To date reviews of studies examining the association between the built environment and physical activity have not summarized findings based on study control of neighborhood self-selection [8, 10, 24, 25]. This review extends prior knowledge by summarizing findings from studies that attempt to address neighborhood self-selection and temporal precedence when examining the association between the objectively-assessed built environment and physical activity. We found that the associations between specific built environmental attributes and physical activity were generally mixed based on this more rigorous evidence. Land use mix, composite walkability indices and neighborhood type were nevertheless consistently associated with higher physical activity levels even after controlling for neighborhood self-selection. Moreover, the built environment was found to be more supportive of walking and cycling compared with physical activity more generally-congruent with previous findings [10, 24, 25] and with current knowledge regarding behavior-specific environmental settings .
Our finding that associations between the built environment and physical activity exist from both cross-sectional studies that adjust for neighborhood self-selection and from quasi-experiments is promising. To date few cross-sectional studies have adjusted for neighborhood self-selection when examining the relation between the built environment and physical activity. This is reflected in the low number of included cross-sectional studies in this review versus elsewhere [8, 10, 24, 25]. Cross-sectional findings that do not account for residential self-selection provide no indication of the extent to which changes in the built environment might be independently associated with changes in physical activity. They may instead just reflect an active individual's preference to live in neighbourhoods that support their existing physical activity levels. In practice, the built environment likely has some effect on the amount of physical activity engaged in by active people who desire to reside in more walkable neighborhoods and the balance between neighborhood supply and demand has the potential to influence population-levels of physical activity.
The purpose of this study was not to describe how neighborhood self-selection and its potential interaction with the built environment influences physical activity but rather to isolate the influence that changes in the built environment might have on changes in physical activity. A unique finding of our review is that the association between the built environment and physical activity likely exists independent of residential location choices. The importance of adjusting for neighborhood self-selection in cross-sectional studies is indicated by reviewed studies that showed an attenuation in the association between the built environment and physical activity after adjustment [47, 51, 55]. Future cross-sectional studies should therefore try to account statistically for neighborhood self-selection in order to obtain more robust estimates of the likely magnitude of any associations between changes in the built environment and physical activity.
Some support for the idea that changes in the built environment precede changes in physical activity is found in the quasi-experimental studies reviewed here that show increased cyclist activity after the installment of cycling infrastructure, increased walking, cycling, and other physical activity after installment of a greenway trail , increased pedestrian activity following installment of street lighting and increased first time park user following park upgrades  (Table 2). These results suggest that modifying one or a few environmental attributes independent of other factors has the potential to encourage more physical activity. However, null and counter-intuitive associations were also found, for example, the reduction in physical activity following the installation of a trail [40, 41]. The overall pattern of findings from quasi-experimental evidence also suggests an increased number of null associations from studies with stronger designs (Table 2). Quasi-experiments offer more robust evidence of causality compared with cross-sectional designs however, they are still subject to biases that might contribute to counter-intuitive findings. The use of quasi-experimental studies in the field is in its infancy. Future quasi-experiments of built environment interventions should take steps to rule out competing explanations and biases such as capturing pre and post intervention data from cohorts, taking multiple pre and post intervention measurements, including multiple intervention-matched control groups and obtaining measures of individual-level dose or exposure to the intervention . Capturing information about intermediate variables (e.g., psychosocial factors, knowledge, awareness) in addition to physical activity might also provide a better understanding about the pathways by which the built environment influences physical activity or clues when interventions provide unexpected results.
To date few potential built environmental interventions and their influence on physical activity have been evaluated and published in the peer-reviewed literature. In addition, environmental attributes that are modified are often those that are only modestly or inconsistently associated with physical activity [8, 10, 24, 25]. There is a dearth of studies examining changes in physical activity among the same respondents in the same neighborhood following changes in pedestrian connectivity, population density, or land uses - which are consistent correlates of walking. This suggests quasi-experiments examining changes in behavior pre and post installment or modification of a wider range of environmental features is needed. For this to occur researchers need to know when changes to the built environment are scheduled to take place to design the study methodology and to obtain resources. Changes to funding opportunities (i.e., rapid reviews for natural experiment opportunities) and building stronger partnerships with urban designers, city planners, and community groups might alleviate some of the barriers to conducting quasi-experiments.
Despite providing preliminary evidence for a causal association between the built environment and physical activity several limitations should be considered when interpreting the findings of our review. The inclusion of only peer-reviewed studies added to the scientific rigour of our review however this approach also means that our findings likely are affected by publication bias thus overestimating the potential influence of environmental attributes on physical activity. The focus on only peer-reviewed evidence might have resulted in the inclusion of fewer quasi-experiments as many urban planning departments likely do not publish evaluations of their own interventions within this forum. We did not exclude or weight study findings based on their individual rigor. This is important to note given that many studies did not provide estimates of the reliability of their physical activity or residential self-selection items. Given the small number of studies included in this review we did not differentiate among quasi-experimental approaches nor method of neighborhood self-selection adjustment. A discussion of the different methods of accounting statistically for neighborhood self-selection, including strengths and limitations is presented elsewhere . We grouped environmental attributes in categories that have face validity, but we acknowledge that some attributes could be located under multiple categories that may not be mutually exclusive.
Data for some environment-physical activity associations were not available from the studies reviewed (e.g., aesthetics and cycling) or if available were examined in only a few studies. Between studies, there was also wide variation in the methods used to measure physical activity and the built environment. Therefore making definitive statements about which environmental attributes should be targeted to increase specific types of physical activities is difficult from our findings. Nevertheless, in support of findings elsewhere [10, 24, 25], our results suggest that some specific environmental attributes might be associated with certain physical activities. For example, connectivity, land use mix, and traffic-related factors are associated with walking for transport but not recreational walking, and population density is associated with walking but not cycling for any purpose or non-specific moderate-to-vigorous intensity physical activity. Consistency in the measurement of built environment attributes and physical activity across studies would improve the comparability of findings and result in stronger statements about the relationship between the built environment and physical activity. This might also allow quantitative synthesis techniques such as meta-analysis to be used.
Evidence regarding the association between the built environment and physical activity found from this review is mixed. Generally, the associations between attributes of the built environment and physical activity from cross-sectional studies appear to remain following statistical adjustment of self-selection although associations attenuate somewhat. However, more rigorous quasi-experimental studies of the relationship between the built environment and physical activity provide less support, with several positive, null, and even counter-intuitive negative associations being found. The mixed nature of findings from the few quasi-experimental studies that have been undertaken on this topic to date, suggest that more quasi-experimental research is needed in order to provide stronger evidence for recommending the creation of walkable neighborhoods as an effective population health intervention for increasing physical activity.
Creating or modifying neighborhoods to make them walkable may not always immediately lead to more physical activity, but could result in other health benefits such as higher social capital [62, 63], improved mental health , and fewer motor vehicle caused pedestrian injuries . However, there is dearth of data on the economic cost of improving neighborhood walkability (or health supportiveness) and the potential subsequent health care savings resulting from improvements in health. Future studies should examine the potential short and long-term effects of built environment interventions on behavior as well the cost-effectiveness of creating walkable neighborhoods. The results of this review reflect previous evidence regarding the importance of street and pedestrian connectivity, land use and destination mix, population density, and overall neighborhood design for supporting physical activity among adults, and adds to the evidence by showing that these associations tend to remain even after taking steps to control for neighborhood self-selection.