Designing Research to Study the Relationship Between the Built Environment and Physical Activity
As discussed in Chapter 4, the built environment can facilitate or constrain physical activity. This chapter is focused on designing research to provide a more rigorous understanding of how the built environment explains physical activity levels—the charge of this study. More important from a policy perspective, the discussion is concerned with issues of causality—the extent to which it can be said that the built environment affects physical activity, and the strength and magnitude of that effect. The chapter starts with an overview of the role of theory in studying the relationship between the built environment and physical activity. It then turns to a discussion of appropriate research designs and availability of data.1
THE ROLE OF THEORY
A theoretical framework that links the built environment to physical activity is critical to good research in this area. Theory provides the basis for formulating testable hypotheses and helps in the interpretation of results. It explains the subjects, variables, and relationships a researcher chooses to study.
Indeed, one of the primary limitations of research to date on the relationship between the built environment and physical activity is the lack of an agreed-upon theoretical framework (Handy 2004). This deficiency is not surprising in view of the relatively recent interest in the topic and the fact that the necessary research must draw on expertise in at least two fields—public health and transportation. A recent review of the literature on the environmental factors associated with physical activity revealed that the conceptualization and measurement of environmental factors “comprise a relatively new area of research,” and these attributes are “among the least understood of the known influences on physical activity” (Humpel et al. 2002, 188). The authors lay the problem squarely at the door-step of inadequate theory: “Currently, even the most relevant theory does not provide sufficiently detailed conceptual tools for differentiating how the separate domains of environmental influences [e.g., accessibility, safety, aesthetics, weather] might impact on different physical activity behaviors” (Humpel et al. 2002, 197).
Research on travel behavior has drawn primarily on demand theory, as pioneered by McFadden (1974) in his Nobel prize–winning work on travel behavior modeling. The basic proposition, derived from economics and psychology, is that individuals make decisions in their self-interest, given the option to do so. In other words, most choices are made on the basis of their feasibility and their relative costs and benefits to the individual. Thus, for example, one would assume that people would be more likely to walk if walking trips became more pleasant, safer, or in any sense easier, or if alternatives to walking became more costly or more difficult.
This approach has been used primarily to forecast travel demand by motorized modes, generally for work trips and often at aggregate (regional) levels, to understand the likely impacts of alternative transportation investments on facility performance (Handy 2004). Demand theory is quite general in principle and can integrate individual perceptions and attitudes, detailed attributes of travel alternatives, and connections between short-term travel choices and long-term decisions about automobile owner-
ship and residential and employment locations in consistent and counterintuitive ways. For example, the new-urbanist literature often states that denser neighborhoods will, intuitively, lead to less driving and more walking (e.g., Duany et al. 2000). Crane (1996a; 1996b), however, uses the demand framework to demonstrate that shorter trips actually stimulate trip taking by car, and the net result for both walking and driving is unclear. Yet walking and cycling have not been the focus of demand modeling to date, and hence the usefulness of the approach as a framework for understanding physical activity behavior has not been realized.
In summary, the main value of the demand approach is its power to explain how complex behaviors change with external circumstances. The way the utility-maximizing framework is commonly applied in modeling travel behavior would need to be altered for it to serve as an appropriate method for analyzing the relationship between the built environment and physical activity. Needed modifications include specifying benefits for walking and cycling as mode choices that are different from those for motorized travel. Minimizing travel time and cost and maximizing comfort are key determinants of motorized travel; the choice of walking or cycling depends more on the importance of combining exercise with utilitarian travel and minimizing the potential for collisions, temperature extremes, rainfall, and adverse terrain. In addition, characteristics of the built environment need to be incorporated into choice algorithms, as does day-to-day variation so that some minimum of walking or cycling time per week, for example, can be included in the overall travel pattern.
Health behavior research, including research on physical activity, has drawn heavily on theories from the field of psychology (Handy 2004). Social cognitive theory, developed by Bandura—an important influence on physical activity research—explains behavior as the interplay among the person, the behavior, and the environment in which the behavior is performed (Bandura 1986). Concepts such as the importance of perceptions and objective factors and the role of motivation and self-efficacy in overcoming barriers have been influential in understanding physical activity
behavior. In general, the theory has emphasized the social rather than the physical environment (Handy 2004).
Ecological models evolved from social cognitive theory. The former emphasize the role of the physical as well as the social environment and thus extend social cognitive theory in ways that are more appropriate for analyzing the link between the built environment and physical activity (King et al. 2002; Sallis and Owen 2002). Ecological models, however, lack specificity about the characteristics of the built environment that might influence behavior. This lack of an agreed-upon conceptualization of the built environment helps explain the inconsistent approach to defining and measuring environmental variables in empirical research in this area, the subject of the next chapter.
Drawing heavily on both demand and ecological models, the committee developed its own conceptual model (see Figure 1-1 in Chapter 1). This scheme emphasizes a more detailed specification of both the built environment (e.g., different geographic scales, potentially relevant environmental characteristics at each scale) and physical activity (by type) (Figure 1-2). However, the specific elements of the model, such as the characteristics of the built environment, are illustrative rather than exhaustive.
Appropriate research designs are also important in testing relationships among variables and in selecting relevant data. This section begins with a brief discussion of research design issues, particularly as they apply to the issue of establishing causal connections. Various research designs are then identified, their strengths and weaknesses are discussed, and their relevance in analyzing the link between the built environment and physical activity is considered.
Making the Causality Connection
The key question from a public health perspective is whether the built environment in place today affects physical activity in ways that are detrimental to health. As documented in Chapter 2, the
causal link between adequate levels of physical activity and health is well established. The other half of the equation, the causal connection between the built environment and physical activity levels, is less well understood (Handy 2004; Boarnet 2004).
In considering research designs and evidence of causality in the relationship between the built environment and physical activity, there are several points to keep in mind. First, conclusions must be based on the results of many studies using a variety of research designs. All studies have weaknesses, and no single study will be sufficient to permit reaching conclusions. Second, the care with which research is performed is more important than the inherent strength or weakness of any given research design. A carefully conducted study with a weaker design generally is to be preferred over a less carefully conducted study with a better design. The care with which a study is performed is demonstrated in the theoretical underpinnings of the research, the use of the most appropriate design for the situation to be examined, the care with which exposures and outcomes are measured, consideration of biases, and the appropriateness of analytic methods.
By nature, causality is a time-ordered process: events or changes, such as an improvement in the built environment (e.g., the addition of a sidewalk), may have a consequence or effect (e.g., an increase in walking). Thus, time-series analyses generally provide the most appropriate research design for investigating cause-and-effect relationships.
The most persuasive scientific evidence of causality usually is derived from experimental studies of individuals. In such studies, subjects are randomly assigned to the exposures of interest and followed for the outcome of interest. The assignment to an exposure group is based on the needs of the study and not the participating individuals, although the risk of harm cannot knowingly be greater for members of any exposure group (Rothman and Greenland 1998). In the
case of determining causal connections between the built environment and physical activity, the exposures would be to certain types of built environments, and the outcomes would be the types and amounts of physical activity performed. The important advantage of experimental studies is that researchers have considerable control over all aspects of the study, including the type of exposure, the selection of subjects, and the assignment of exposure to the subjects. When they are conducted well, experimental studies ensure that the exposure precedes the outcome, at least two doses of exposure are administered, and subjects are randomly assigned to exposure groups. This procedure minimizes the probability that the results and conclusions will be biased. Randomized clinical trials for drug testing are well-known examples of experimental studies.
Despite their advantages, experimental studies of individuals are not always possible. It is difficult to imagine, for example, how experimental studies of the relationship between the built environment and physical activity behaviors could be used to examine more than a small portion of the areas of interest. Modifying or creating new built environments just to conduct experimental research is, for practical purposes, impossible. Likewise, randomizing participants to specific residential or employment locations is implausible. Even if these barriers could be overcome, the limited, artificial, and intrusive nature of the experiment would likely jeopardize the generalizability of the results (Caporaso 1973). Similar limitations apply to laboratory experiments, which in this context could refer to asking subjects about their responses to hypothetical situations (e.g., preferences for different types of residential locations whose characteristics were systematically varied by the analyst’s design)—referred to in the travel behavior literature as “stated preference” or “stated response” studies (Louviere et al. 2000). Fortunately, there are alternatives to experimental studies, commonly termed observational studies.
Nonexperimental research designs are often referred to as observational studies because the researcher has little or no control over many aspects of the study and instead becomes a careful observer.
The terminology applied to the numerous designs of such studies varies across disciplines. A few of the most common designs and their main characteristics are reviewed here.
Longitudinal Studies Also called cohort, concurrent, follow-up, incidence, and prospective studies, longitudinal studies are those in which individuals have different levels of exposure to a variable of interest and are followed over time to determine the incidence of various outcomes. Two categories of longitudinal studies—quasi-experimental designs and natural experiments—deserve specific mention. Quasi-experimental designs are those in which the exposure is assigned but not according to a randomized experimental protocol. Investigators lack full control over the dose, timing, or allocation of subjects but conduct the study as if it were an experiment (Cook and Campbell 1979; Last et al. 2001). Natural experiments are situations in which differing groups in a population have differing exposures and can be observed for different outcomes (Last et al. 2001). Neither type of design is really an experiment because researchers have not randomly assigned the individuals to exposure groups. The terminology, while not strictly accurate, does call attention to the fact that human groups normally have different exposures and that these naturally occurring events can provide useful information.
An example of a natural experiment is discussed by Boarnet (2004). The California Safe Routes to School Program (see Box 4-1 in Chapter 4) awarded construction funds to numerous communities to improve the safety and viability of walking and cycling to school. A large number of projects (186) that involved changing the built environment were funded within a period of a few years (two annual award cycles). This created the opportunity for a natural-experimental research design. The results are reported in Chapter 4, but in his commissioned paper, Boarnet (2004) makes several suggestions concerning research design that are relevant to the present discussion. First, several projects should be studied because single projects may encounter practical difficulties (e.g., construction delays). In addition, studying an array of projects improves the
ability to generalize the results. Second, before-and-after studies must have baseline data. Ideally, these data should be collected before the intervention occurs. An alternative, second-best approach is to ask subjects retrospectively to compare their activity levels before and after the improvement. Finally, natural experiments may involve groups that have different exposures, but in many such studies, this does not occur by design. For example, Boarnet (2004) found that most but not all of the construction projects he reviewed had been located in places where children would come into contact with them on the way to school. That distinction enabled the researchers to develop ad hoc intervention and nonintervention groups on the basis of whether the children would pass the project on their usual route to school.
Case-Control Studies In case-control studies, exposure to an acknowledged risk factor is compared between individuals from the same population with and without a condition. As opposed to longitudinal studies, in which participants are enrolled and grouped according to exposure status, in case-control studies participants are grouped according to their outcome status. This could mean, for example, sorting individuals on the basis of their activity level (e.g., active versus sedentary) into case and control groups to see whether there are statistically significant differences in environmental characteristics that may influence the propensity of the two groups to be physically active.
Cross-Sectional Studies Also called prevalence studies, cross-sectional studies examine the relationship between conditions (e.g., physical activity behaviors) and other variables of interest in a defined population at a single point in time. For example, the physical activity behavior of matched pairs of individuals and communities could be compared at a particular point in time. Thus the walking and cycling behavior of individuals in a more pedestrian-oriented neighborhood could be compared with that of individuals in a typical suburban planned unit development. The communities could be matched by income, location, accessibility to transportation
services, and topography to isolate the characteristics associated with the friendliness of the different communities to walking and cycling and the potential effect of those characteristics on these activities. Such cross-sectional studies can quantify the presence and magnitude of associations between variables. Unlike longitudinal studies, however, they cannot be used to determine the temporal relationship between variables, and evidence of cause and effect cannot be assumed. As discussed in the following chapter, most studies of the built environment and physical activity have been cross-sectional.
Other Research Design Issues
Level of Aggregation and Geographic Scale
Aggregate data are rarely helpful in illuminating causal links. Because physical activity manifests itself at the individual level, one could argue that the individual is the proper unit of analysis. As shown in the committee’s detailed conceptual model (Figure 1-2 in Chapter 1), physical activity is undertaken at many geographic scales—in and around the home, at work, in facilities such as schools and recreation centers, in the neighborhood, and in the region. This adds a layer of complexity.
As hypothesized in Figure 1-2, at each geographic scale different features of the built environment may have different effects on the individual’s propensity to be physically active. Short distances providing easy access to multiple destinations and a pleasant and safe environment may be important facilitators of physical activity in the neighborhood. By comparison, the size and distribution of activities in a metropolitan area and the availability of transportation alternatives to the automobile (e.g., transit) may dominate the extent to which one chooses a physically active mode of transportation for regional travel, such as commuting or traveling to a shopping center.
Furthermore, the effect of the built environment is likely to differ by type of physical activity. Safety and access to parks and other recreational facilities, for example, may be important in encouraging leisure-time physical activity outside the home. On the other hand, time and distance are likely to be more important factors in
the decision to use nonmotorized transport for destination-oriented travel and may also affect destination choice.
Because most physical activity is spatially constrained and bounded by peoples’ time budgets and physical limitations, smaller geographic units of analysis (e.g., neighborhoods, areas around work sites) are likely to yield more information on the attributes of the built environment that influence physical activity. In general, as discussed in more detail in the next chapter, issues of geographic scale have been underexamined in recent studies linking physical activity behavior to the built environment (Boarnet 2004).
A basic research challenge is distinguishing the role of personal attitudes, preferences, and motivations and of external influences in observed behavior. For example, do people walk more in a particular neighborhood because of pleasant tree-lined sidewalks, or do they live in a neighborhood with pleasant tree-lined sidewalks because they like to walk? This “self-selection” problem potentially confounds the ability of researchers to distinguish how much walking and cycling in an activity-friendly neighborhood is associated with the built environment and how much reflects the attitudes and lifestyle preferences of those who choose to live there. In his paper, Boarnet (2004, 4) raises this point specifically:
Persons might choose their environments in part based on their desired level of physical activity. It does not take much imagination to believe that an avid surfer would choose to live near the beach or that a ski enthusiast would move near the mountains. Generalizing to other, more common forms of physical activity, do persons who wish to walk choose residences in pedestrian-oriented neighborhoods near parks? If so, the association between physical activity and urban form might represent persons’ residential location choices rather than an influence of the built environment on activity.
If researchers do not properly account for the choice of neighborhood, their empirical results will be biased in the sense that features of the built environment may appear to influence activity
more than they in fact do. (Indeed, this single potential source of statistical bias casts doubt on the majority of studies on the topic to date; see Chapter 6.) Various researchers have tried to control for the possibility of self-selection bias in a number of ways. Boarnet and Sarmiento (1998) and Boarnet and Crane (2001a; 2001b) used instrumental-variables techniques to control for choice of residential location in studying how neighborhood features shape motorized travel.2 Bagley and Mokhtarian (2002) used structural-equations modeling3 to account simultaneously for multiple directions of causality, such as the influence of attitudes on both travel and residential location, and the influence of residential location on travel behavior once attitudes were controlled for. Cervero and Duncan (2002) examined mode choice among residents of transit-oriented developments by using nested logit techniques. In their analysis, mode choice was expressed hierarchically as a function of residential location, which in turn was expressed as a function of workplace location.
Another strategy for coping with self-selection bias is to observe when a person moves and to draw associations between changes in the built environment near that person’s new residence vis-à-vis the old and changes in physical activity levels. Krizek (2003) employed such an approach in studying influences of urban design on travel behavior. Research in progress by Handy and Mokhtarian is also focusing on the travel behavior of recent movers to a variety of types of neighborhoods compared with that of a similar group of non-movers in the same neighborhoods. However, moving is often associated with other life changes—marital status, job change, family size, and age of children—that can confound the effect of the new environment on changes in physical activity levels (Boarnet 2004).
Another approach is to focus on children under the assumption that children do not choose their residential location. An important consideration, however, is that parents may impart attitudes about physical activity to their children. For example, if parents prefer and choose to live in an activity-friendly location, correlations between the child’s level of physical activity and the built environment may not demonstrate an independent causal effect, but rather reflect the parental attitudes that have been transmitted to the child (Boarnet 2004). Natural experiments, discussed earlier in this chapter, are another way of circumventing potential self-selection bias. The change in the built environment in such a study, however, cannot be so large as to induce residential relocation, thus confounding the independent effect of the change in the built environment on physical activity levels (Boarnet 2004).
Finally, Schwanen and Mokhtarian approached this issue in a series of studies by comparing the travel behavior of “matched” (or “consonant”) residents of urban and suburban neighborhoods (that is, those who are living in the type of neighborhood they prefer) with that of “mismatched” or “dissonant” residents. They examined the question of whether the travel behavior of mismatched individuals is more like that of the matched residents of the neighborhood in which they actually live or that of the matched residents of the kind of neighborhood in which they would prefer to live. The former outcome would suggest that the effects of the built environment outweigh personal predispositions, while the latter would suggest the converse. Schwanen and Mokhtarian (2003) compared non-commute-trip frequencies of matched and mismatched urban and suburban residents, Schwanen and Mokhtarian (2004a) compared the commute mode choice of consonant and dissonant workers, and Schwanen and Mokhtarian (2004b) completed the picture by examining the role of dissonance in mode-specific distances traveled for all purposes.
Efforts to address the issue of self-selection specifically have only recently been incorporated into research on the influences of the built environment on physical activity. Thus, much remains to be learned about the issue’s relative importance. (Economics and po-
litical science have a considerable literature on residential self-selection that could be drawn upon.4) Knowledge gained through a combination of analytical methods, whether econometric tools or natural experiments, should shed considerable light on this question. Until a body of evidence takes form and the importance of this issue is better understood, the ability to link features of the built environment to physical activity levels will necessarily remain limited.
Two other important hallmarks of good research design are internal and external validity. Internal validity is the degree to which the research design accurately and faithfully reflects the conceptual model that guides the empirical study. Most important, all necessary control variables are used to remove confounding influences and reduce the chances of spurious inferences. Data limitations rarely allow this to be done for a topic such as built environments and physical activity—an issue addressed in the following section. The validity of the data, which is also a concern, is discussed below.
External validity speaks to the generalizability of the research. Data drawn from a single case (e.g., one city or a particular neighborhood) have often been used in past empirical research on the link between the built environment and physical activity levels. As a substantial body of research drawn from many cities and settings accumulates, the external validity of research in this area should improve.
AVAILABILITY OF DATA
Lack of data is one of the main barriers to further progress in examining the causal links between the built environment and physical activity levels. Just as the development of an appropriate theoretical framework will require the joint involvement of the public health and transportation communities, so, too, will the development of appropriately linked data sets. The first grants from the Robert
Wood Johnson Foundation’s Active Living Research Program were earmarked entirely for the development of reliable measures of both the built environment and physical activity, an illustration of the importance of data and measurement to research.5
Many measurement issues affect the ability of a researcher to measure the link between the built environment and physical activity (see also Chapter 2). The first such issue relates to the trade-off between the precision and accuracy of the data and their breadth and accessibility. For example, physical activity can be measured directly as energy expenditure by methods that gauge metabolic energy rates. Pedometers, which count steps and measure distance, and accelerometers, which measure the intensity of an activity, provide objective but somewhat less precise measures of physical activity than those obtained with metabolic methods.6 Indirect measures of physical activity rely on self-reports from surveys or diaries. Of course, direct laboratory measures are the most precise, but they are also the most costly, inconvenient, and artificial (i.e., the results may not be representative of real-life contexts). Use of pedometers and accelerometers, particularly to supplement and corroborate survey data, is less demanding but risks a possible “Hawthorne effect”: respondents must wear the devices and may change their activity patterns because they know they are being monitored.7 Surveys are the most efficient way of collecting data
needed from large numbers of respondents in a range of environments if the objective is to examine the effect of the built environment on physical activity and health outcomes. However, the data collected from surveys and diaries are self-reported, so accuracy and bias are concerns. One commonly used technique to address the accuracy issue is to calibrate survey-collected measures of physical activity against laboratory measures (Boarnet 2004).
A second measurement issue relates to the need for objective as well as subjective measures. This distinction is more important for measures of the built environment than for measures of physical activity.8 Geographic information systems (GIS) now widely available can provide objective measures of many features of the built environment, such as street connectivity and the presence and location of sidewalks, parks, open spaces, and schools (Boarnet 2004).9 As discussed in the preceding chapter, however, individuals’ perceptions of their environment are also important to their choices about being physically active. Thus data on subjective factors, such as individuals’ perceptions of neighborhood safety and the quality of amenities that encourage them to walk and cycle, are an important complement to objective measures. Research is under way to develop more standardized protocols for measuring the perceived qualities of the built environment (see Winston et al. 2004, for example).
A third measurement issue relates to the scale of the data, in particular, the need for fine-grained data on features of the built environment because much physical activity is undertaken near one’s home and workplace. GIS measures of the built environment have become common in studies of land use and travel behavior. They yield geographic-linked data on population and employment densities, mix of commercial and residential land uses, and characteristics of street networks (e.g., street grids, four-way intersections) (Boarnet 2004). However, just because objective data exist in a GIS for an area does not mean that the data are on appropriate variables
or at scales that are useful for analyzing the relationship between the built environment and physical activity.
Two other measurement issues relate to the reliability and validity of the data. Reliability refers to the likelihood that a data measure or survey instrument will provide the same result when it is used by a different researcher or in a different test. For example, interrater reliability is frequently measured when the research involves environmental audits—direct observation of the built environment that is accomplished by walking neighborhoods and recording information about selected environmental characteristics. (See Box 5-1 for an example of an environmental audit instrument.) Va-
The Systematic Pedestrian and Cycling Environmental Scan Instrument
The Systematic Pedestrian and Cycling Environmental Scan Instrument, or SPACES, is one of the first environmental audit instruments developed to measure features of the built environment associated with physical activity (Pikora et al. 2002). To use the SPACES audit tool, observers walk through neighborhoods answering questions that prompt them to record information about street width, sidewalks, traffic volume, lighting, aesthetics, parks and shops, and various other factors that might be linked to physical activity. Information is recorded for individual blocks and thus can be aggregated to higher geographic levels or analyzed at the block level. The SPACES audit tool has been reliability tested, and Pikora et al. (2002) report that many of the questions have high interrater reliability.
A similar audit tool, also applied at the block level, was developed to measure the built environment near school sites in the evaluation of the California Safe Routes to School Program (see Box 4-1 in Chapter 4) (Boarnet et al. 2003).
SOURCE: Boarnet 2004.
lidity refers to an assessment of whether the data collected are accurate relative to some objective standard or measure. For example, as noted above, accelerometers are considered an objective measure of the intensity of physical activity, but their accuracy falls short when validated with portable metabolic units; accelerometers may underestimate energy expenditure by one-third to two-thirds of the more objective metabolic measurement (Welk 2002 in Boarnet 2004). Likewise, the validity of self-reported survey or diary data on travel and other forms of physical activity is problematic because subject recall may be faulty or biased or both. The Strategies for Metro Atlanta’s Regional Transportation and Air Quality (SMARTRAQ) project is an example of an approach that supplements self-reported data on walking and other nonmotorized transport collected from travel diaries with the use of more objective Global Positioning System (GPS)10 data and accelerometers as a check on the location and intensity of the physical activity (see Box 5-2).
The above measurement issues affect the quality of the data available to study the link between the built environment and physical activity. Another, more fundamental challenge is to link disparate databases and data on the built environment and physical activity for research purposes. Currently, these data are spread across a variety of data sources from different fields that have often been developed to address different questions (Boarnet 2004).
Data on Physical Activity
Despite the limitations noted above and in Chapter 2, surveys are the most promising sources of data for studying the links among the built environment, physical activity, and health (Boarnet 2004). The principal public health surveys, identified and discussed in Chapter 2, offer national-level data on a range of physical activity that are tracked over time and are readily available to researchers.11
Strategies for Metro Atlanta’s Regional Transportation and Air Quality
SMARTRAQ was an attempt to link travel diary surveys with information on physical activity. Typically, travel diaries collect self-reported data on walking and cycling but have little objective data on these or other types of physical activity. Nor is specific information generally available about the location where the activity occurred.
As part of a comprehensive travel diary study, SMARTRAQ equipped 500 respondents with GPS transponders and accelerometers. The GPS units were shoulder-mounted systems that tracked walking and other nonmotorized travel. Accelerometers provided measures of activity that did not rely on self-reports. GPS provided information on location that allowed a detailed linking to the built environment, while the accelerometer gave information about the intensity of the physical activity. These two data sources offered the potential to yield information about the link between physical activity and the built environment while also providing a prototype for future studies. More information about SMARTRAQ is presented in the appendix of the paper by Boarnet (2004).
SOURCE: Boarnet 2004.
One limitation of these databases is incomplete coverage of physical activity. As noted in Chapter 2, the major focus of public health surveys has been on leisure-time physical activity. Only recently have data on utilitarian travel begun to be included, but these data are not recorded separately, and physical activity at the workplace is not reported at all. Another key gap, which reflects the early stage of interest in and research on the links among physical activity, health, and the built environment, is the lack of any reported geographic or environmental data that would enable researchers to link survey information on physical activity
levels with details about the respondent’s location and physical environment.
These limitations can be addressed. More comprehensive measures of physical activity are being collected in many public health surveys (see, for example, the discussion of the Behavioral Risk Factor Surveillance System in Chapter 2). Geocoding of physical activity and health survey data using GIS is probably the most efficient way to provide the needed links to data on the built environment (Boarnet 2004). Of course, issues of subject anonymity and confidentiality must be addressed, but there are precedents for doing so.
Data on the Built Environment and Travel
Data on the built environment are not as well developed as data on physical activity and health. Standardized data sets on the built environment are rare, even at the metropolitan area level (Boarnet 2004) (see Box 5-3 for two examples of local land use databases). Typically, researchers must construct such data sets by using available GIS data supplemented with observational environmental audits when necessary and feasible.
Testing of the accuracy of measures of the built environment is at a preliminary stage relative to measures of physical activity. In particular, as discussed in the next chapter, additional research is needed to determine which elements of the built environment are most useful for studying the environmental determinants of physical activity (Boarnet 2004). Complicating this task is the need for fine-grained measures of environmental features (e.g., size and orientation of parking lots, availability and condition of sidewalks), as well as of related mediating variables that may affect individual decisions to be physically active, such as local crime rates and the amount and speed of traffic (Boarnet 2004).
Data on land use and travel behavior are available but typically have not focused on the full range of physical activities and offer limited geographic detail. The National Household Transportation Survey—the primary source of data on nationwide travel behavior—covers commuting as well as nonwork travel, including
Examples of GIS-Based Land Use and Built Environment Databases
One of the better examples of GIS-based land use databases is the Regional Land Information System (RLIS) for metropolitan Portland, Oregon. The Portland Metro has been developing this sophisticated GIS land use database for more than 10 years. The RLIS database includes GIS-based data on sidewalks, bicycle routes, rivers, paths, vegetation cover, slopes, parks, and open spaces, linked to both street and census geography. This is an advanced set of geographic data that enables researchers to use measures of the built and natural environments without having to develop those measures on their own. (See Bolen 2002 for more detail on RLIS.)
Montgomery County, Maryland, is also a leader in making advanced GIS-based land use data available to the research community. The county has developed a website that provides land use information and, in many cases, GIS-compatible land use data. These data include parcel-level information on land uses, aerial photographs that can illustrate detailed historical land use patterns, street maps, bicycle paths, parks, ball fields, watersheds and other natural resources, open spaces, job access, and school boundaries.
The experience in Portland and Montgomery County suggests that evolving best practices for metropolitan land use data will include the following: parcel-level land use and zoning data supplemented with aerial photographs and remote sensing data; land use information for individual parcels that allows the calculation of land use mix; street networks; sidewalk coverage; bicycle paths; parks and other recreation areas; natural resources such as waterways, lakes, and open spaces; accessibility measures that include access to jobs and shopping; school boundaries; crime rates; street lighting; and street tree coverage and other features that provide shelter from the elements. As communities digitize existing databases, such detailed information will become increasingly common. Best practices in GIS-based land use and built environment databases will include user-friendly website access and download capability, data that are compatible with common GIS programs, and historical land use and built environment data that enable changes to be tracked.
SOURCE: Boarnet 2004.
walking and cycling. As discussed in Chapter 2, the 2001 survey attempted to get a better handle on walking, the most common form of physical activity. Plans are to release block-level and census tract data to researchers in 2004, pending completion of confidentiality agreements and the availability of funding to support the data release (Boarnet 2004). Thus for the first time in the United States, researchers will be able to study the relationship between travel and the built environment at the neighborhood level.
At the local level, many metropolitan planning organizations (MPOs) conduct periodic travel diary surveys to provide input into the development and updating of regional travel forecasting models. Typically, the larger MPOs have geocoded the data so a researcher can link them to data on the built environment. Geographically based information about the survey respondent—either residential location or the locations of trip origins and destinations—can be linked to census data on socioeconomic characteristics or to other information about the built environment.12 Use of travel diaries is limited by the fact that most diary surveys collect self-reported data on walking and cycling but few data on other types of physical activity (e.g., gardening, housework, stair climbing) (Boarnet 2004). The SMARTRAQ project, described in Box 5-2, is a promising exception.
Linking Data on the Built Environment and Physical Activity
Modifying existing national survey data on physical activity and health so they can be linked geographically to measures of the built environment is the most immediate improvement likely to provide researchers with the necessary data to better understand potential causal links between physical activity and urban form (Boarnet 2004). Refining measures of both physical activity and the built environment is another important step (Boarnet 2004). Measures of the former are more advanced, but completeness is an issue if total physical activity levels are to be studied. Greater attention should
be focused on capturing physical activity for utilitarian purposes as well as for recreation and exercise, and both types of data should encompass activities in and around the home and workplace, which currently are understudied areas. Devising appropriate and valid measures of the built environment—in particular, developing a better understanding of which features are likely to influence physical activity levels—is a greater challenge. Fine-grained measures of features of the built environment that support physical activity—pedestrian and bicycle paths, public spaces, street lighting at both the neighborhood and workplace levels—may not be available in GIS maps in some localities and may require additional data collection (Boarnet 2004).13 Once more standardized measures are developed, it should be easier to test hypotheses about the relative effect of various characteristics of the built environment on physical activity levels in a range of settings.
NRC National Research Council
RWJF Robert Wood Johnson Foundation
Bagley, M. N., and P. L. Mokhtarian. 2002. The Impact of Neighborhood Type on Travel Behavior: A Structural Equations Modeling Approach. Annals of Regional Science, Vol. 36, No. 2, pp. 279–297.
Bandura, A. 1986. Social Foundations of Thought and Action: A Social Cognitive Theory. Prentice-Hall, Inc., Englewood Cliffs, N.J.
Boarnet, M. G. 2004. The Built Environment and Physical Activity: Empirical Methods and Data Resources. Prepared for the Committee on Physical Activity, Health, Transportation, and Land Use, July 18.
Boarnet, M. G., and R. Crane. 2001a. The Influence of Urban Form on Travel: Specification and Estimation Issues. Transportation Research A, Vol. 35, No. 9, pp. 823–845.
Boarnet, M. G., and R. Crane. 2001b. Travel by Design: The Influence of Urban Form on Travel. Oxford University Press, Inc., New York.
Boarnet, M. G., K. Day, C. Anderson, T. McMillan, and M. Alfonzo. 2003. Safe Routes to School, Vols. 1 and 2. California Department of Transportation, Sacramento, Dec.
Boarnet, M. G., and S. Sarmiento. 1998. Can Land Use Policy Really Affect Travel Behavior? A Study of the Link Between Non-Work Travel and Land Use Characteristics. Urban Studies, Vol. 35, No. 7, pp. 1155–1169.
Bolen, R. 2002. GIS: Essential for Urban Growth Management: Portland, Oregon Metropolitan Area. Presented at RLIS@Ten Symposium, Portland, Ore., March. www.metro-region.org/library_docs/maps_data/gis_and_planning.pdf. Accessed Nov. 18, 2003.
Caporaso, J. A. 1973. Quasi-Experimental Approaches to Social Science: Perspectives and Problems. In Quasi-Experimental Approaches: Testing Theory and Evaluating Policy (J. A. Caporaso and L. L. Roos, Jr., eds.), Northwestern University Press, Evanston, Ill., Chapter 1.
Cervero, R., and M. Duncan. 2002. Residential Self Selection and Rail Commuting: A Nested Logit Analysis. University of California Working Paper. www.uctc.net/papers/604.pdf.
Cook, T. D., and O. T. Campbell. 1979. Quasi-Experimentation. Rand McNally, Chicago.
Crane, R. 1996a. Cars and Drivers in the New Suburbs. Journal of the American Planning Association, Vol. 62, pp. 51–65.
Crane, R. 1996b. On Form Versus Function: Will the New Urbanism Reduce Traffic, or Increase It? Journal of Planning Education and Research, Vol. 15, pp. 117–126.
Duany, A., E. Plater-Zyberk, and J. Speck. 2000. Suburban Nation: The Rise of Sprawl and the Decline of the American Dream. North Point Press, N.J.
Greenwald, M. J., and M. G. Boarnet. 2001. Built Environment as Determinant of Walking Behavior: Analyzing Nonwork Pedestrian Travel in Portland, Oregon. In Transportation Research Record: Journal of the Transportation Research Board, No. 1780, TRB, National Research Council, Washington, D.C., pp. 33–42.
Handy, S. 2004. Critical Assessment of the Literature on the Relationships Among Transportation, Land Use, and Physical Activity. Prepared for the Committee on Physical Activity, Health, Transportation, and Land Use, July.
Handy, S., and K. Clifton. 2001. Local Shopping as a Strategy for Reducing Automobile Travel. Transportation, Vol. 28, No. 4, pp. 317–346.
Humpel, N., N. Owen, and E. Leslie. 2002. Environmental Factors Associated with Adults’ Participation in Physical Activity. American Journal of Preventive Medicine, Vol. 22, No. 3, pp. 188–199.
King, A. C., D. Stokols, E. Talen, G. S. Brassington, and R. Killingsworth. 2002. Theoretical Approaches to the Promotion of Physical Activity: Forging a Transdisciplinary Paradigm. American Journal of Preventive Medicine, Vol. 23, No. 2S, pp. 15–25.
Krizek, K. 2003. Residential Relocation and Changes in Urban Travel: Does Neighborhood-Scale Urban Form Matter? Journal of the American Planning Association, Vol. 69, No. 3, pp. 265–281.
Last, J. M., R. A. Spasoff, S. S. Harris, and M. C. Thuriaux. 2001. A Dictionary of Epidemiology, 4th ed. Oxford University Press.
Louviere, J. J., D. A. Hensher, and J. D. Swait. 2000. Stated Choice Methods: Analysis and Application. Cambridge University Press, Cambridge, United Kingdom.
McFadden, D. L. 1974. The Measurement of Urban Travel Demand. Journal of Public Economics, Vol. 3, pp. 303–328.
NRC. 1999. Governance and Opportunity in Metropolitan America (A. Altshuler, W. Morrill, H. Wolman, and F. Mitchell, eds.), National Academy Press, Washington, D.C.
Pikora, T. J., F. C. L. Bull, K. Jamrozik, M. Knuiman, B. Giles-Corti, and R. J. Donovan. 2002. Developing a Reliable Audit Instrument to Measure the Physical Environment for Physical Activity. American Journal of Preventive Medicine, Vol. 23, No. 3, pp. 187–194.
Rothman, K. J., and S. Greenland. 1998. Modern Epidemiology. Lippincott-Raven, Philadelphia, Pa.
RWJF. 2002. Active Living Policy and Environmental Studies Program Special Solicitation. Princeton, N.J., May.
Sallis, J. F., and N. Owen. 2002. Ecological Models of Health Behavior. In Health Behavior and Health Education: Theory, Research, and Practice (K. Glanz, B. K. Rimer, and F. M. Lewis, eds.), Jossey-Bass, San Francisco, Calif.
Schwanen, T., and P. L. Mokhtarian. 2003. Does Dissonance Between Desired and Current Neighborhood Type Affect Individual Travel Behavior? An Empirical Assessment from the San Francisco Bay Area. Proceedings of the European Transport Conference, Strasbourg, France, Oct. 8–10. www.its.ucdavis.edu/publications/2003/RP-03-18.pdf.
Schwanen, T., and P. L. Mokhtarian. 2004a. What Affects Commute Mode Choice: Neighborhood Physical Structure or Preferences Toward Neighborhoods? Journal of Transport Geography, forthcoming.
Schwanen, T., and P. L. Mokhtarian. 2004b. What If You Live in the Wrong Neighborhood? The Impact of Residential Neighborhood Type Dissonance on Distance Traveled. Submitted for publication.
Tiebout, C. 1956. A Pure Theory of Local Expenditures. Journal of Political Economy, Vol. 64, Oct., pp. 416–424.
Welk, G. J. 2002. Use of Accelerometry-Based Activity Monitors to Assess Physical Activity. In Physical Activity Assessments for Health-Related Research (G. J. Welk, ed.), Human Kinetics, Champaign, Ill.
Winston, E., R. Ewing, and S. Handy. 2004. The Built Environment and Active Travel: Developing Measures of Perceptual Urban Design Qualities. Presented at 83rd Annual Meeting of the Transportation Research Board, Washington, D.C.
Current State of Knowledge
A review of the empirical evidence on the relationship between the built environment and physical activity levels indicates this to be a relatively new field of inquiry. The work conducted to date has embodied two strands of research—one from urban planning and travel behavior and the other from public health and physical activity. Addressing the topic from a broad range of perspectives, areas of expertise, and measures of the variables of interest has stimulated the contributions of a wide range of researchers. In the absence of a common conceptual framework, a more standardized vocabulary, and better linked data sets, however (see Chapter 5), the majority of studies remain at the correlates stage.
The literature provides a growing body of evidence that shows an association between the built environment and physical activity that bears further investigation. However, it is difficult to sort out which characteristics of the built environment have the strongest association. Nor does the literature illuminate the strength of the associations or the populations affected. (For example, an environmental attribute may be strongly associated with higher levels of physical activity but affect only a small subpopulation; conversely, the environmental attribute may have a small association but affect a large population.) More important, as of this writing, the evidence falls short of establishing causal connections.
Nevertheless, the literature provides preliminary evidence that some characteristics of the built environment may affect physical activity levels, or at least certain types of physical activity (e.g., destination-oriented travel or recreational physical activity).
These characteristics include certain land use measures (e.g., density, diversity of uses), accessibility, certain design features, and certain aspects of the transportation infrastructure (sidewalks in particular). Feeling safe and secure from crime and traffic, although obviously not a physical attribute of the built environment, was found to be closely linked to the decision to be physically active for many population groups—women, including minorities; children; and older adults—and thus warrants further investigation. Personal attitudes, motivation, and social support systems were also found to be critical for physical activity and, in the limited number of studies that included these variables, more important than the physical environment as motivating influences. Thus, the evidence to date suggests that a supportive built environment alone is not sufficient to influence physical activity but plays a facilitating role.
The very limited evidence from the handful of studies that addressed causal connections between the built environment and physical activity suggests a complex relationship. When individual attitudes and residential location preferences are taken into account, the autonomous effects of the built environment (e.g., walkability) on physical activity behavior are often exhibited, but much less strongly and in a more nuanced way. Research that attempts to test causal connections in a wide range of settings is important in advancing the understanding of these effects.