Abstract
A growing literature has indicated a relationship between social capital and certain aspects of the built environment with contributions from various disciplines, including environmental psychology, urban design, and health geography. In this systematic review and research synthesis, we summarize the literature in this domain using existing sociological and design frameworks to ascertain the effect of specific built environment domains on social capital. Our review shows that there is a significant relationship between social capital and the built environment, specifically between social cohesion and access to destinations/walkability. Positive relationships exist between social capital, design, and diversity, whereas the effect of population density on social capital is negative and unclear. We find significant methodological limitations and gaps in the published literature, including the absence of longitudinal studies and the use of a plethora of social capital and built environment measures.
Introduction
Over the last decade, evidence has been accumulating on the relationship between social capital and health (Kawachi & Berkman, 2001, 2014; Kawachi, Kennedy, & Glass, 1999). Social capital has been conceptualized as the features of social organization—such as trust between citizens, norms of reciprocity, and group membership—that facilitate collective action (Kawachi et al., 1999; Putnam, 2001).
A number of individual and ecological factors have been associated with social capital of which one of the least researched is the built environment (Kawachi et al., 1999). For example, urban sprawl, a feature of the built environment of many postwar cities in the developed world, is implicated for observed decreases in social capital (Wood et al., 2008). To counter this trend, urban designers have adopted the “New Urbanist” paradigm that seeks to enhance social capital through the creation of pedestrian friendly, walkable neighborhoods with easy access to parks, public transport, and retail outlets, which may also require a high density of dwellings. While sprawl, car dependency, destination unavailability, and inaccessibility decrease avenues of informal social interaction and discourage formal interaction, New Urbanist neighborhoods seek to maximize opportunities for social interaction through better design. This could manifest through multiple pathways such as more time for socialization from decreased commuting time, greater opportunities for social interaction in walkable densely settled communities, a greater feeling of belonging in green walkable suburbs, and ease of access to avenues and centers of formal interaction such as clubs and recreational places. Nevertheless, the evidence in support of decreased social capital from increased urban sprawl has been conflicting (Putnam, 2001) with some researchers questioning the assumptions and the effectiveness of the New Urbanist paradigm (Nguyen, 2010). In addition to neighborhood design and features of the neighborhood that encourage the creation of social capital such as parks and paths, the presence of libraries, schools, or community centers could enhance social capital in the neighborhood through increased interactions and as venues for collective activities (Leyden & Goldberg, 2015; Liu & Bearman, 2012). Conversely, the presence of liquor stores and neighborhood disorder could have an opposite effect (Keizer, Lindenberg, & Steg, 2008).
While a number of researchers have investigated the effect of specific aspects of the built environment on specific domains of social capital (as discussed later in this article), uncovering a variety of relationships in the process, there has been no attempt to synthesize and summarize this accumulation of evidence in a coherent manner. A previous commentary (not a systematic review) by Wood and Giles-Corti (2008) discussed some of the environmental drivers of social capital. This important commentary illustrated the role of three intertwined domains: neighborhood contextual trends, neighborhood design, and neighborhood attributes in creating and supporting neighborhood social capital. Thus, for example, the researchers discuss the role of neighborhood crime and stability in supporting or influencing neighborhood social capital. While social environmental drivers such as crime play an important role in the formation and maintenance of neighborhood social capital, our article, in the form of a systematic review, focuses specifically on the built environment’s role in neighborhood social capital. We address the specific question “How are aspects of the built environment and neighborhood design related to social capital?”
A systematic review of the built environment drivers of social capital could be useful in multiple ways. First, given the strong relationship between social capital and health, it could help public health researchers and planners understand the ways in which the physical environment helps drive health behaviors and consequent health, psychological and social outcomes. Second, it could be useful for planners in designing and redesigning/retrofitting (at various scales) neighborhoods to enhance social capital. Finally, it could help researchers in identifying features of the built environment that could be used as predictors of area-level social capital. A secondary goal of this review is to summarize the quality of the extant built environment and social capital literature.
We thus conducted a systematic review of the available literature on social capital and built environment. Our objectives were to
summarize the specific aspects of the built environment that relate to social capital at various scales and
clarify the relationships between these aspects of the built environment and the various components of social capital.
Method
A protocol for the systematic review was developed and registered with the International Prospective Register of Systematic Reviews (PROSPERO) at http://www.crd.york.ac.uk/prospero/, registration number CRD42015019233. We searched PubMed, PsycINFO, CINAHL Plus with Full Text (using the EBSCOHost integrated database), and ISI Knowledge (Figure 1) for published peer-reviewed research articles. Our choice of databases was motivated by existing systematic reviews in the built environment area (Dunton, Kaplan, Wolch, Jerrett, & Reynolds, 2009; Renalds, Smith, & Hale, 2010). We also scanned the reference lists of articles that passed the eligibility test for further articles. We did not set a particular cutoff date for the earliest articles, and all databases were scanned through to November 21, 2015. Our search strings were ((Residence Characteristics [MeSH Terms]) OR built environment [Title/Abstract]) AND X [Title/Abstract]. The social-capital-related terms that were inserted into X, one at a time, were sense of community, community attachment, social ties, social interaction, volunteering, collective action, collective efficacy, social capital, community participation, trust, social support, place attachment, and knowledge of neighbors. The social-capital-related terms were decided by the authors based on their knowledge of the social capital literature.
Summary of the literature review process.
We did not impose any restrictions on our searches other than articles be published in English. Note that the Medical Subject Headings (MeSH) list of terms does not have a significant representation of built environment terms. For example, the word built environment is not a MeSH term.
Inclusion and Exclusion Criteria
We included studies that investigated features of the built environment and their relationship with social capital. We included all articles that measured social capital, collective efficacy, acts of neighboring, and various other components of neighborhood social capital. Articles that evaluated either or both of the behavioral/structural components or the cognitive/perceptual components of social capital were included. Studies that measured social capital at neighborhood and up to county scales were reviewed. The dynamics of social capital generation beyond these scale boundaries are likely to be very different. Thus, we excluded studies that investigated the effects of specific design features within a building or building complex (such as the presence of vegetation within a building complex common space) on social capital. In addition, we excluded articles that investigated the quality or aesthetics of various features of the built environment. If studies included the effect of built environment quality in addition to other metrics such as design, the quality results were ignored, but the study was included. We excluded these because quality is often difficult to quantify and/or generalize across multiple studies. Studies measuring social capital in specific groups such as the disabled or the aged were excluded as these are, once again, difficult to generalize. Review or discussion articles were not considered in detail other than as potential sources of relevant publications. Materials not published in peer-reviewed outlets were excluded, as were books and book chapters.
We included studies that either objectively evaluated the built environment or asked survey respondents about their perceived built environment. Thus, studies that measured walkability using Geographical Information Systems (GIS) as well as studies where survey responders subjectively ascertained the walkability of neighborhoods were included. In addition, we included studies, especially from the urban design domain, that compared neighborhoods based on their own perceptions of the built environment.
Data Analysis
Quality assessment
We assessed if the articles had clear definitions of social capital, appropriately defined study objectives, and used appropriate methods. The studies in our review were either quantitative or mixed method. Table 1 summarizes the analyzed studies in chronological order. We have attempted to abide by Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines for systematic reviews. A number of the assessed papers, in addition to evaluating relationships between built environment and social capital, reported other relationships; for example, between built environment and health outcomes. When this occurred, we extracted only social-capital-related results and ignored the other aspects of the study.
Summary of Included Studies in Order of Publication Year.
Note. GIS = Geographical Information Systems.
Next, we implemented a quality assessment to score each paper on metrics of both general quality and potential for bias. First, we used the National Institutes of Health, Heart Lung and Blood Institute (NIH-NHLBI; 2014) systematic review quality assessment tool for Observational Cohort and Cross-Sectional Studies to assess the quality of each study (Table A1 in the appendix). The tool provides a means to assess the internal validity and risk of bias for each study. It consists of 14 questions with yes/no answers. Detailed guidance on answering these questions is available on the tool website (NIH-NHLBI, 2014). We modified this tool to accommodate a gradient of numeric responses ranging from 0 as “definitely no” to 1 as “definitely yes” (Table A2 in the appendix). We also removed Questions 5, 6, 10, 12, and 13, which relate to cohort studies and statistical power (or the probability of false null hypothesis rejection, which is usually not provided in cross-sectional studies). This exercise was implemented by two of the authors who scored the papers independently and then compared scores, resolving differences through discussion and consensus. Finally, Spearman’s rank correlation coefficient and Cohen’s kappa were calculated to evaluate the level of score agreement between the two reviewers.
We mapped each question in the NIH-NHLBI quality assessment questionnaire to each bias domain in the Cochrane Risk of Bias Assessment Tool: for Non-Randomized Studies of Interventions (ACROBAT-NRSI) bias assessment tool (Table A3 in the appendix). We are cognizant that these tools are generally intended for longitudinal studies. Given the cross-sectional nature of the assessed papers, we believe that this analysis exposes the shortcomings of the current built environment and social capital literature. The final scores were an average of the scores of the two reviewers published as a percentage fraction of the maximum possible score. We calculated average scores for each question in the NIH-NHLBI tool and each item in the ACROBAT-NRSI tool (scores summed and averaged across papers), and also for each paper (scores summed and averaged across questions and items). This provides a profile of the “quality” gaps in the reported literature.
Data condensation
The articles appraised were heterogeneous with different conceptualizations of social capital and the built environment. Studies measured different components of the built environment and various domains of social capital. We therefore condensed both the built environment and the social capital domains into specific theory-driven subdomains and subsequently into an easily interpretable matrix. We used a simple “vote” method of data condensation to summarize both significant and nonsignificant relationships (Cooper, 2010). Note that while we use the term relationships, as this is what the majority of researchers investigate, some researchers have actually investigated associations. When a researcher used different regression models, we reported results from the most complex model or the model with the greatest number of independent variables. Thus, multivariate models were preferred to univariate models. Occasionally a simpler model included a built environment parameter that the complex model did not include, in which case we reported the results from the simpler model.
Social capital measures
To summarize the social capital domain, we followed the theoretical framework provided by Carpiano (2006), which is based on the work of the French sociologist Bourdieu. This framework is relevant to the built environment and health domain (Carpiano, 2006; Wood & Giles-Corti, 2008). Using Carpiano’s framework, we mapped the different terms used to describe the different components of social capital in the various papers into three specific items: social capital, social cohesion, and neighborhood attachment. In this framework, social cohesion is the antecedent of social capital, or it represents the “networks and values from which social capital can be developed and used for action” (Carpiano, 2006, p. 641). Thus, network ties, familiarity, trust, and informal social interactions come under the purview of social cohesion. The Carpiano framework delegates the action-oriented proactive forms of social capital under the label “social capital.” Thus, participating in neighborhood organizations or helping neighbors (offering “social support”) is social capital. The other item, “neighborhood attachment,” is generated through social cohesion.
We also incorporated collective efficacy as a fourth item. Collective efficacy was proposed by Sampson as a means of measuring action-oriented social capital, especially as it relates to the social control of crime and disorder (Sampson, McAdam, MacIndoe, & Weffer-Elizondo, 2005; Sampson, Raudenbush, & Earls, 1997). Collective efficacy is defined as “social cohesion among neighbors combined with the willingness to intervene on the behalf of common good” (Cohen, Inagami, & Finch, 2008, p. 199). In contrast to other measures of social capital, it is not dependent on specific neighborhood networks and the individual’s relationships with these networks (Cohen et al., 2008). While group participation is also included under the social capital domain of Carpiano, we left the collective efficacy/collective action item as separate when specifically defined as such by researchers. Thus, if a researcher investigated “community participation” we categorized this under the umbrella of social cohesion but if collective efficacy was specifically mentioned we left it as such. Some studies measured social capital or social cohesion as a single metric, and did not measure its components. These studies were mapped directly to the social capital item or social cohesion item of Carpiano’s framework. We use the term overall social capital to indicate all domains of social capital (social cohesion, social capital, etc.) together.
Built environment measures
We similarly summarized built environment metrics following the 3D framework suggested by Cervero and Kockelman (1997). The 3Ds consist of Density, Diversity, and Design. While originally developed to predict travel demand, the framework has been adopted by built environment and health researchers in a variety of studies (Giles-Corti, Hooper, Foster, Koohsari, & Francis, 2014; Handy, Boarnet, Ewing, & Killingsworth, 2002; Lee, 2007). However, we added an additional dimension—“Destination”—following the 6D extension by Giles-Corti et al. (2014) of the 3D framework. We were specifically interested in destination because of its strong correlation with walkability, which is in turn reported to enhance social capital (Duncan, Aldstadt, Whalen, Melly, & Gortmaker, 2011; Leyden, 2003). The other two domains in 6D—distance to transit and demand management—were considered unlikely to have been assessed for their effect on social capital in the literature. The destination domain measures the proximity to a range of destinations inclusive of retail, sports, recreation, health clinics, hospitals, fast food, and alcohol outlets, and studies investigating walkability were mapped onto the destination domain. In Kockelman’s paper, the density domain consisted of residential and employment density (Cervero & Kockelman, 1997), though some researchers used population density, a closely related metric. The diversity domain measures diversity of land use through the use of land for residential, commercial, recreational purposes in a shared nonsegregated manner. In this framework, parks and green space also belong to the diversity domain. The design domain captures street network patterns, street widths, speed limits, and the prevalence of cul-de-sacs. In this regard, urban planners refer to grid patterned Manhattan style street networks as “Traditional,” while street networks in modern U.S. suburbs with multiple loops and cul-de-sacs are referred to as “Conventional.” In this article, we expand the design domain to indicate New Urbanist design of neighborhoods as this paradigm seeks to incorporate many of the positive features of the design domain. For example, a central theme of New Urbanist designs is the development of networked, accessible street-centered neighborhoods in addition to mixed housing types, rear parking, pedestrian, and cycling provisions.
Data Synthesis
Raw score table
For each research study, we generated a score table (three columns, many rows). Each row in the raw score table consisted of three records. One record described the social capital component measured in a specific research study such as “trust in neighbors,” whereas the other record described the built environment component found to be significantly associated with the social capital component such as “walkability.” If the relationship was in the expected direction, for example, walkability is found to enhance trust in neighbors or the presence of alcohol outlets reduces social cohesion, the relationship was scored +1, otherwise a −1 score was given. This score occupied the third column. If multiple relationships were investigated in a study, they populated additional rows in the raw score table. Finally, all nonsignificant relationships were also tabulated similarly in a separate table. It is important to note that this score was based on what we considered the expected direction of the relationship to be, which did not always coincide with expectation of the authors of the papers reviewed. Nevertheless, the authors’ expectations (when mentioned) and our expectations were generally the same with one exception (Hipp, Corcoran, Wickes, & Li, 2014).
Condensed score table
We created a second score table. For each row in the raw score table, this table had a corresponding row. This table incorporated the summary built or social capital measure corresponding to each component of social capital or built environment into the first table. Thus, continuing with the above example, “trust in neighbors” is mapped to “social cohesion,” while walkability is replaced with “destination.” The score occupied the third record. This condensed score table could then be easily collapsed into a matrix with summary social capital measures as rows and summary built environment measures as columns, and the sum of scores representing the overall strength of the relationship. If during the condensation process the same relationship was repeated in a study, for example, social cohesion-design +1, we repeated this relationship in our table. Thus for a given study, a given combination of a specific condensed social capital domain and a specific condensed built environment domain can appear more than once. The raw and condensed score tables are published together in our article as one table. Nonsignificant relationships were similarly tabulated to be used in significance testing (described next). While nonsignificant relationships are included in the synthesis, for the sake of brevity, they are described in lesser detail than the significant results. Also, the combined raw and condensed score table is not published for nonsignificant results but available from the first author on request. The summary matrices that are generated from this table are referred to as the Social Capital-Built Environment (SC-BE) matrix henceforward.
Significance test
To test if the condensed relationships between the social capital and built environment domains were significant, we implemented a significance test for vote counts (Cooper, 2010) as follows:
where all relationships regardless of significance are included, and
Results
Overview
Figure 1 summarizes the data flow for this review following, as far as feasible, PRISMA guidelines. Studies are summarized in Table 1. Of the 2,042 studies found in our initial search, 2,003 were excluded after duplicate removal and initial screening. Generally, the studies removed at the screening stage were unrelated to either built environment and/or social capital. Thirty-nine studies were assessed in detail for eligibility, and 16 excluded with reason (reasons summarized in Figure 1). Thus, we finally reviewed and synthesized a total of 23 studies with a combined total of 90,002 participants from seven countries. Two studies (Rogers, Gardner, & Carlson, 2013; Rogers, Halstead, Gardner, & Carlson, 2011) reported the same data, and we reported them together where practicable. The majority of studies (13) were conducted in the United States; five were from Australia, and one each from Norway, Ireland, the Netherlands, the United Kingdom, and Japan. The earliest study was conducted in 1997 and the latest in 2014. The articles were published in 15 different journals with the largest representation (5) in Health & Place. The majority of studies used a cross-sectional survey to address their research questions. Studies had sample sizes ranging from 48 to 22,191. Most studies reported a participant response rate, with around half reporting a response rate of 50% or above.
The combined raw and condensed score table for significant relationships is Table 2. Methods ranged from simple tabulations without any statistical testing to complex multilevel models (Tables 1 and 2). One paper utilized percent enumerations of social capital measures by neighborhoods for comparison (Podobnik, 2011). Another used simple correlations between the number of library visits and social capital measures (Johnson, 2010). Yet, another researcher used t tests and Wilcoxon rank sum tests to compare social capital measures between neighborhoods with varying walkability (Rogers et al., 2013; Rogers et al., 2011). Multivariate regressions with built environment variables as the independent variables and social capital measures as the dependent variables were common, with some researchers using a continuous dependent variable (Wood et al., 2008) and others using binary dependent variables with logit or probit models (Hanibuchi et al., 2012; Leyden & Goldberg, 2015). The most common means of modeling the effect of built environment on social capital was multilevel or hierarchical models. A common conceptual model was that of social capital at the neighborhood level with various covariates at the individual and neighborhood levels (Cohen et al., 2008; Du Toit, Cerin, Leslie, & Owen, 2007; Hipp et al., 2014; Nguyen, 2010; Skjaeveland & Garling, 1997). One paper used a multilevel model not for conceptual ease but to account for clustering in the data (Maas, van Dillen, Verheij, & Groenewegen, 2009). Of all studies, one did not report significance testing (Podobnik, 2011), while another found no social-capital-related significant results (Nasar, 2003). We were unsure about the directionality of the findings in a third study (5-15 min subjective distance to destination related to less sense of community), and did not include it in our matrices (Francis, Giles-Corti, Wood, & Knuiman, 2012).
Condensation of Significant Relationships: Combined Raw and Condensed Score Table.
Note. NR = not reported; NA = not available.
1 indicates that the relationship is in the expected direction, while −1 indicates a counterintuitive relationship.
Beta: Slope coefficient estimate in a regression.
SE: Standard error, standardized betas are beta coefficients that have been standardized, such that the variances of the dependent and the independent variables are equal to 1, partial beta coefficients provide the amount by which the dependent variable increases when a specific independent variable increases by one unit, holding other independent variables constant.
MANCOVA: Multivariate analysis of covariance, GLM: Generalized linear model.
95% confidence interval.
Rogers (2011) is based on the same data as Rogers (2013) and are presented together.
Overall, we found 22 different metrics of social capital in the reviewed papers (Table 2). Three papers used a direct and specific measure of social capital (Johnson, 2010; Nguyen, 2010; Wood et al., 2008). A further group of papers measured collective efficacy and collective action (Cohen et al., 2008; Du Toit et al., 2007; Theall et al., 2009). Sense of community was utilized by four researchers (Du Toit et al., 2007; French et al., 2014; Podobnik, 2011; Wood, Frank, & Giles-Corti, 2010).
Assessment of Quality and Bias
Generally, research papers stated their research objectives, specified and selected an appropriate study population, defined exposures clearly, measured a range of exposures, and adjusted for confounders, though some research papers had a response rate below 50%. The median quality score on our version of the NIH-NHLBI scale for research papers was 67%. The highest scoring paper (79%) was well designed, had a high survey response rate, and adapted its measures of exposure and outcome from standardized questionnaires. On average, papers performed moderately on our version of the Cochrane Bias Assessment Scale, with a median score of 94.0% for detection, 73.0% for selection, 67.0% for performance, and 94.0% for reporting (tables not shown). The overall median score for individual papers was 75.0%. However, note that these scores were calculated after removing items pertaining to longitudinal studies, and thus biases introduced by the cross-sectional nature of these studies are not taken into consideration.
There was moderate agreement between the two reviewers regarding the quality of the research papers. Median Cohen’s kappa and median Spearman’s rank correlation coefficient was .49 for agreement of scores of individual papers. Similarly, median Cohen’s kappa and Spearman’s rank correlation coefficient was .61 for each question in the NIH-NHLBI tool across different papers. These tables are not shown here but are available from the authors on request.
Condensation and Significance Testing Results
A total of 66 significant and 69 nonsignificant relationships were tabulated (Tables 3 and 4). While Table 3 shows all relationships, Table 4 summarizes our findings in terms of directionality of the findings and significance. Of significant relationships, 27 involved destinations, 12 density, 13 diversity, and 14 design. The vast majority of significant relationships mapped to either the social capital or social cohesion with one each mapping to neighborhood attachment and collective efficacy. Overall, social capital (all domains together) was found to be significantly associated with the built environment (Tables 3 and 4) with the overall total of 84 positive relationships and 51 counterintuitive relationships, or relationships that were not in the expected direction. The small number of relationships between built environment and collective efficacy/neighborhood attachment investigated made inferences on these domains difficult. Overall, social capital was significantly related to the design domain. A significant relationship also existed between access to destinations, social cohesion, and overall social capital. In addition, social cohesion was related to built environment in general (all built environment domains together). We discuss each domain in detail and specifically report significant relationships from each domain next.
Matrix With Total Counts of All Significant and Nonsignificant Relationships and Significance of Condensed Results.
Note. Each cell presents the total number of relationships inclusive of nonsignificant relationships. The numbers in brackets are as follows: (−number of counterintuitive relationships, +number of positive relationships). The numbers in brackets also include negative relationships. Bold lettering implies q* < .00227, where q* is the false discovery rate adjusted p value or the significance of the condensed relationships between the rows and columns. Thus, for example, social cohesion is significantly related to destinations.
Final SC-BE Matrix of Relationships.
Note. SC-BE = social capital-built environment.
↓ = negative nonsignificant relationship.
↑ = positive nonsignificant relationship.
▲ = positive significant relationship.
Destinations
Various researchers report an increase in social capital from better destination access (Leyden & Goldberg, 2015; Rogers et al., 2013; Rogers et al., 2011), including increased knowing of neighbors, social engagement (Leyden & Goldberg, 2015; Rogers et al., 2013; Rogers et al., 2011), and an increased sense of community (Du Toit et al., 2007). Ten studies reported significant relationships between destinations and social capital. Of the 10 studies, four investigated the relationship between walkability and social capital, while the remaining investigated the association of local destinations such as libraries and interesting sites with social capital. Studies used both objectively determined measures of walkability and destinations, and subjective measures. Access to destinations/walkability was found to be significantly associated with social cohesion. In addition, there was an overall significant relationship between social capital (all domains) and access to destinations (Table 4). Approximately 15% of significant relationships and 30% of all relationships between access to destinations and overall social capital are not in the expected direction. It is instructive to note, however, that almost half of the contribution to the significant relationships came from three papers with a very similar design, two of which shared the same dataset (Leyden, 2003; Rogers et al., 2013; Rogers et al., 2011).
Density
Four studies demonstrated significant relationships between density and social capital (Tables 1 and 2). We mapped three different density-related measures into the density domain. These included dwelling density (French et al., 2014; Skjaeveland & Garling, 1997), population density (Brueckner & Largey, 2008), and urbanization (Hanibuchi et al., 2012). The summary SC-BE matrix (Tables 3 and 4) shows that the overall relationship between social capital and density is negative, though not a strong, consistent relationship. Approximately 75% of all significant and nonsignificant relationships were in the negative direction (Tables 3 and 4).
Of the significant relationships, one researcher (Hanibuchi et al., 2012) found conflicting relationships between urbanization and social capital, while French (French et al., 2014) reported a decreased sense of community with increasing residential density. In addition, one Norwegian study at the Local Neighborhood scale found dwelling density to be negatively related to supportive acts of neighboring (Skjaeveland & Garling, 1997), whereas another study from Utah found fewer interactions in high-density census tracts (Brueckner & Largey, 2008; Skjaeveland & Garling, 1997). The Utah study also observed no effect of density on group involvement. Note that the Norwegian and Utah studies, while being at very different scales, found similar, negative results.
Diversity
The diversity domain is used to indicate land use mix, including the presence of green space. One researcher found a direct negative relationship between land use mix and social capital (Wood et al., 2010). Nevertheless, an overall positive relationship between diversity and the various social capital domains is indicated in the SC-BE matrix (Tables 3 and 4). Around 60% of all relationships (Table 3) between increased diversity and overall social capital are in the expected direction.
Of the significant relationships, the majority (3) of studies investigated relationships between green space access and social capital. A fourth study reported the effect of environmental features, which the researchers conceptualized as “barriers to socialization,” such as parks, industrial areas, rivers, and highways (Hipp et al., 2014). While in this study parks were not considered as facilitators but barriers to socialization, we coded parks, following the rest of the literature, as facilitators. Increased distance to parks was found to have a positive effect on social capital in this study, which we considered a counterintuitive finding.
Specifically, there was a positive effect of access to green space on social capital. Lund (2003) found that access to parks increased unplanned interactions and had a positive effect on local social ties, though the overall explanatory power of physical environment attributes was small. Another study from Los Angeles reported a positive association between neighborhood parks and neighborhood collective efficacy (Cohen et al., 2008). In contrast, a study from the Netherlands reported that people with more green space around them did not have significantly more supportive interactions but were less likely to experience a shortage of social support when there was more green space within 1 km around them (Maas et al., 2009).
Design
Gridded or traditional street networked New Urbanist neighborhoods were expected to generate more social capital than areas with conventional suburban streets. A handful of papers that investigated the design domain were from the urban design discipline (Brown & Cropper, 2001; Kim & Kaplan, 2004; Nasar, 2003; Podobnik, 2011). These papers generally contrasted social capital in New Urbanist neighborhoods with social capital in conventional neighborhoods, with the key distinguishing feature being neighborhood design.
While two papers reported that New Urbanist neighborhoods demonstrated a greater sense of community (Kim & Kaplan, 2004; Podobnik, 2011), one showed no relationship (Brown & Cropper, 2001). Greater group participation in New Urbanist neighborhoods was described by one study (Podobnik, 2011), and another reported more social contacts (Brown & Cropper, 2001). There was some support that these findings were not the result of self-selection (Kim & Kaplan, 2004), as the choice of living in a New Urbanist neighborhood is not necessarily driven by the desire for better sense of community (Nasar, 2003). In spite of these generally positive findings, other papers have been less optimistic about the New Urbanist approach. Some of the key features of New Urbanism are mixed use development with both retail and residential zoning in addition to close access to public transport. Another feature is the presence of gridded “traditional” street networks as opposed to “conventional” curvy cul-de-sac intensive suburban street networks. One study (Wood et al., 2008) found that people residing in neighborhoods with conventional street networks have higher social capital than residents in neighborhoods with traditional or mixed street networks. Urban sprawl, a by-product of conventional street design in neighborhoods, has been implicated with causing decreased social capital (Putnam, 2001). One study that investigated the direct effect of urban sprawl on social capital reported conflicting findings (Nguyen, 2010). Thus, support for Putnam’s hypothesis of decreased social capital with increased sprawl comes from the finding that less sprawl is related with political participation, political group membership, and interest in national affairs at the county level of geography (Nguyen, 2010). However, this study also reported more social interaction, participation in religious groups, giving and volunteering in sprawled areas (Nguyen, 2010). Overall, the weight of evidence supported a significant positive relationship between design and overall social capital with 75% of all relationships in the expected directions (Tables 3 and 4). Of the significant relationships, nine were in the expected direction, while five were counterintuitive.
Discussion
A diverse range of studies have attempted to investigate links between social capital and aspects of the built environment. This systematic review reveals a significant positive relationship between social cohesion and destination accessibility/walkability. It also reveals a significant relationship between social cohesion and the overall built environment (all domains of built environment considered together). In addition, access to destinations and design is related to overall social capital (all domains of social capital considered together). Finally when considered together, there is a significant relationship between overall social capital and the built environment (all domains together).
Existing conceptual frameworks hypothesize relationships between the built environment and social capital through two separate pathways, which some researchers (Kim & Kaplan, 2004) term as “activity-based” and “meaning-based.” Discussing the activity-based pathway, Wood et al. (2010) quoted another author: “The trust of a city street formed over time from many, many little public sidewalk contacts . . .” Thus, chance and formal encounters facilitated by the built environment can enhance social capital. The meaning-based pathway emphasizes the role of perceptions in the creation of belonging, attachment, and social capital. Thus, for example, a sense of safety and low levels of vehicular traffic (Wood et al., 2010) or green space (Maas et al., 2009) can affect perceptions of area friendliness and helpfulness. The overall policy environment can influence the structure of the built environment (Francis et al., 2012) and the existence of destinations that enhance or harm social capital (Theall et al., 2009).
Not all relationships between social capital and built environment were found to be in the expected direction. Indeed, the negative relationship between density and social capital is surprising as, for example, both density and design variables were found to load on to an intensity factor related to walking (Cervero & Kockelman, 1997). It has been argued that intermediate variables such as the homogeneity of inhabitants may play an important role in actually creating social capital (Talen, 1999). Also, social capital is often created from interactions within geographically diverse social networks raising questions about the role of the local neighborhood in the process (Talen, 1999). It is possible that some of the densely settled areas draw a transient population with minimal local ties and who do not have access to a car, who remain in the neighborhood for a short period of time only, and therefore require easy access to various amenities. Wood et al. (2010) discussed this in the context of similar negative findings with mixed land use, calling this the “outsider or stranger hypothesis.” Indeed, a threshold (Wood et al., 2008) could exist for density beyond which gains in social capital from better access to destinations, green space, and so on are offset by the loss of social capital from the presence of large numbers of individuals with minimal local ties. Other high-density areas may be low socioeconomic status (SES) inner city neighborhoods beset with urban decay or incivility. The only study that does find a significant positive effect of density is from Japan, a country known to be relatively ethnically homogeneous (Hanibuchi et al., 2012). It has been reported that ethnic diversity has a negative effect on social capital (Hero, 2007), and the effect of density, demographic diversity and social capital needs to be explored along with a deeper investigation of the density domain. Finally, high-density areas may draw significant foot and vehicular traffic from elsewhere, which may detract individuals from spending time in the street or in their front yard (if present, though unlikely in high-density areas), thereby reducing potential interaction with neighbors (Wood et al., 2010).
While the density domain in the reviewed articles is measured objectively using statistical/GIS measures, the remaining domains could be measured either objectively or subjectively. Thus, for example, while individuals could be asked about the walkability of their neighborhoods, walkability could also be measured by querying the GIS-based Walk Score (Duncan et al., 2011) measure at an individual’s residence. This could account for some of the variations in relationships with social capital in the remaining domains. For example, research has shown that the size of the perceived neighborhood varies with SES (Vallée, Roux, Chaix, Kestens, & Chauvin, 2015), and that, depending on attributes such as SES, individuals may interpret their neighborhood as being less walkable even when it is actually quite walkable. Research on access to destinations or walkability and social capital could benefit from the use of diverse but standardized measures of walkability and social capital, as half of the significant findings reported in this domain came from one specific design (Leyden, 2003).
The studies discussed in this review are cross-sectional studies that encompass a multiplicity of scales from single neighborhood comparisons (Brown & Cropper, 2001; Nasar, 2003; Podobnik, 2011; Skjaeveland & Garling, 1997) to county-level comparisons (Brueckner & Largey, 2008; Nguyen, 2010). It is important that studies be implemented at multiple scales as some relationships such as the inverse relationship between density and social capital seem to hold across scales indicating a degree of robustness in this finding. However, this review also highlights the shortcomings of the social capital and built environment literature. First, studies use a variety of conceptualizations and operationalizations of social capital, in addition to a variety of metrics of the built environment. It is important that future studies use existing benchmarked measures of both social capital and built environment. Some examples of existing measures include the Social Capital Benchmark Survey (Harpham, Grant, & Thomas, 2002), the Neighborhood Environment Walkability Scale (NEWS; Cerin et al., 2013), and the methodology used to create county sprawl indices in the United States (Ewing & Hamidi, 2003). Indeed, differences in various measurement schemes could explain some of the variations in relationship direction.
A second shortcoming is the cross-sectional nature of all the studies included in this review. The absence of analyses that incorporate temporal precedence makes it difficult to infer causal relationships. Thus, there remains a need for high-quality longitudinal studies of the effect of built environment on social capital, such that causal relationships can be inferred and interventions tested. Finally, an important predictor of neighborhood social capital—residential tenure—is often not incorporated into many studies, though some studies include this (French et al., 2014; Lund, 2003; Nguyen, 2010; Theall et al., 2009; Wood et al., 2008). One study that did account for the age of the neighborhood (but not the tenure of individuals) found it to be a strong predictor of social capital, more so than resident characteristic adjusted walkability (Hanibuchi et al., 2012). Thus, future research on this topic should incorporate benchmarked and well-accepted measures of social capital and built environment. In addition, longitudinal designs and residential tenure adjustment should be considered. Future researchers should also specifically investigate the role of density in social capital in depth, as discussed earlier.
This review is subject to various limitations. First, it does not include unpublished or non-peer-reviewed articles. Second, while we made extensive efforts to capture the relevant published literature through systematic searching and scanning the reference lists of papers, there is a possibility that some publications may have been overlooked. In this regard, a more comprehensive set of search terms could have captured more papers though it is unlikely that this would significantly change the overall findings of this review. While the quality of the reviewed papers was assessed, the agreement between reviewers on paper quality was moderate. Our summarization method requires the mapping of built environment and social capital measures onto specific domains. In addition, the mappings were based on our expectation of the direction of the relationship, which may not always be accurate. While we attempted to be accurate in creating these mappings, there is subjectivity involved in the process. It could be argued that the process of condensing many diverse variables into specific narrow domains may introduce various biases. Thus, for example, walkability was forced into the destination domain. While this is generally supported by the existing literature (Duncan et al., 2011; Lee & Moudon, 2006), a rigorous meta-analysis identified intersection density as the strongest predictor of walking (Ewing & Cervero, 2010) among many built environment parameters. Thus, using alternative frameworks may have produced somewhat different results. Nevertheless, it is unlikely that the general associations uncovered in our analyses would have been significantly different.
A further limitation is the vote counting procedure and subsequent significance test (Cooper, 2010). Vote counting procedures are known to be conservative and err toward not rejecting the null hypothesis when it is false (Type 2 error; Cooper, 2010). Conversely, as sets of relationships share the same dataset, they are not necessarily independent, resulting in Type 1 errors when distributional assumptions (normal/binomial) are made, though bootstrapped confidence intervals, which are robust to nonindependence, supported our results. Nevertheless, in spite of these limitations and the implementation of appropriate multiple hypothesis testing correction, we found various significant positive relationships between the built environment and social capital, underscoring the strength of the extant relationship. Also, some of the tests are underpowered, because very few relationships were tested, and the inclusion of a larger set of relationships could have generated greater degrees of statistical significance.
Conclusion
To our knowledge, this is the first systematic synthesis of the literature on built environment and social capital. Using a rigorous vote counting method, sound theory-driven sociological and design frameworks complemented by statistical testing and quality assessment, the review synthesized a broad and diverse literature into a small set of take-home results. Specifically, the research synthesis highlighted a significant relationship between access to destinations and social cohesion. In addition, an overall significant relationship between the built environment and social capital was indicated. This review also showed the considerable, though not consistent and strong, negative relationship between increased density and low social capital. While the evidence pointed toward associations, conclusions about causality can be best drawn with studies that establish temporal precedence. No such studies were identified in this review.
Footnotes
Appendix
Mapping of NIH-NHLBI Quality Assessment Tool to Cochrane Risk of Bias Assessment Tool: For Nonrandomized Studies of Interventions.
| Solution | Performance | Detection | Attrition | Reporting | Other | |
|---|---|---|---|---|---|---|
| Question 1 | X | |||||
| Question 2 | X | |||||
| Question 3 | X | |||||
| Question 4 | X | |||||
| Question 5 | X | X (Design) | ||||
| Question 6 | X | |||||
| Question 7 | X | |||||
| Question 8 | X | |||||
| Question 9 | X | |||||
| Question 10 | X | |||||
| Question 11 | X | |||||
| Question 12 | X | |||||
| Question 13 | X | |||||
| Question 14 | X |
Note. Note that Questions 5, 6, 10, 12, and 13 were removed from assessment. Thus, the performance and attrition and design domains could not be measured. As all studies assessed were cross-sectional studies, these domains are not relevant. NIH-NHLBI = National Institutes of Health, Heart Lung and Blood Institute.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.