Objective To synthesise the literature on the effects of neighbourhood environmental change through residential relocation on physical activity, walking and travel behaviour.
Design Systematic review following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines (PROSPERO registration number CRD42017077681).
Data sources Electronic databases for peer-reviewed and grey literature were systematically searched to March 2017, followed by forward and backward citation tracking.
Eligibility criteria A study was eligible for inclusion if it (1) measured changes in neighbourhood built environment attributes as a result of residential relocation (either prospectively or retrospectively); (2) included a measure of physical activity, walking, cycling or travel modal change as an outcome; (3) was quantitative and (4) included an English abstract or summary.
Results A total of 23 studies was included in the review. Among the eight retrospective longitudinal studies, there was good evidence for the relationship between relocation and walking (consistency score (CS)>90%). For the 15 prospective longitudinal studies, the evidence for the effects of environmental change/relocation on physical activity or walking was weak to moderate (CS mostly <45%), even weaker for effects on other outcomes, including physical activity, cycling, public transport use and driving. Results from risk of bias analyses support the robustness of the findings.
Conclusion The results are encouraging for the retrospective longitudinal relocation studies, but weaker evidence exists for the methodologically stronger prospective longitudinal relocation studies. The evidence base is currently limited, and continued longitudinal research should extend the plethora of cross-sectional studies to build higher-quality evidence.
- physical activity
Statistics from Altmetric.com
The health benefits of physical activity are well established.1–4 However, globally, large proportions of the population are not sufficiently active or are completely inactive.5 6 Walking is the most popular kind of physical activity7 8 and typically occurs in neighbourhood environments, which may facilitate or hinder physical activity through their design.9 10 Over the last two decades, there has been an exponential increase in studies which, based on social-ecological models of health,11 have investigated the relationships between built environment attributes and physical activity, particularly walking.12 13 This research has found that physical activity and walking are associated with a range of built environment attributes, such as walkability (street connectivity, land use mix and population density), access to green space and recreational facilities, safety from crime and traffic, aesthetics and access to public transport.14–16 However, despite the substantial policy interest,17–22 nearly all the studies in this field of research are cross-sectional15 16 23 and therefore do not provide causal evidence about the effects of the built environment on physical activity. If cross-sectional studies report an association between environmental attributes, for example, walkability and physical activity, it is not clear to what extent this is due to the effect of the environment or to alternative explanations, such as residual confounding, where people living in high walkable neighbourhoods are different to people residing in low walkable neighbourhoods.
For ethical and practical reasons, randomisation is virtually impossible in research examining the impact of neighbourhood built environments on walking and physical activity.24 Several alternative designs may be considered to extend the current evidence base built primarily on cross-sectional studies. For example, longitudinal analysis of people who remain in their neighbourhoods (eg, examination of environmental predictors of physical activity initiation/maintenance among ‘non-movers’), longitudinal analysis of people who relocate to neighbourhoods with different environmental attributes (ie, relocation studies) and evaluations of environmental interventions are all longitudinal by nature, which allows for establishing the temporal sequence of cause and effect, a key criterion for causation. Further, these study designs are better at accounting for confounding than cross-sectional studies because they provide opportunities for comparing exposures and/or outcomes within an individual, instead of comparing people living in different types of neighbourhoods at one point in time. Still, these alternative designs have their advantages and limitations. For example, opportunistic evaluations of environmental interventions are less subject to self-selection bias (ie, people choose to live in neighbourhoods to accommodate their lifestyle preferences, such as their propensity for active travel25) compared with the other two longitudinal study designs discussed here. However, researchers do not have control over the timing, location and nature of the intervention.26 Neither do they have control over the dose of the intervention. As environmental change is usually slow and incremental,27 it may not provide a sufficient ‘dose’ required for behavioural change during the time frame of the evaluation. In fact, some evaluations of environmental interventions on physical activity had mixed findings14 possibly due to these challenges. In longitudinal studies of non-movers, one may expect little changes in the outcomes because behaviours tend to habituate over time. Relocation studies, on the other hand, follow the concept of ‘mobility biographies’, where stabilised behavioural patterns are ‘interrupted’ by life events, including environmental changes as a result of residential relocation.28 Moreover, because environmental exposures pre-relocation and post-relocation can be quantified, changes in exposures can be evaluated as a ‘natural experiment’, and there have been calls for such designs to evaluate effects of neighbourhood environments on health behaviour and outcomes.24 29 30 However, relocation studies are still subject to confounding, such as reasons and motivations for relocation.
In summary, evaluations of walking, physical activity and travel behaviour before and after people relocate between neighbourhoods that differ in environmental attributes offer a unique opportunity to examine the role of neighbourhood environments, not only within the context of residential relocation and mobility biographies, but also extend the current evidence on built environments and physical activity/travel behaviour in general by addressing some critical methodological limitations of cross-sectional studies. To the best of our knowledge, no other study has systematically reviewed the evidence on the effects of residential relocation on walking, physical activity or travel behaviour. In the present systematic review, we aim to synthesise the current evidence on the association between neighbourhood built environments and walking, physical activity and travel behaviour within the context of residential relocation.
Data sources and searches
The protocol for this systematic review was registered with the International Prospective Register of Systematic Reviews (PROSPERO; registration number CRD42017077681, available at https://www.crd.york.ac.uk/PROSPERO/display_record.php?RecordID=77681). This systematic review follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (online supplementary table 1).31
Supplementary file 1
Systematic searches were conducted from database inception to March 2017, in the electronic databases MEDLINE, The Cochrane Library, EMBASE, CINAHL, SPORTDiscus, PsycINFO, Informit, Avery and RIBA for peer-reviewed papers, and The Grey Literature Report and ProQuest Dissertations and Theses Global for grey literature. Additional articles were identified through backward and forward citation tracking of included publications, and using the authors’ own reference libraries. The list of search terms used in our MEDLINE search, which was adapted for searches in other databases, can be found in online supplementary table 2.
A study was eligible for inclusion if it (1) measured changes in neighbourhood built environment attributes as a result of residential relocation (either prospectively or retrospectively); (2) included a measure of physical activity, walking, cycling or travel modal change as an outcome; (3) was quantitative and (4) included an English abstract or summary.
A study was excluded if it (1) was based on simulation data only32; (2) was conducted in the context of relocation on a university campus or at work33; (3) focused on international migration34; (4) examined social environments only35 or (5) did not clearly define or measure the built environment attributes.36 37 Specifically, exclusion criterion 3 was chosen because individuals and their environments may not be comparable pre-immigration and post-immigration. Exclusion criterion 5 applies to relocation studies where the environment was vaguely defined or not measured (eg, moving to a ‘New Urbanist-inspired’ development,36 or a mixed-use development37) and therefore one cannot determine how built environment attributes changed after relocation.
Following a standard protocol, two authors (BN and DD) independently screened studies for eligibility based on the title, abstract and full text. Uncertainty was discussed involving a third author (KG), and any disagreement was resolved by consensus. A PRISMA flow diagram presents the summary of the study selection process (figure 1).
Information about each paper was extracted by BN and DD independently for quality assurance. Any disagreement was discussed until consensus was reached.
At the study level, the following information was extracted: study name (if any), study design, setting and follow-up, sample recruitment, sample characteristics, neighbourhood environmental attributes (perceived or objectively measured), covariates, whether accounted for residential self-selection, potential moderators/effect modifiers tested and main findings.
At the result level, information about each finding was extracted based on the combination of environmental exposure and walking/physical activity/travel behaviour outcome. For studies that reported both cross-sectional and prospective longitudinal analyses,38–42 we extracted findings from longitudinal and quasi-experimental analyses only because of the inability to ascertain changes in environmental attributes from cross-sectional analyses. Further, in addition to physical activity, walking and cycling outcomes, we also extracted results regarding public transport and car use43 44as these can serve as important secondary outcomes. This is because a modal change from driving to public transport use is relevant to an active lifestyle.43 44 For studies that involve modal change,40 45 it is important to present information on all transport modes to provide a complete picture. Finally, although our search protocol excluded studies that exclusively examined the neighbourhood social environment, we also extracted results regarding perceived safety and sociability40–42 46 47 because both these attributes are closely linked to aspects of the built environment.48 49
General characteristics about each selected study, including country, study name, study design, neighbourhood environment measures (objective and/or perceived), walking/physical activity/travel behaviour measure (objective or self-reported) and whether residential self-selection was accounted for, and if so, how, were summarised and tabulated.
Extracted study results were synthesised in separate tables for retrospective longitudinal and prospective longitudinal studies. Retrospective longitudinal studies (often referred to as quasi-longitudinal studies in the planning and transportation literature) refer to a study design where participants retrospectively report for a defined time in the past (eg, 1 year ago, prior to relocation) and the present to determine the effects of a change in an attribute (eg, neighbourhood walkability) on behaviour (eg, walking).40–42 47 50 Although this type of research design was named differently by some studies, such as a quasi-longitudinal pre–post design, for the purpose of this systematic review, we define them consistently as retrospective longitudinal studies. Longitudinal prospective studies include observational (cohort)38 and natural experimental studies,39 51 where naturally occurring events, such as moving to a new community, are evaluated as defined interventions in a prospective fashion.
For each study, all tested associations (unadjusted and adjusted) involving change in neighbourhood environmental attributes (exposures) and walking/physical activity/travel behaviour (outcomes) were considered. Because of the heterogeneity in the exposure and outcome measures, we could not quantitatively synthesise the effect sizes, instead, we grouped the results into categories and semiquantitatively summarised them based on the direction and significance of the associations. For environmental attributes, we developed a grouping scheme similar to that used in previous literature reviews,52 53 where attributes were allocated to subcategories under ‘recreation environment’, ‘neighbourhood design’, ‘transportation environment’, ‘aesthetics’, ‘crime-related safety’, ‘social environment’ and ‘aggregated characteristics’. For the outcomes, we categorised walking into recreational/leisure, transport and total walking; physical activity into recreational, transport and total physical activity; and other travel behaviour into cycling, public transport use and driving. We did not separate cycling for recreational and transport because only one study included some recreational cycling outcomes.54 We developed a matrix to tabulate the extracted results using ‘+’ to denote statistically significant (P<0.05, unless noted otherwise) associations in the expected direction, ‘–’ to denote significant associations in the unexpected direction and ‘0’ for non-significant associations.52 The expected direction is based on the existing evidence base and the concept that activity friendly neighbourhood environments characterised by mixed land use and well-connected streets with good access to parks and recreation facilities, public and alternative transportation options, and low traffic and crime, are conducive to walking, physical activity and active travel, while discouraging car driving. Specifically, the expected direction for each association is presented in online supplementary table 3.
We allowed each study to contribute more than one finding to each combination of neighbourhood environmental attribute and outcome. When a study included different exposure measures for the same category of environmental attribute, we considered these as distinct findings. For example, Knuiman et al measured land use mix objectively using a Geographic Information System (GIS) within a 1600 m street network from participants’ homes and participants’ perceptions about the number of types of destinations within their neighbourhood.55 In this case, both findings were counted as two separate associations. Similarly, when a study included multiple outcomes for the same environmental attribute that did not overlap, such as cycling for leisure and cycling for transport,54 we considered them as distinct associations.
Given that studies present their results differently (eg, some present only the final models while others present unadjusted and different versions of adjusted models), to ensure that results from one study are not inflated as a result of duplication, we adopted the following protocols for assigning ‘+’, ‘–’ or ‘0’ to each comparison. (1) When different models for the same association (using the same exposure and outcome) were presented, we determined the significance and direction of association based on the model that at least adjusted for demographic characteristics, socioeconomic status and neighbourhood self-selection. Alternatively, if the authors explicitly discussed that one model is less biased than the other, we then coded this association based on the less biased model. For example, Braun et al tested the association between a walkability index and walking outcomes using both random and fixed effects models.27 They argued that estimates from random effects models were more biased because of residual self-selection bias; we therefore coded this association based on results from the fixed effects model. (2) When we could not select the least biased model based on criterion 1, we coded this association based on the consistency of results. For example, if at least 60% of the adjusted results were significant in the expected direction, we coded this result as ‘+’, if the pattern of the results was inconsistent (eg, 50% ‘+’, 50% ‘0’), we coded it as ‘?’ to denote the uncertainty of the association.
Finally, we summarised results regarding each environmental attribute across different outcome measures by calculating a consistency score as the percentage of total associations being significant in the expected direction.52 Two consistency scores were developed. The first one refers to the number of associations coded ‘+’ as a proportion of the total number of associations, which denotes the overall consistency of an environmental attribute with different outcomes at the level of associations (findings). Using the same scoring system, we summarised results about environmental attributes across different outcomes and each outcome across different environmental attributes. Due to the small number of studies overall, particularly regarding domain-specific walking and physical activity, we combined different subcategories of walking and physical activity within the larger categories.
The second score applied weights to associations reported from the same study (in the same or different publication), so that the overall consistency of associations was not driven by single studies. Specifically, we applied a weighting scheme similar to that reported in a systematic review by Cerin et al.16 For example, Handy et al reported associations between three land use mix indicators and overall walking,41 and Cao et al reported two50; given that the two publications were based on the same study, we assigned each of the five findings a weighting of 0.2. Applying weighting, the second summary statistic indicates the overall consistency of an environmental attribute with different outcomes at the study level.
Quality appraisal and risk of bias analysis
We developed a quality appraisal checklist (online supplementary table 4) based on previous systematic reviews16 56 with additional items designed particularly for relocation studies (eg, ‘Did the study assess whether the participants experienced life changing events which may have led them to relocate and did they account for these events?’). Two authors (BN, KG) independently performed quality appraisal, and any disagreement was resolved by consensus.
We conducted the following three risk of bias analyses for the data synthesis by recalculating the consistency scores after (1) excluding all studies with low-quality scores (<5) based on quality appraisal, (2) excluding all studies that did not adjust for self-selection bias (see online supplementary table 5 for details) and (3) limiting to findings involving objectively measured neighbourhood environmental attributes. These risk of bias analyses aim to examine how sensitive study findings are to the quality of the included studies, self-selection bias and the measurement mode of the neighbourhood environment. Previous studies suggest attenuated associations after accounting for self-selection and considerably different levels of consistency in associations by the measurement mode of neighbourhood environmental attributes.57
Selection of studies
The database searches yielded 3324 records (figure 1). After removing duplicates, 2846 records remained. After excluding 2817 records based on reading the titles and abstracts, the full texts of the remaining 29 were examined and an additional 14 full texts were excluded. With an additional 2 studies identified through backward and forward citation tracking, and 6 from the authors’ own reference libraries, a total of 23 publications were appraised and synthesised.
Fifteen of the publications were based on longitudinal prospective studies27 28 38 39 45 46 51 54 55 58–63 and eight based on retrospective longitudinal/quasi-longitudinal studies.40–42 47 50 64–66 Altogether, the publications were based on studies conducted in six countries (table 1), with the USA (n=10) and Australia (n=5) contributing to most of the publications. Five publications were based on the RESIDential Environment Project (RESIDE) in Perth, Australia, and four were based on a study conducted in Northern California, USA. In terms of the measurement of neighbourhood environmental attributes, 11 reported objective measures only, mostly based on a GIS, 3 included perceived measures and 9 included both objective and perceived environmental measures. All but one study39 relied on self-reported measures of walking/physical activity/travel behaviour. More than half of the publications reported some measures of residential preferences to account for self-selection bias. Details about each study, including relevant findings, are presented in online supplementary table 5. Overall, the quality scores varied, ranging from 1 to 7 on a scale from 0 to 9, with 10 of the 23 studies scoring five or more points. Longitudinal prospective studies scored much higher (range: 2–7, mean: 5) than retrospective longitudinal studies (range: 1–3, mean: 2.6). The results of the critical appraisal are presented in online supplementary table 6.
Summary of findings from retrospective longitudinal studies
As shown in table 2, overall transportation access, social environment, crime-related safety and accessibility were among the most assessed environmental attributes and walking was the most commonly used outcome. Overall, there was consistent support for the effects of change in neighbourhood environmental attributes through residential relocation on the change in a range of outcomes, particularly walking, where the consistency scores were >90%. Most environmental attributes yielded a consistency score of ≥50% across outcomes and the score was particularly high for overall transportation access (access to a range of specific or non-specific transportation options, such as sidewalks, bike paths, public transport and roads), aesthetics and crime-related safety. After accounting for multiple findings from the same study, the weighted consistency scores were slightly lower than the unweighted ones. It is important to note that given the small number of retrospective longitudinal studies many of the environmental attribute–outcome combinations were not examined, and most of those that were examined involved a small number of studies.
Summary of findings from longitudinal prospective studies
Given the larger number of longitudinal prospective studies, a broader range of environmental attribute–outcome combinations were explored (table 3). The most examined environmental attributes were land use mix/destinations and public transport access/services and the most commonly used outcomes were transport walking and cycling. Compared with results from retrospective longitudinal studies, those from prospective longitudinal studies were much less consistent. Among all environmental attributes, walkability/pedestrian friendliness had the highest weighted and unweighted consistency scores, although the findings only involved three studies. Most environmental attributes had consistency scores of 25%–40%, providing less consistent evidence for the effects of change in neighbourhood environments through residential relocation on change in walking/physical activity/travel behaviour. A few attributes had a consistency score of 0%, including traffic, aesthetics, neighbourhood type, sprawl, all of which were based on a small number of studies. Across outcomes, associations involving a walking outcome had the highest consistency scores while those involving physical activity and cycling had much lower scores.
Risk of bias analysis
Three risk of bias analyses were conducted separately for retrospective longitudinal and prospective longitudinal studies (online supplementary table 7). First, when excluding studies with a quality score of ≤4, 0 retrospective longitudinal and 10 prospective longitudinal studies remained. The consistency scores were very similar in the risk of bias analyses, and in some cases slightly higher, among the higher-quality studies compared with all studies. Second, when limiting to studies that accounted for self-selection, the consistency scores from retrospective longitudinal studies remained nearly identical while those from longitudinal prospective studies slightly fluctuated, though the overall level of consistency remained similar. Finally, when limiting to findings involving objectively measured neighbourhood environmental attributes, consistency scores remained similar or slightly lower in retrospective longitudinal studies and similar (or in some cases slightly higher) in longitudinal prospective studies. Overall, results from risk of bias analyses showed robustness in our findings, but are somewhat limited by the small numbers of studies/findings after exclusion.
To our knowledge, this is the first systematic review that synthesises the evidence on the effects of change in neighbourhood environments through residential relocation on walking, physical activity and travel behaviour. Given the potential for walking to increase total physical activity levels and health, efforts to implement environment-changing interventions seem logical and may have an impact at the population level. Our review found a scarcity of literature on residential relocation, with only 23 publications from 16 studies in six countries (five high-income countries and one upper-middle-income country) meeting our inclusion criteria. Overall, the studies are heterogeneous in terms of design and measures, making it difficult to draw conclusions about specific associations. Summarised across different exposure and outcome measures, the overwhelming pattern of associations suggests a much stronger evidence for the effects of change in neighbourhood environment through residential relocation in retrospective longitudinal (quasi-longitudinal) than prospective longitudinal studies; and for both study designs, the most consistently significant associations involved walking as an outcome.
The differences in findings between prospective longitudinal and retrospective longitudinal studies highlight the importance of research design. In principle, although prospective longitudinal studies are not perfect, a retrospective longitudinal/quasi-longitudinal design is more subject to bias, with participants more prone to recall and social desirability biases (ie, they report in favour of an improvement).40 42 In cases where individuals were prompted to report the change in both neighbourhood environment and physical activity/travel behaviour,47 50 common source bias may be an additional concern. Previous literature has also documented the ‘honeymoon effect’ where recent movers are likely to rate their new neighbourhood more favourably.67 The common source bias and honeymoon effect combined may particularly bias the associations away from null among those who recently relocated compared with those who relocated further in the past, or the control group who did not relocate. Taking these potential biases into consideration, the high consistency in findings from retrospective longitudinal studies should be interpreted with caution.
The small number of studies on residential relocation is in contrast to the vast and ever-growing body of literature on built environments and physical activity in general.16 68 To contextualise our review, we have summarised all literature reviews on built environments and physical activity among adults that we identified through previous reviews of reviews24 68 and we updated this list through systematically searching literature databases (see table 4 and online supplementary table 8 for unabridged information). Nearly 30 reviews have been published, with some reviews including a large number of empirical studies,15 16 69 indicating the popularity of the field. However, the current evidence base predominantly relies on cross-sectional studies, and some literature reviews have even excluded longitudinal or experimental studies a priori to solely focus on cross-sectional studies.70–74 While cross-sectional studies are important for generating hypotheses at an early stage of scientific field development, and have contributed to understanding the plausibility, consistency and the specificity of the associations between the built environment and physical activity,53 they are inherently subject to ambiguity in temporality, residual confounding and self-selection bias. Given that the ultimate goal of research on built environments and physical activity is to inform urban planning, transportation and public health policy and practice, we must consider evidence that is based on stronger research designs, including, but not limited to, residential relocation studies.
Residential relocation: opportunities and challenges
Residential relocation provides a unique opportunity for improving the evidence base. One of the key limitations of cross-sectional studies is that those living in high and low walkability neighbourhoods may be substantially different (eg, socioeconomic status, propensity to be physically active), which violates the ‘exchangeability’ assumption for causal inference, and statistical methods cannot ensure total control of confounding.75 Longitudinal studies (including residential relocation studies), on the other hand, compare an outcome within the individual. When time-varying variables are accounted for, a participant could serve as her/his own control,60 which better accounts for residual confounding. Furthermore, studies on ‘mobility biographies’ argue that individuals are likely to be ‘open-minded’ to changing habitual travel behaviour and to resynchronise their behaviour with their new environment after relocation.28 64 The implication is that residential relocation not only provides an opportunity for understanding the impacts of neighbourhood environments on behaviour during a period susceptible to behavioural change, but also serves as an ideal window of opportunity for interventions. For example, recently relocated residents should be made aware of the local facilities and opportunities for active living, as previous evidence suggests a mismatch between perceived and objectively measured neighbourhood environment and that perceived environmental attributes may be more strongly associated with physical activity than objectively measured neighbourhood attributes.76
However, as demonstrated by the overall low-quality score in our quality appraisal, relocation studies have methodological challenges. First and foremost, endogeneity of neighbourhood selection biases the estimate of associations between the built environment and physical activity in observational studies. Relocation studies, whether prospective longitudinal or retrospective longitudinal, are still subject to the same self-selection bias where individuals who are predisposed to lifestyle change (eg, those who are environmentally concerned) select their new residential neighbourhood to facilitate the change. Such unmeasured preferences or constraints that impact both neighbourhood selection and physical activity will lead to erroneous associations between neighbourhood environments and physical activity. Compared with cross-sectional studies, longitudinal studies (including relocation studies) provide the opportunity for establishing the temporality of residential preferences, exposures to neighbourhood environments and changes in travel behaviour/physical activity.42 Such study designs paired with appropriate methods for accounting for self-selection bias, as outlined by Cao et al,25 could potentially provide stronger evidence towards causality. Second, a unique challenge to relocation studies is confounding by concurrent life events that cause or accompany relocation and neighbourhood reselection. For example, people relocate in response to other life events, such as changing jobs, employment status or household size. Previous studies found that residential relocation was no longer associated with travel modal change when adjusted for other life events, such as birth of the first child and changing employer.28 77 Therefore, it is important to account for major life events when assessing the association between relocation and walking/physical activity/travel modal change. Of all the studies in this review, less than half (n=9, 39%) explicitly adjusted for life events. Third, previous evaluations of environmental interventions suggest that significant behavioural change may be more likely to occur over a longer follow-up period,14 78 possibly due to a ‘lag time’ to adapt to a new environment. Therefore, residential relocation studies need to be planned with longer-term follow-up in mind. Fourth, residential relocation studies, along with other longitudinal studies, are subject to loss to follow-up. For example, most of the longitudinal studies reviewed had a drop-out rate of >30%. Therefore, appropriate handling of missing data is critical to prospective evaluations of residential relocation studies. Fifth, as researchers cannot influence the relocation process, studies involving residential relocation may encounter practical challenges. For example, some participants may have moved prior to the pre-move data collection leading to a smaller sample size than envisaged and a loss of power,39 or due to unforeseen circumstances, pre-move data had to be collected retrospectively rather than prospectively as initially planned.37 Such unexpected and uncontrollable events challenge the researchers to react promptly and pragmatically with the minimal compromise of research quality. Finally, in relocation studies, behaviour change is catalysed by relocation. It is unknown whether similar changes in environmental attributes will lead to changes in outcomes among non-movers. Hence, it is important to supplement evidence from relocation studies with longitudinal studies of non-movers and evaluations of environmental interventions.
Strengths and limitations
One of the strengths of this systematic review is that it adheres to the PRISMA statement for systematic reviews,31 which is not standard practice in the field of built environments and physical activity/travel behaviour.24 In addition, we developed methodologies to account for multiple publications of the same study along with several risk of bias analyses to determine how sensitive our overall findings are to specific studies, measurements, study design and quality. Our review is limited by the small number of studies, the relatively low quality of most studies, heterogeneous exposure and outcome measures, and not being able to take into account effect sizes in our synthesis. Finally, summarising across diverse environmental attributes and outcomes is methodologically challenging. While synthesising evidence by categorising these measures provides a ‘big picture’ perspective of the evidence, it also inevitably introduces biases in interpretation when lumping measures together.
Overall, we found a paucity of studies on the associations between changes in neighbourhood built environment and walking/physical activity/travel behaviour outcomes in the context of residential relocation. The findings of these studies differ dramatically by study design, with retrospective longitudinal/quasi-longitudinal studies supporting a significant association whereas findings from prospective longitudinal studies were less consistent, but possibly also less biased. Further research should focus more on well-designed ‘natural experiments’. Residential relocation provides a unique opportunity for studying environment-induced changes in physical activity. The literature reviewed here represents steps towards incremental improvement in quality evidence to inform policy and practice regarding urban design and transportation planning. However, the inadequate evidence base limits specific policy recommendations regarding how changes in a particular environmental feature or infrastructure will ‘cause’ health-promoting change in residents’ physical activity and travel behaviour. Continuous improvement of the research evidence is critical to the field. Future studies could benefit from using longitudinal data sources, such as travel panels45 and cohort studies,38 evaluating relocation effects over longer follow-up periods and apply appropriate research designs and statistical approaches to account for self-selection and concurrent life events. Additional data from geographically diverse areas, particularly from low-income and middle-income countries, could also add to the current literature. In summary, this review appraises environmental changes for walking, physical activity and travel behaviour in a methodologically sound manner, aiming to refocus the research agenda of the built environment beyond cross-sectional studies to provide higher-quality evidence.
What is already known?
Attributes of neighbourhood built environments are associated with walking and physical activity based on a large body of literature that mainly consists of cross-sectional studies.
Cross-sectional studies are particularly subject to biases and cannot provide the strongest and most policy-relevant evidence.
In studies of neighbourhood environments, randomisation is virtually impossible, ‘natural experiments’ that evaluate effects of neighbourhood environments on health behaviour and outcomes provide opportunities for high-quality evidence.
A number of studies examined the effects of environmental changes through residential relocation on walking, physical activity and travel behaviour. However, the evidence has not been synthesised or appraised.
What are the new findings?
There is a paucity of relocation studies examining effects of built environments on physical activity/travel behaviour.
The quality of the studies varied, with prospective longitudinal studies rating higher than retrospective longitudinal studies.
There was encouraging evidence for the relationship between residential relocation and walking from retrospective longitudinal studies, but much weaker evidence from prospective longitudinal relocation studies.
Future studies could benefit from using longitudinal data, such as travel panels and cohort studies, evaluating relocation effects over longer follow-up periods and accounting for self-selection and concurrent life events.
Contributors DD and KG conceptualised the study, were involved in research supervision, wrote the first draft, and all the other authors provided critical input in the interpretation of data and writing of the manuscript. BN and DD conducted the literature search. BN, DD, VL and KG extracted data. All authors approved the final version for submission.
Funding The study was funded by a Heart Foundation Cardiovascular Research Network Project grant awarded to Ding et al and
Competing interests DD is supported by an Australian Heart Foundation Future Leader Fellowship. BN is supported through an Australian Postgraduate Award and a University of Sydney Merit Award.
Patient consent Not required.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.