Measuring 'neighborhood': Constructing network neighborhoods

This study attempts to measure neighborhood boundaries in a novel way by creating network neighborhoods based on the density of social ties among adolescents. We create valued matrices based on social ties and physical distance between adolescents in the county. We then perform factor analyses on these valued matrices to detect these network neighborhoods. The resulting network neighborhoods show considerable spatial contiguity. We assess the quality of these aggregations by comparing the degree of agreement among residents assigned to the same network neighborhood when assessing various characteristics of their “neighborhood”, along with traditional definitions of neighborhoods from Census aggregations. Our findings suggest that these network neighborhoods are a valuable approach for “neighborhood” aggregation. © 2011 Elsevier B.V. All rights reserved. Neighborhoods constitute a fundamental unit of interest for many social scientists. Indeed, sociologists since nearly the dawn of the discipline have focused on neighborhoods as genuine phenomena, as exemplified by the Chicago School in the early part of the 20th Century (Park and Burgess, 1921; Shaw and McKay, 1942). Psychologists have focused on the effect of neighborhood contexts on numerous individual-level processes (Bronfenbrenner, 1977). In the latter part of the 20th Century, the advent of a particular statistical technique—multilevel modeling—coincided with an explosion of interest in the effects of neighborhoods on innumerable individual behaviors of adolescents and adults alike. This “neighborhood effects” literature has looked at the effects of various neighborhood characteristics on delinquent behavior (Osgood and Anderson, 2004; Silver and Miller, 2004), educational achievement (Ainsworth, 2002), low birth weight (Morenoff, 2003), and depression (Ross et al., 2000), to name just a few. These studies have used numerous conceptualizations of what constitutes a neighborhood, including such varied ecological units as census-defined blocks, block groups, tracts, postal zip codes, or neighborhoods as defined by the cities or residents themselves. All of these studies face a common challenge: how exactly should we conceptualize “neighborhoods”? A challenge for ecological theories positing such neighborhood effects is that nowhere in this literature is there a very clear This research is supported in part by NSF Grant BCS-0827027. ∗ Corresponding author at: Department of Criminology, Law & Society, University of California, Irvine, 3311 Social Ecology II, Irvine, CA 92697, United States. Tel.: +1 949 824 8247. E-mail address: [email protected] (J.R. Hipp). definition about what we mean when measuring a “neighborhood”. This is not a trivial issue, given that such theories are explicitly ecological, requiring the construction of aggregated measures. Although numerous studies have tested the effects of various neighborhood characteristics on various outcomes, nearly all of these studies are constrained to aggregating the contextual measure of interest to geographic units of analysis that have been designated by the U.S. Census Bureau. There is no reason to assume that these are the only possible neighborhood aggregations that could be constructed. Such arbitrary aggregations can cause the researcher to fail to find an effect that is actually present given a different aggregation of the social context of interest (Hipp, 2007). We propose here one strategy to creating neighborhoods that incorporates information on the social ties within the broader community. Although some adopt a purely geographic conception of the neighborhood, we argue that the presence of social ties is a characteristic of neighborhoods—and are implicit in many existing definitions of neighborhoods—and thus it is reasonable to incorporate the structure of community social ties into a definition of neighborhood boundaries. Although some may believe that our approach conflates the hypothesized positive effects of social ties with the very definition of neighborhood, we point out that our approach: (a) considers only the presence of ties and does not presume they have a pro-social character, and (b) allows for great variation in tie density. We therefore propose creating a valued sociomatrix of the residents within the community in which the valued relations are some combination of the physical distance between the persons and whether or not they are socially tied. Given the novel nature of our procedure, and the fact that there 0378-8733/$ – see front matter © 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.socnet.2011.05.002 Author's personal copy J.R. Hipp et al. / Social Networks 34 (2012) 128– 140 129 are myriad options that could be adopted at several of the decision points in our study, we adopt an exploratory approach to demonstrate the utility of this strategy. In what follows, we first discuss the issues involved in measuring neighborhoods. After discussing why social relations among residents likely play an important role in any definition of neighborhoods, we discuss the several issues that must be addressed when constructing network neighborhoods. We then describe the data and present illustrative examples of our approach, and finally present the results when using various possible configurations of “neighborhoods”. We conclude with a discussion of how our “network neighborhoods” have potential utility for both social network and neighborhood scholars. 1. Conceptual background 1.1. Measuring neighborhoods The numerous studies that have studied ecological processes in neighborhoods, or the effects of neighborhoods on various individual-level outcomes, contain as a theme the acknowledgement that defining “neighborhoods” is a difficult task. Only occasional studies have offered serious treatments on how we define neighborhoods (Hunter, 1974; Schwirian, 1983). One theme in this literature, although often implicit, is that neighborhoods are geographic entities that are essentially always defined as contiguous units. Thus, the notion of physical closeness is inherently part of the notion of neighborhood. The notion of neighborhoods implies the existence of boundaries in the social environment. Being able to define boundaries is necessary when identifying any type of ecological units, ranging from counties to cities to neighborhoods. For ecological units that are political entities—such as cities or counties—the boundaries are usually precisely defined. A challenge for non-political entities such as neighborhoods is that such clear boundaries generally do not exist. The physical environment of cities can appear to be a uniform space in which one street of homes fades into the next. It therefore becomes a challenge for the researcher to identify these boundaries. Previous research has adopted various boundary definitions. Some researchers have used boundaries defined by the U.S. postal service’s zip codes (Harris, 2001; Tatlow et al., 2000). These are particularly limited given that they were never meant to capture real ecological units, but instead were constructed for the explicit purpose of delivering mail. Other researchers have used units that were designated through some process occurring in the city in which the neighborhoods exist. These “named” neighborhoods often appear to constitute “real” neighborhoods (given that the residents of them often are aware of their names), though researchers have generally not attempted to validate whether they are indeed more effective measures based on some particular set of criteria. Furthermore, some scholars have observed that these named neighborhoods are created through a social process in which the boundaries between certain neighborhoods can be contested. That is, the residents of a lower status neighborhood may claim residence in a higher status adjacent neighborhood; at the same time, the residents of the higher status neighborhood have an interest in defining boundaries that exclude less desirable blocks (Halperin, 1998). An analogous process occurs as political parties bicker over the boundaries of districts every 10 years as part of the redistricting process. Most commonly, researchers use boundaries defined by the U.S. Census Bureau. Although it is common to lament the use of such officially designated units, it is nonetheless the case that these have the desirable feature of being created with the purpose of constructing something akin to “neighborhoods”. Thus, researchers often use block groups or tracts as proxies for neighborhoods (Morenoff, 2003; Sampson and Raudenbush, 2004; Wooldredge, 2002). When defining the boundaries of neighborhoods, nearly all definitions create boundaries that maximize homogeneity of the residents within a neighborhood, and maximize the degree of heterogeneity across neighborhoods. This approach is generally taken whether the neighborhoods are named by the residents within the city or whether they are defined by the U.S. Census Bureau. In fact, the Census Bureau adopted an approach explicitly maximizing homogeneity within neighborhoods based on certain key characteristics such as race/ethnicity and socioeconomic status. Named neighborhoods likewise often draw boundaries at points in which the characteristics of residents change. Indeed, a body of research in the geography literature has developed a host of algorithms that create “neighborhoods” based explicitly on the notion of maximizing homogeneity within neighborhoods and maximizing heterogeneity across neighborhoods (for a review, see Duque et al., 2007). It is useful to ask why nearly all algorithms that create “neighborhoods” attempt to cluster together into neighborhoods residents who are similar on social characteristics. We suggest that this strategy is implicitly based on the presence of social relations, and it is therefore useful to consider the process of tie formation. Although social ties in principle can form between any residents within the larger community, the geography and neighborhoods literature offers two key insights: (1) residents will be more likely to form ties with others who live closer to them in physical space (propinquity); (2) certain social categories create social distance between residents that can also create disjunctions in this structure of social ties (homophily).1 Indeed, Mayhew and colleagues (Mayhew et al., 1995) suggested conceptualizing a more general concept of distance, with social distance and physical distance as two dimensions of this more general concept. Prior research has documented the tendency to create social ties with others closer in physical space (Butts, forthcoming; Caplow and Forman, 1950; Festinger et al., 1950; Hipp and Perrin, 2009), and this tendency underlies the notion that neighborhoods will have a geographic component to them. We suggest that the common strategy in the literature of creating boundaries based on the break points in the social landscape based on the characteristics of residents is done under the implicit assumption that these represent break points in the social relations among residents. Thus, we argue that an implicit assumption underlying this approach is the notion of homophily: residents are more likely to associate with others who are more similar to themselves (McPherson et al., 2001). Studies have documented that social distance between individuals can be created by various social categories, including race/ethnicity, economic class, age, marital status, and the presence of children, and this impacts tie formation within neighborhoods even controlling for their physical location (Hipp and Perrin, 2009). There are several reasons why the pattern of social ties is important for defining neighborhoods. First, social ties can affect residents’ perceptions of neighborhood cohesion. Thus, the notion of “neighborhood” often carries with it both a sense of place as well as a community—perhaps partly imagined—of intertwined relationships. The neighborhood literature posits that cohesion is based on residents’ perceptions of attachment and similarity of values, and a large body of research has focused on the extent to which 1 There is also some evidence that certain physical boundaries can create disjunctions in the structure of social ties throughout the larger community, although we do not explore this here. Author's personal copy 130 J.R. Hipp et al. / Social Networks 34 (2012) 128– 140 residents feel a sense of attachment to the neighborhood.2 Furthermore, some theories posit that this cohesion can bring about a sense of the neighborhood as a collective unit and can impact various neighborhood-level outcomes such as the level of crime and delinquency (Sampson and Groves, 1989; Sampson et al., 1997). Second, these social ties allow residents to convey information to one another. Residents of neighborhoods can gain information about problems in their neighborhood through these ties. Given the theoretical interest in how residents respond to problems in the neighborhood through various forms of collective action, the flow of information regarding such problems is a necessary precondition for such behavior. A third mechanism through which social ties might operate is by helping residents provide various forms of collective action, such as informal social control. For instance, social disorganization theory posits that network ties are an important facilitator of residents intervening when they observe delinquent behavior in the neighborhood. For residents to actually engage in collective action behavior in response to such problems requires the flow of information to coordinate such behavior. By providing a sense of cohesion in the neighborhood, social ties can create a sense that others in the neighborhood are also willing to intervene and provide informal social control and sanctioning when observing delinquent behavior in the neighborhood. This sense of general willingness to engage in such behavior—what Sampson and colleagues refer to as collective efficacy (Sampson et al., 1997)—arguably rests on the existence of social ties. Indeed, social disorganization studies often posit that such characteristics as residential stability and racial/ethnic homogeneity affect the level of informal social control because they give rise to more social ties among residents. We therefore argue that the notion of network neighborhoods is not really novel, but rather that the idea that networks of relations to define neighborhoods actually underlies many existing approaches, at least implicitly. However, we argue for making explicit that networks are important when creating neighborhood boundaries. Arguably, the fact that few studies have taken such an approach is largely due to the difficulty of collecting such data, and not for theoretical reasons. 1.2. Network neighborhoods An initial question we need to address for constructing network neighborhoods is how to actually measure social ties. That is, what kind of social ties should we focus on when measuring neighborhoods? One approach adopted by Grannis (2009) focused only on neighborly ties. This is, of course, a very narrow definition of the social relationships of persons. An advantage of this approach is that such ties will likely be constrained to very narrow geographic areas. Indeed, Grannis (2009) found that most neighborly ties (defined as the type of informal behavior that occurs with geographic proximity) tended to be constrained to the same block and possibly a few nearby blocks. Of course, this sacrifices quite a bit of information about the total social relations of residents to focus on this one particular type of tie. To assess whether focusing only on neighborly ties is indeed reasonable requires a careful consideration of what exactly it is we wish to measure when accounting for the social relations among residents. In measuring the social ties of residents, we need to consider what these ties are theoretically expected to accomplish. Given the 2 Note that this is distinguished from the social network literature, which often measures cohesion with structural measures based on social interaction. The extent to which these structural measures and perceptual measures are related has occasionally been explored in the literature (Hipp and Perrin, 2006; Paxton and Moody, 2003). earlier considerations regarding the roles of social ties, we should determine which social ties will aid in bringing about information flow and a sense of cohesion with the neighborhood. The importance of physical closeness for defining a neighborhood suggests that information flow between persons may not be salient when people live far apart from one another. For example, social ties with persons at a distant work environment may have little meaning for the neighborhood context. Such work ties do not create information flow within the neighborhood, nor do they create a sense of cohesion. It is possible that they may impact the resident’s ability to interact with fellow neighborhood residents due to time constraints, though this would arguably be captured by the lack of ties to residents of the neighborhood (Bellair, 1997; Hipp and Perrin, 2006). Likewise, ties to friends who live further away geographically will have little impact on the neighborhood. However, focusing only on ties to residents on the same block may miss the important ties that link into the broader neighborhood, or even nearby neighborhoods that some have suggested have important effects (Bellair, 1997; Gans, 1962). Thus, it may be that it is important to focus both on neighborly ties that are extremely localized, as well as ties to nearby areas. An additional issue to consider is the strength of the ties. For example, should we focus only on strong ties? Or is it important to focus on weak ties, as some neighborhood scholars have suggested (Bellair, 1997)? If information flow is of particular interest, it may be desirable to assess the frequency of interaction for ties. If cohesion and attachment are of particular interest, it may be desirable to assess activities of ties that bring about a stronger sense of cohesion. Regardless, it is worth considering whether the valence of the ties should be measured along these various dimensions rather than simply focusing on the presence or absence of ties. If the researcher indeed has access to all appropriate social ties, then clustering into “neighborhoods” based on social relations is straightforward. However, if locally-based neighborly ties are very common, very geographically constrained (Grannis, 2009) and not captured in the measured social relations, then we suggest that one approach to constructing network neighborhoods might use the physical distance between residents as a proxy for neighborly ties, and the actual presence or absence of other social ties to further capture general social relations. Adopting an approach that accounts for both social and physical distance then raises the question of how to relatively weight social and physical distance (Butts and Carley, 1999, 2000). That is, how relatively important are each of these measures of distance? There is no clear answer, as the answer can vary based on the geographic scale, the entity of interest (e.g., people versus neighborhoods) and the characteristics of the ties (e.g., strong versus weak). Little evidence exists regarding this question, as we are aware of only one case study attempting to measure the relative contributions of social and physical distance on neighborhood social ties (Hipp and Perrin, 2009). An obvious challenge is that measuring the location of neighborhoods based on the local density of social ties requires information on all of the social ties among residents within a broader area, such as a city or county. With such information, one in principle could then estimate the boundaries of neighborhoods based on the density of these social ties. That is, we would expect to observe a high density of ties among residents who live within the same neighborhood, and a low density of ties across residents living in different neighborhoods. For example, Fig. 1 shows social ties among residents in hypothetical neighborhoods if indeed ties are more likely to form among residents within the same neighborhood. In contrast, if ties form only based on a particular physical distance function, the pattern of ties across the neighborhoods would not show such discrete breakpoints. Instead, ties would be linked to others closer in space, but there would not be evidence of such clustering. Furthermore, there would not necessarily be a tendency Author's personal copy J.R. Hipp et al. / Social Networks 34 (2012) 128– 140 131 Fig. 1. Hypothetical social network ties if neighborhood effects are present. for ties to be more likely within whatever geographic area is defined as a neighborhood rather than across neighborhoods. Once the network is defined containing valued ties between the residents of the broader community (based on some combination of physical and social distance), it is then necessary to cluster the egos. There are numerous possible algorithms that can be employed for this problem, and scholars have studied the properties of various clustering approaches and algorithms (for a nice discussion, see Fortunato, 2010). It is worth emphasizing that the bulk of these clustering algorithms are designed for dichotomous tie measures (0/1 indicators of the presence or absence of a tie), whereas our approach creates a network of valued relations. These various clustering routines generally yield a solution in which residents are clustered into various groups, or what we consider “neighborhoods”. Thus, the boundaries between neighborhoods would be constructed based on the results of these clustering approaches. A further question when employing such clustering routines for our explicitly spatial problem is whether certain constraints should be placed on possible solutions. Specifically, some have argued that residential blocks are a fundamental unit of geography (Taylor, 1997; Taylor et al., 1984), given that social ties are often particularly dense on blocks, and neighborly ties are sometimes constrained to just a single block, at least in urban environments (Grannis, 2009). This implies constraining the solution such that households on same block are always considered part of the same neighborhood. Another question that should be addressed is the often implicit assumption of much prior research that each household must be located in one neighborhood. Must it be the case that a person must live in a neighborhood? Or is it possible to have isolates, and might we consider them to reside in their own neighborhood? In the case of rural areas, it may be reasonable to suppose that geographic isolates indeed live in their own neighborhood. Grouping them into a “neighborhood” with other households that are quite geographically distant arguably does not make conceptual sense. On the other hand, it may not be reasonable to consider those living in more urban areas to be isolates in their own neighborhood. In urban areas, residents are rarely very far away from others. This suggests that the possibility of social ties between a person and other residents is plausible, and that even if such residents choose to socially isolate themselves from nearby residents, there is no reason to suspect that they indeed constitute their own neighborhood. Indeed, such residents are still subject to the same perils in the environment (e.g., crime) that other nearby residents encounter. Another possibility that is infrequently considered in the neighborhood effects literature is that persons might be considered part of more than one neighborhood. Arguably, a neighborhood is at least in large part constituted by the social and spatial presence of persons. But these persons need not be residents.3 For example, a person who spends their evenings in and around their home clearly live in the neighborhood surrounding their home, whatever its boundaries might be. To the extent that they talk to their neighbors, they can enhance the information flow in the neighborhood, help foster a sense of cohesion, and provide informal social control by intervening when they observe others engaging in delinquent behavior. But this same person may also spend the daytime several days a week at a workplace that is far removed from their neighborhood. To the extent that their time is exclusively spent within the confines of this work environment while they are at work, they will not have much impact on the neighborhood surrounding their workplace (for example, see Duneier, 1999; Jacobs, 1961). But to the extent that they are out and about in this area—either walking to and from their car, walking about during their lunch hour, etc., they might be considered part of this neighborhood. Furthermore, consider a person who frequently chooses to spend their evenings in an area not near their home. This can occur for several reasons: it might occur because they have a particular friend or group of friends who live in this other area; it might occur because this 3 Further examples are persons who regularly are in certain neighborhoods as part of delivery jobs. For example, newspaper delivery, mail carriers, and trash collectors all spend time on a regular basis in some neighborhoods. Of course, it is unlikely that they develop a sense of attachment to the neighborhood, and therefore are more likely to be classified as visitors rather than neighborhood members. Author's personal copy 132 J.R. Hipp et al. / Social Networks 34 (2012) 128– 140 area has amenities that are enticing to the person; it might occur because the area has particular characteristics with which the person identifies. If they only consider themselves to be visitors to this area, this will arguably not be considered their own neighborhood. If they in fact identify with the neighborhood, then this might be considered a second neighborhood for them. This idea has rarely been considered in prior research. Of course, it is extremely difficult to obtain information on all of the social ties within a larger community. The approach we adopt here uses information on social ties among the adolescents within the schools of a county. Although such data will provide us some interesting insights, they also raise several methodological challenges as we describe below. Nonetheless, we view this as an exploratory study of how network neighborhoods might be constructed. We will compare the results from our approach (giving varying weights to social ties) with Census-defined neighborhood boundaries, as well as a purely spatial approach. We turn to a description of our data next. 2. Research design and methods