variable have been, in our opinion, neglected. Comprehensive examinations of walking behavior among people and places require more sophisticated measurements of time-measurements that reflect the multiple definitions and impacts of this currently elusive dimension of human experience.

EMERGING APPROACHES TO UNDERSTANDING MULTILEVEL IMPACTS ON WALKING BEHAVIOR

Although the ecological model has become a focus of discussion, the development of multilevel modeling methods that capture the complex and changing web of factors influencing walking in different population subgroups has been challenging (20). In the following sections, we discuss examples of statistical and computer simulation approaches that may advance our understanding of relationships among these multilevel factors and the interplay of those factors over time.

STATISTICAL APPROACHES TO MULTILEVEL MODELING

Numerous statistical approaches have been applied to better understand the interrelationships among factors influencing walking and similar physical activity patterns, including variants of multiple regression, path analysis, and hierarchical modeling methods. However, identifying approaches that yield information that could potentially be applied in developing targeted public health interventions has been difficult. One example of an exploratory decision-oriented statistical method that may be helpful in capturing the complex multilevel interactions during walking and similar behaviors is signal detection methodology (23). Signal detection belongs to the family of nonparametric recursive partitioning (risk classification or decision tree) methods that have been used extensively in the medical literature to better understand subgroup-intervention effects. First applied in the evaluation of the sensitivity and the specificity of diagnostic tests, the method defines distinct subgroups of individuals that are mutually exclusive and maximally discriminated from one another with respect to a specific dichotomous outcome (23). This is accomplished through the application of algorithms consisting of a series of simple "and/or" decision rules (i.e., recursive partitioning) (27). As applied in the physical activity field, the dichotomous outcome can be operationalized as meeting the surgeon general guidelines of 150 min or more of moderate to vigorous physical activity per week (yes or no) (46). Strengths of signal detection methods for behavioral research include elimination of problems of collinearity and missing data across predictor variables (because each variable is evaluated in turn to determine its sensitivity and specificity in predicting success) (23). Signal detection methods are particularly useful for understanding higher-order interactions among predictor variables that are the foundation of multilevel, transactional models of health behavior change (4,23,37).

The few studies that have applied signal detection methodology as part of randomized clinical trials to better understand the confluence of factors influencing physical activity participation have been promising (18,19,21). In these studies, participants were initially underactive with respect to meeting the current national physical activity recommendations and the vast majority of participants engaged in walking as their primary physical activity. In exploring baseline predictors of intervention success, the most successful subgroups have typically been defined by acombination of personal, social, environmental, and program-related characteristics. For example, in a signal detection analysis conducted as part of the Activity Counseling Trial (45), the subgroup that had the highest 24-month success rates with respect to meeting the national recommendations of at least 150 min·wk−1 in moderate or more vigorous physical activity was that who had been randomly assigned to the physician advice + mail + telephone counseling intervention (a program variable) and who at baseline reported seeing people walking in their neighborhood (a social environmental variable) in combination with higher initial self-efficacy ratings (>63 out of 100) (a cognitive variable) (19). Over 77% of individuals in this subgroup were successful at meeting the national physical activity recommendations at 24 months. When the same subgroup was formulated, using the original signal detection-chosen variables and cutpoints, in the other two program arms (physician advice only, physician advice + mail), the 24-month success rates in these other two arms were significantly lower (58% and 47%, respectively) than that in the counseling arm, indicating the potential impact of program type in combination with the baseline cognitiveandsocial environmental factors described above on 24-month physical activity success (19). This subgroup also demonstrated the highest success rates at each measured time point (6, 12, 18, and 24 months), suggesting behavioral consistency over time (19).

A recently completed signal detection analysis across five randomized controlled trials evaluating varied physical activity interventions in different regions of the country also suggests the potential utility of such decision tree methods in understanding how variables may interact in influencing physical activity levels within and across different locales (21). Cross-sectional pooled signal detection analysis across the five clinical trial sites (i.e., San Francisco peninsula, CA; Eugene, OR; Atlanta, GA; Kingston, RI; and Memphis, TN) indicated that people who were most likely to meet current national physical activity recommendations reported living in neighborhoods with more attractive scenery (i.e., neighborhood aesthetics) in combination with greater ease or pleasantness of their neighborhood for walking-67% of this subgroup achieved the national recommendations, whereas only 36% of those reporting attractive neighborhood scenery yet lesser ease or pleasantness of their neighborhood for walking met the current national physical activity recommendations (21). Although neighborhood aesthetics and ease and pleasantness of one's neighborhood for walking each have been associated with physical activity levels in other studies (15), this study is the first to suggest a potentially important interaction between these two environmental variables in explaining moderate or more vigorous physical activity. Notably, the combinations of perceived environmental variables were found to be better discriminators of physical activity levels in the signal detection analysis than were demographic variables (i.e., age, gender, years of education, annual household income, ethnicity, site location) (21).

The potential importance of the social environment (specifically, the program delivery agent used) for different subgroups of people has been suggested in a recent physical activity intervention trial comparing physical activity advice delivered by trained health educators versus an automated telephone-linked counseling system (17). In this randomized trial, 218 community-dwelling adults ages 55 years and older who were healthy but initially physically underactive went through baseline evaluation and were subsequently randomized to a 12-month attention-control (health education) arm, a telephone-based physical activity human advice arm, or a telephone-based physical activity automated advice arm. The two physical activity programs focused primarily on brisk walking, and participants were counseled, based on social cognitive theory-derived strategies (5), to increase their physical activity levels commensurate with the national physical activity recommendations (46). We speculated, based on social influence theory (6), that individuals who were comfortable interacting with other people would do particularly well with the human advice intervention, whereas those who were not (e.g., were socially anxious) would perform less well with that intervention (17). Similarly, we speculated, based on self-determination theory (34), that individuals who were oriented toward independence and autonomy and who reported less willingness to receive direction and advice from other people would do particularly well with the automated advice arm (17). Relevant baseline measures reflecting the above dispositional tendencies (i.e., social anxiety, an orientation toward not wanting to be controlled or directed by others) were entered into a signal detection model along with an array of other psychosocial as well as demographic and physical or health variables. The dichotomous outcome being predicted was success or failure in achieving the national physical activity recommendations at 6 months, measured by the Stanford Physical Activity Recall (36). For those participants randomized to the human advice program, the subgroup that had the lowest success rate in achieving the national recommendations was defined by having low baseline barriers self-efficacy ratings (<59 on a 100-point scale) (11) in combination with above-average baseline social anxiety ratings (>8 on a 0-18 scale) (10,26). Only 14% of this subgroup achieved the national recommendations at 6 months compared with 58% of participants overall who had been assigned to the human advice program. Meanwhile, a signal detection analysis conducted with participants randomized to the automated advice program indicated that the subgroup with the lowest 6-month success rate with that program had baseline performance self-efficacy ratings <89 out of 100 (only 24% of that subgroup succeeded). The subgroup that achieved the best success rate with that program had performance self-efficacy levels of at least 89 (11) in combination with low baseline controlled orientation scores (indicating a preference for not being told what to do by other people) (34). Eighty-four percent of that subgroup achieved the national recommendations at 6 months (compared with 49% of the participants overall who were assigned to the automated advice program).

Together, the above studies demonstrate the promise of signal detection methods, and similar approaches in promoting our understanding of how variables may interact in influencing physical activity levels, setting the stage for more targeted interventions aimed at specific population subgroups. It is important to note, however, that similar to multiple regression and other prediction-oriented methods, signal detection and other classification or decision tree methods are exploratory in nature. That is, they can be useful in generating hypotheses related to which subgroups of people might fare best with which types of physical activity interventions. The next step consists of experimentally testing whether certain subgroups identified by such exploratory methods would indeed become more regularly active with one intervention versus another. Such experiments need to be performed within the same population targeted in the original exploratory analysis, given the population specificity of the predictors identified in such exploratory studies.

EXPLORING MULTILEVEL IMPACTS ON WALKING THROUGH THE USE OF COMPUTER-BASED SPATIAL SIMULATIONS

The ecological model allows us to look beyond traditional physical activity outcomes, such as leisure-time exercise, to include spatial behavior (48). As noted previously, all physical activity includes a specific environmental context to meet an individual's biological, psychological, and/or sociological needs. This means that in addition to recording physical activity types and energy expenditure amounts, there is also interest in knowing when and where the activity was undertaken, for what purpose, and the connections occurring among different activities. Considering walking, for example, we are not only interested in how many steps a person has taken, but also where and when the steps were taken, for what purpose (e.g., social, household-related, transport, exercise, leisure), and links between walking and other activities. Because environmental obstacles, both physical and psychological, are often barriers to physical activity, assessing and understanding physical activity as a spatial behavior should be beneficial. Commonly used physical activity methods (e.g., accelerometers or questionnaires), however, have focused primarily on measuring the accumulated outcomes of the physical activities, for example, energy spent or total movements recorded, and provide little information about the interactions or relationships between environments and specific physical activities.

To address this need, Zhu (48) recently proposed a new concept or measure called "physical activity space" (PAS), which is defined as "the area or space where an individual spends time and engages in physical activities." An individual's PAS is comprised of a home base or a residence and other activity sites, for example, workplace, schools, stores, restaurants, fitness clubs, recreational areas, and pathways traveled to and from home and other activitysites. The size and shape of an individual's PAS will vary depending on the activities the individual is obligated or chooses to perform, the modes of transportation available, and the geographical locations.

Time, space, and corresponding activities are three key elements in measuring PAS. Time simply refers, in this context, to a moment, and space refers to a location connected to the behavior. Based on this conceptualization, human daily activity patterns are comprised of three major components: a) the time during which an activity occurs; b) the space through which the activity takes place; and c) the type of activity. Time and space are thus inseparable from the intricacies of human behavior, including physical activity. When connecting space with time, a human activity "pathway" can be generated.

With recently developed satellite technology-based devices [i.e., global positioning system (GPS)], it has become possible to capture an individual's physical movements through space and time using GPS in combination with geographic information systems. Preliminary efforts have been made in Zhu's kinemetrics laboratory to develop measurement tools that can measure PAS. Using a global positioning system (GPS) device and an ambulatory physical activity recorder (e.g., accelerometer) placed on individuals across a 21-d period, some pilot data have been collected relative to individuals' PAS. Figure 2 shows the walking spaces of a 79-yr-old woman on two different days. A rather consistent walking space, according to the time, the location, and the size of the space, was observed.

FIGURE 2
FIGURE 2:
Walking space example.

Agent-based modeling (ABM).

To fully understand the rich information collected by a space measure, traditional analytical methods may no longer be adequate, including several advanced multilevel statistical procedures (e.g., hierarchical linear models), because many of these procedures' assumptions are often violated in attempts to connect individuals with the spaces that they occupy (e.g., assuming physical activities of the residents in a neighborhood are clustered within the neighborhood being studied). Fortunately, these limitations can be eliminated by applying agent-based modeling (ABM) and complex adaptive system (CAS) techniques. The heart of ABM is the "agent," the subject or the individual with a set of characteristics or attributes. An agent's behavior, for example, how he or she responds to the environment or interacts with other agents (i.e., individuals) in the system, is determined by a set of plans that simulate daily activities and rules that govern the interactions between the simulated individuals or agents. A plan might include daily dog walking, whereas rules would limit this behavior during rainy weather. An ABM is most often realized as a large computer program that makes exploration of these different plans and rules possible.

Important characteristics of agent-based models.

Agent-based models attempt to simulate humans in their environment and, with sufficient numbers, derive characteristics from their aggregate behaviors. Archeologists, economists, sociologists, and biologists use these techniques to simulate activities ranging from ancient Anasazi village traffic in the Southwestern United States (22) to recreational river use of the Grand Canyon (14). Although hierarchical modeling examines human behavior derived from measurable behaviors with increasing levels of detail, ABM layers simple physical behavioral models in ever increasing levels of abstraction. An ABM's greatest level of detail might simulate the steps and the route a subject takes from the office to the nearest restaurant while higher levels would control the route taken, the decision to drive or walk, or what to do if the restaurant is too crowded when the individual gets there. Other programs (e.g., a CAS program) might also be developed to modify the route if the subject learns of construction or traffic congestion.

An ABM's agents are represented by computer code that models some real-world behavior, such as walking. Typically, we implement these in the object-oriented programming style: humans are a particular object type as are their cars, children, and houses. We extend this programming style to include other agents-some are physical entities whereas others are concepts such as road and sidewalk networks, aesthetic areas, plans, goals, and behaviors (7). Agents pursue activities over time through events: walking to the park, driving to work, and driving the children to school. These events can be modeled across days, weeks, months, and even years using the discrete-event simulation technique. The computer does not simulate activities occurring between events-only the events themselves. For example, if an agent's goal is to get to work, a plan programmed into the computer might involve having the agent walk to work but drive if the weather is bad.

Uses of agent-based models.

With sufficient data, specified rules, plans, and goals can be built into a computer programmed ABM so that it functions as a type of human behavior laboratory. The student can gain insights into populations en masse through computer animation and subsequent measurement of outcomes of interest. For example, varying the average distances various subpopulations will walk for entertainment, the researcher can examine the average walking distance of the entire simulated population in their computer-based environment. In this manner, the research can model a physical activity intervention strategy (e.g., having people walk or bicycle rather than drive to work) and analyze the outcome before committing to human subject testing.

Important components of an agent-based model.

Successfully exploiting an ABM requires four activities: a) initial data collection with the actual population and environment(s) of interest to develop a basic appreciation of the physical activity patterns of potential relevance to that population (examples of types of useful data to collect include measured daily activities from subpopulations collected through ecological momentary assessment (EMA) or similar assessment techniques occurring in the natural environment); b) scenario generation, which involves converting the collected information into rules, plans, goals, and other parameters of the agent-based model; c) execution of the simulation to establish bounds on the outcomes; d) analysis of results; and e) validation of the results. For a human activity simulation, one needs geographic data and population demographics. The researcher can download free, high-quality geographic data from the USGS World Wide Web site (www.usgs.gov), including local topographic data (elevation) and ground cover-trees, fields, urban areas, concrete, and water. Roads and other geographic features can also be taken from the US Census Tiger files. Many states and municipalities maintain detailed maps of their streets, sidewalks, American Disabilities Act ramps, bike lanes, and even traffic patterns that can be useful with varying levels of effort. Other environmental information should also be considered to meet the requirements of an ecological model as discussed earlier. Types of useful information will typically include sufficient details of subjects' activities beyond just their location and physical activity levels. In addition to programming in actual locations individuals may visit (based, for example, on information recorded in the "real world" using GPS units and similar devices), a sophisticated ABM system must also know the reason for the agents' movements, for instance, for grocery shopping, exercise, walking to the bank, driving to the airport, and so on.

A scenario specifies the agents' activities over time. Because there are as many activities as there are humans, specifying this is difficult: we typically resort to simple plans for population subgroups that focus on essential activities of interest. The computer plan for a person that walks to work might resemble the following:

  • At the designated time to walk to work ± 5 min, walk towork.
  • If not raining, at 11:30 a.m. ± 5 min, walk to Liberty Park, then after 30 min ± 5 min, walk back to work.
  • At 5 p.m. ± 10 min, walk home.

The scenario generation program's job is to assign this plan to only the percentage of the population that walks to work (based on such data for one's locale and population of interest) and only to those in areas where it is likely to occur (e.g., urban areas as opposed to suburban locales). Such scenarios can reflect time (e.g., walking or driving to work in the morning hours) as well as place parameters.

The computer model is executed many times, both to encompass the possible range of outcomes and to vary the values of some of the parameters that drive it. Modern computers are capable of this activity without great expense. Where many hundreds of thousands of runs are required, networks of computers can be used, with results communicated over public networks.

The data collected from the many executions of an ABM program can represent more than one possible sequence of the day's events. If a model simulates and measures the distance a subpopulation walks in a day, the results can be stated for each individual as "on an average Monday in June, Subject 500 walked between 5 and 15 kilometers with an average of 12 kilometers," thus encompassing executions when it rained and the subject walked little, as well as sunny, temperate days when walking was the norm. One can also view the results as an animation, with routes and trajectories for walkers, automobile traffic, and others maneuvering on a map of the area.

Finally, both the model and its program must be validated. This can range from simply viewing the animations to gain insights into walking patterns to the application of sophisticated statistical techniques. The goal is to demonstrate that measurements of the model's activities are sufficiently close to measurements taken from actual subpopulations to inspire confidence.

ABM: limitations and future directions.

Although GPS-based physical activity space (PAS) measures have provided the field with a new means of understanding physical spatial behaviors, they have several limitations. First, GPS can only capture locations outdoors. Second, although technologies are emerging that combine GPS with physical activity measurement devices (e.g., accelerometers, pedometers), such combination devices are currently expensive and require further testing and validation. Third, most exercise science researchers have not been trained on spatial analytical methods. Although the research community has successfully used ABM in other scientific arenas, much remains to be learned with respect to its applications in the physical activity or walking arena. Many samples of human activity patterns need to be collected in such a way that they can be used to drive the simulated behavior of large population subgroups. This requires that better data collection devices be developed that can be worn unobtrusively (e.g., as part of clothing), require no intervention from the user, and collect multilevel data such as location and energy expenditure. Researchers in the field need to continue to develop better measures of environmental aesthetics (everything from weather to graffiti) and similar parameters, along with ways of collecting such objective and perceptual data in real time. This will include assessments of time in terms of the duration of the activity as well as assessments of time in terms of when the activity took place, that is, the time of day, week, month, orseason of the year. Fourth, we need better computing environments than the eclectic collection of geographic information, weather generation, statistical analysis, and simulation systems currently in use. These should be usable by the general research community rather than requiring many specially trained individuals for their operation. Finally, ethical issues have been raised by some researchers with respect to the appropriateness of monitoring people's locations or positions without their express permission. Current laws generally allow the monitoring of people's positions in public places as long as individuals cannot be identified. However, heightened sensitivity with respect to privacy issues and similar concerns among the American public may deter the use of GPS-based technologies for public health purposes in at least some locales. These and related issues await further investigation.

SUMMARY AND RECOMMENDATIONS FOR FUTURE RESEARCH IN THE FIELD

Ecological frameworks have proven to be heuristic models for capturing the complex and dynamic factors influencing walking and other physical activity behaviors. The above discussion highlights three areas (i.e., adding the concept of time to ecological models; applying clinical decision-oriented statistical methods to better capture combinations of multilevel variables predictive of intervention success or failure; and developing activity space models that take advantage of GPS and other emerging technologies). All three of these areas aim to broaden our conceptualizations of the multilevel determinants of walking and other forms of physical activity through applications of concepts and methods from other disciplines and fields. Of relevance to the issues discussed in this article, the physical activity promotion field could benefit from further investigation in many areas, including the following:

  • More frequent measurement of potentially influential factors across multiple levels of impact (i.e., personal, organizational, environmental, policy related), so as to better determine the time course through which such factors exert their effects. Emerging ecological momentary assessment (EMA) techniques and similar approaches have made more frequent measurement both more feasible and less costly. Although it is possible to provide electronic devices to a large population sample, it may not be financially or logistically reasonable at this time to do so. Alternatively, it may be possible to recruit a representative sample of subjects from the larger study cohort to complete assessments of walking between the times of the regular assessments. The results of this "nested study" could be generalized to the larger study cohort through data imputation methods (29). It may be possible to coordinate the timing of the EMA with portable GPS devices. As an individual walks in a neighborhood, a change in a direction or a sustained stop may cue the EMA device, via an accompanying GPS device, to prompt the study participant to report the reasons and circumstances of the change in walking pattern. This is a developing area of needed research that holds promise for both understanding as well as intervening in walking behaviors within the actual time and place context in which they are occurring.
  • Related to the above, development of more dynamic, behaviorally driven definitions of physical activity "adoption" and "maintenance," as opposed to the arbitrary periods currently in use to define these important behavioral processes (e.g., the traditional labeling of the first 6 months of physical activity increases as "adoption").
  • Inclusion of other health behaviors as part of the ecological framework being applied in the field to better understand the dynamic interplay between physical activity and other important health behaviors (e.g., dietary intake, smoking behavior, alcohol use). There is evidence, for example, that health behaviors are not mutually exclusive but rather follow specific patterns. Behaviors of this kind are found to "cluster" among subgroups in the population (8).
  • Application of methods that facilitate the direct translation of research findings to public health policy and practice. For example, the identification, via signal detection methodology, of specific subgroups based on public health-relevant outcomes allows for the subsequent systematic evaluation of interventions targeted to such subgroups, which could, in turn, set the stage for public health applications.
  • Continued development of computer systems that can be used by the research community to exploit available human and environmental data and models for analysis. For example, if EMA data were available relevant to walking, such data could be used in combination with demographic and environmental data to develop computer simulations of human walking behavior. ABM and similar simulation methods provide a potentially powerful tool for understanding how the different domains of individual behavior may interact with physical, social, cultural, and spatial environments to influence daily decisions related to walking and other forms of physical activity.
  • Recognition that physical activity is a human spatial behavior that is closely associated with time and location. To fully understand walking behaviors and their interaction with the environment, individuals' PAS needs to be taken into account, and factors that affect this PAS must be investigated.

Finally, a research-practice agenda should be established to better coordinate simulations and other scientific investigations with field studies in diverse populations. This should include a determination of the situations under which what has been learned in the laboratory can best contribute to the development of community-based interventions to enhance walking in a range of populations.

By continuing to expand the current knowledge base to include the impacts of time, space, and similar domains, the promise of the ecological model and similar multidimensional frameworks will be better realized. With such advancements, more powerful and better-targeted interventions, taking into account the most relevant aspects of individuals' lives, may emerge.

REFERENCES

1. Atienza AA, Collins R, King AC. The mediating effects of situational control on social support and mood following a stressor: a prospective study of dementia caregivers in their natural environment. J Gerontol B Psychol Sci Soc Sci. 2001;56B:S120-39.
2. Atienza AA, King AC. Comparing self-reported versus objectively measured physical activity behavior: a preliminary investigation of older Filipino American women. Res Q Exerc Sport. 2005;76:358-62.
3. Ball K, Bauman A, Leslie E, Owen N. Perceived environmental aesthetics and convenience and company are associated with walking for exercise among Australian adults. Prev Med. 2001;33:434-40.
4. Bandura A. Social cognitive theory: an agentic perspective. Annu Rev Psychol. 2001;52:1-26.
5. Bandura A. Toward a psychology of human agency. Psychol Sci. 2006;1:164-80.
6. Basic Behavioral Science Task Force of the National Advisory Mental Health Council. Social influence and social cognition. Am Psychol. 1996;51:478-84.
7. Benenson T, Torrens PM. Geosimulation: Automata-Based Modeling of Urban Phenomena. Chichester, England: John Wiley and Sons, Ltd, 2004.
8. Berrigan D, Dodd K, Troiano RP, Krebs-Smith SM, Barbash RB. Patterns of health behavior in U.S. adults. Prev Med. 2003;36:615-23.
9. Dannefer D. Cumulative advantage/disadvantage and the life course: cross-fertilizing age and social science theory. J Gerontol B Psychol Sci Soc Sci. 2003;58B:S327-37.
10. Fenigstein A, Scheier MF, Buss AH. Public and private self-consciousness: assessment and theory. J Consult Clin Psychol. 1975;43:522-7.
11. Garcia AW, King AC. Predicting long-term adherence to aerobic exercise: a comparison of two models. J Sport Exerc Psychol. 1991;13:394-410.
12. Giles-Corti B, Broomhall MH, Knuiman M, Collins C, Douglas K, Ng K, Lange A, Donovan RJ. Increasing walking: how important is distance to, attractiveness, and size of public open space? Am J Prev Med. 2005;28:169-76.
13. Giles-Corti B, Donovan RJ. The relative influence of individual, social and physical environment determinants of physical activity. Soc Sci Med. 2002;54:1793-812.
14. Gimblett HR, Roberts CA, Daniel TC, et al. An intelligent agent-based model for simulating and evaluating river trip scenarios along the Colorado River in Grand Canyon National Park. In: Integrating Geographic Information Systems and Agent-Based Modeling Techniques for Simulating Social and Ecological Processes. New York: Oxford University Press, 2002.
15. Humpel N, Owen N, Leslie E. Environmental factors associated with adults' participation in physical activity. Am J Prev Med. 2002;22:188-99.
16. Kahn EB, Ramsey LT, Brownson RC, Heath GW, Howze EH, Powell KE, Stone EJ, Rajab MW, Corso P. The effectiveness of interventions to increase physical activity. A systematic review. Am J Prev Med. 2002;22:73-107.
17. King AC, Friedman R, Marcus B, Castro C, Forsyth L, Napolitano M, Pinto B. Harnessing motivational forces in the promotion of physical activity: the Community Health Advice by Telephone (CHAT) Project. Health Educ Res. 2002;17:627-36.
18. King AC, Kiernan M, Oman RF, Kraemer HC, Hull M, Ahn D. Can we identify who will adhere to long-term physical activity? Application of signal detection methodology as a potential aid toclinical decision-making. Health Psychol. 1997;16:380-9.
19. King AC, Marcus BH, Dunn AL, et al. Identifying subgroups who succeed or fail with three physical activity interventions: the Activity Counseling Trial. Health Psychol. 2006;25:336-47.
20. King AC, Stokols D, Talen E, Brassington GS, Killingsworth R. Theoretical approaches to the promotion of physical activity: forging a transdisciplinary paradigm. Am J Prev Med. 2002;23(2S):15-25.
21. King AC, Toobert D, Ahn D, et al. Perceived environments as physical activity correlates and moderators of interventions in five studies. Am J Health Promot. 2006;21:24-35.
22. Kohler TA, Van West CR, Carr EP, Langton CG. Agent-based modeling of prehistoric settlement systems in the northern American Southwest. In Proceedings of the Third International Conference/Workshop on Integrating GIS and Environmental Modeling. Santa Barbara, CA: National Center for Geographic Information Analysis, 1996.
23. Kraemer HC. Assessment of 2 × 2 association: generalization of signal detection methods. Am Stat. 1988;42:37-49.
24. Kuh D, Ben-Shlomo YE. A Life Course Approach to Chronic Disease Epidemiology, 2nd ed. New York: Oxford University Press, 2004.
25. Langlois JA, Keyl PM, Guralnik JM, Foley DJ, Marottoli RA, Wallace RB. Characteristics of older pedestrians who have difficulty crossing the street. Am J Public Health. 1997;87:393-7.
26. Leary MR. Social anxiety, shyness, and related constructs. In: Measures of Personality and Social Psychological Attitudes. Robinson JP, Shaver PR, Wrightsman LS (Eds.) San Diego, CA: Academic Press, Inc., 1991, pp. 161-94.
27. Lemon SC, Roy J, Clark MA, Friedmann PD, Rakowski W. Classification and regression tree analysis in public health: methodological review and comparison with logistic regression. Ann Behav Med. 2003;26:172-81.
28. Li F, Fisher KJ, Brownson RC, Bosworth M. Multilevel modeling of built environment characteristics related to neighborhood walking activity in older adults. J Epidemiol Community Health. 2005;59:558-64.
29. Mendes DL, Guralnik CJM, Bandeen-Roche K. Short term change in physical function and disability: the Women's Health and Aging Study. J Gerontol B Psychol Sci Soc Sci. 2002;57:S355-65.
30. Oakes JM, Kaufman JSE. Methods in Social Epidemiology. San Francisco, CA: Jossey-Bass, 2006.
31. Owen N, Humpel N, Leslie E, Bauman A, Sallis JF. Understanding environmental influences on walking; Review and research agenda. Am J Prev Med. 2004;27:67-76.
32. Pate RR, Pratt M, Blair SN, et al. Physical activity and public health: a recommendation from the Centers for Disease Control and Prevention and the American College of Sports Medicine. J Am Med Assoc. 1995;273:402-7.
33. Pivarnik JM, Reeves MJ, Rafferty AP. Seasonal variation in adult leisure-time physical activity. Med Sci Sports Exerc. 2003;35(6):1004-8.
34. Ryan RM, Deci EL. Self-determination theory and the facilitation of intrinsic motivation, social development, and well-being. Am Psychol. 2000;55:68-78.
35. Saelens BE, Sallis JF, Black JB, Chen D. Neighborhood-based differences in physical activity: An environment scale evaluation. Am J Public Health. 2003;93:1552-8.
36. Sallis JF, Haskell WL, Wood PD, Fortmann SP, Rogers T, Blair SN, Paffenbarger RS, Jr. Physical activity assessment methodology in the Five-City Project. Am J Epidemiol. 1985;121:91-106.
37. Sallis JF, Owen N. Ecological models. In: Health Behavior and Health Education: Theory, Research, and Practice. Glanz K, Lewis FM, Rimer BK (Eds.) San Francisco: Jossey-Bass, 1997, pp. 403-24.
38. Sallis JF, Owen N. Ecological models of health behavior. In: Health Behavior and Health Education: Theory, Research, and Practice, 3rd ed. Glanz K, Rimer BK, Lewis FM (Eds.) San Francisco: Jossey-Bass, 2002, pp. 462-84.
39. Salmon J, Owen N, Crawford D, Bauman A, Sallis JF. Physical activity and sedentary behavior: A population-based study of barriers, enjoyment, and preference. Health Psychol. 2003;22:178-88.
40. Satariano WA, McAuley E. Promoting physical activity among older adults: from ecology to the individual. Am J Prev Med. 2003;25(3Sii):184-92.
41. Shiffman S, Hufford MR. Ecological momentary assessment. Appl Clin Trials. 2001:1-4.
42. Smyth JM, Stone AA. Ecological momentary assessment research in behavioral medicine. J Happiness Stud. 2003;4:35-42.
43. Task Force on Community Preventive Services. Increasing physical activity: a report on recommendations of the Task Force on Community Preventive Services. Morb Mortal Wkly Rep. 2001;50:1-14.
44. Taylor W, et al. Childhood and adolescent physical activity patterns and adult physical activity. Med Sci Sports Exerc. 1999;31(1):118-23.
45. The Writing Group for Activity Counseling Trial Research Group. Effects of physical activity counseling in primary care: the Activity Counseling Trial: a randomized controlled trial. J Am Med Assoc. 2001;286:677-87.
46. US Department of Health and Human Services. Physical Activity and Health: A Report of the Surgeon General. Atlanta, GA: US Department of Health and Human Services, Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion, 1996.
47. Young TK. Population Health: Concepts and Methods. New York: Oxford University Press, 1998.
48. Zhu W. Assessing physical activity space: issues and challenges. In Proceedings II of the 2003 Daegu Universiade Conference: Facing the Challenge. Yeungnam University, Korea: Daegu Universiade Conference Organizing Committee, 2003, 601-8.
Keywords:

PHYSICAL ACTIVITY; ECOLOGICAL MODEL; MULTILEVEL MODELING; SIGNAL DETECTION METHOD; TIME; PHYSICAL ACTIVITY SPACE; AGENT-BASED MODELING

©2008The American College of Sports Medicine