Activity Locations, Residential Segregation, and the Significance of Residential Neighborhood Boundary Perceptions

Abstract

The inadequacies of residential census geography in capturing urban residents’ routine exposures have motivated efforts to more directly measure residents’ activity spaces. In turn, insights regarding urban activity patterns have been used to motivate alternative residential neighborhood measurement strategies incorporating dimensions of activity space in the form of egocentric neighborhoods—measurement approaches that place individuals at the center of their own residential neighborhood units. Unexamined, however, is the extent to which the boundaries of residents’ own self-defined residential neighborhoods compare with census-based and egocentric neighborhood measurement approaches in aligning with residents’ routine activity locations. We first assess this question, examining whether the boundaries of residents’ self-defined residential neighborhoods are in closer proximity to the coordinates of a range of activity location types than are the boundaries of their census and egocentric residential neighborhood measurement approaches. We find little evidence that egocentric or, crucially, self-defined residential neighborhoods better align with activity locations, suggesting a division in residents’ activity locations and conceptions of their residential neighborhoods. We then examine opposing hypotheses about how self-defined residential neighborhoods and census tracts compare in socioeconomic and racial composition. Overall, our findings suggest that residents bound less segregated neighborhoods than those produced by census geography, but self-defined residential neighborhoods still reflect a preference toward homophily when considering areas beyond the immediate environment of their residence. These findings underscore the significance of individuals’ conceptions of residential neighborhoods to understanding and measuring urban social processes such as residential segregation and social disorganization.

How residents experience and perceive their residential or home neighborhoods has been of central concern to scholars since the inception of urban studies (Du Bois 1899; Simmel 1903; Park and Burgess 1925; Wirth 1938; Durkheim 1951; Lynch 1960). Indeed, decades of research regarding the multifaceted influences of neighborhood contexts on human development and wellbeing has yielded a wealth of insights (Faris and Dunham 1939; Shaw and McKay 1969; Sampson, Raudenbush, and Earls 1997; Leventhal, Dupéré, and Brooks-Gunn 2009; Chetty, Hendren, and Katz 2016). Nevertheless, an ongoing issue facing the field is the indeterminate nature of the “neighborhood” concept. Prevailing approaches to neighborhood operationalization have largely relied on the widespread availability of administrative units or census geography to define neighborhoods. Yet, extant conceptual approaches to neighborhoods have emphasized alternative bases for defining neighborhoods. For instance, neighborhoods have been conceptualized as capturing a wide variety of residents’ routine activities such as schools, religious institutions, recreational and leisure activities, and interactions with friends and neighbors (Shaw and McKay 1969; Kornhauser 1978; Wilson 1987; Anderson 1999; Small and McDermott 2006). Residents’ own neighborhood perceptions have also have been underscored as important to the conceptualization of neighborhood space, measurement of neighborhood social processes such as social disorganization, and mobilization of residents in defense of the neighborhood area (Sampson et al. 1997; Gieryn 2000; Coulton et al. 2001; Sampson 2012; Burdick‐Will 2018; Pratt et al. 2019). In this view, the boundaries of residents’ self-defined neighborhoods reflect local organization of behavior and sentiments of place (Park and Burgess 1925; Janowitz 1967; Suttles 1972; Hunter 1974; Campbell et al. 2009).

In light of these more theoretically informed conceptualizations of neighborhood space, census geography has been called into question as an accurate representation of how urban residents actually experience and view their own residential neighborhood environments (Coulton et al. 2001; Grannis 2009; Sperling 2012). One especially notable critique of census geography is the observation that residents’ routine activity locations tend not to be within the boundaries of their census tract of residence, with potentially significant consequences for the interpretation of neighborhood research (Golledge and Stimson 1987; Sastry, Pebley, and Zonta 2002; Kwan 2012; Browning and Soller 2014). In response to this concern, contemporary urban researchers have called for more direct measurement of urban residents’ extra-neighborhood exposure patterns in the form of activity space approaches (Kwan 2009, 2012; Matthews and Yang 2013; Browning and Soller 2014; Crawford et al. 2014; Tana, Kwan, and Chai 2016; Wang et al. 2018; Zenk et al. 2019).

While measurement of activity space and its composition is a growing area of research in urban studies, understanding the nature and consequences of residential neighborhood environments remains a central concern, particular in research on the contexts of child and adolescent development (Sampson 2012; Chetty et al. 2016). Features of residential neighborhood social organization such as the extent of social network interaction, local sentiments of attachment, mutual trust and shared pro-social normative orientations (Park 1915; Park and Burgess 1925; Jacobs 1961; Suttles 1972; Sampson 2012) may exert uniquely powerful influence on youth outcomes by comparison with features of non-neighborhood settings. Moreover, residents themselves tend to disagree with the notion that census geography accurately represents the boundaries of their residential neighborhoods, highlighting the need for better understanding of the lived experience of neighborhood settings and the intersection of everyday routines and neighborhood boundaries (Coulton et al. 2001; Campbell et al. 2009; Burdick‐Will 2018; Pratt et al. 2019). This is best evidenced in the few studies that have asked respondents to report the boundaries of their self-defined residential neighborhoods, with resulting areas exhibiting substantial heterogeneity in neighborhood boundary perceptions even among proximate residents (Coulton, Jennings, and Chan 2013).

Insights from the literature on urban activity patterns have in turn been used to motivate alternative residential neighborhood measurement strategies incorporating dimensions of activity space (Jones and Pebley 2014; Hasanzadeh, Broberg, and Kyttä 2017). To this end, egocentric residential neighborhoods—which place individuals at the center of their own residential neighborhood units—have been proposed as alternatives to census geography in the absence of resident-specific activity data (Gómez et al. 2004; Lee et al. 2008, 2018; Hipp and Boessen 2013; Matthews and Yang 2013; Crawford et al. 2014). Use of egocentric neighborhoods is inspired by evidence suggesting that residents’ activity locations and social ties tend to be within relatively close range of their home address, highlighting the potential for error when designating a census unit as a residential neighborhood for residents who live close to the edge of census boundary. For example, Sastry et al. (2002) find that only 16 percent of residents’ routine grocery stores and 12 percent of residents’ places of worship are included in their census tract of residence. At the same time, these same routine destinations tended to be within a two-tract range of their home address. Similarly, in a study of 55 adolescent males from Philadelphia, Basta, Richmond, and Wiebe (2010) find that while 70 percent of respondents’ non-home time was spent beyond their residential census tract, the average maximum distance travelled away from home did not exceed 1.5 miles (2,414 meters). These findings underscore that though residents’ census tracts may poorly approximate residents’ non-home exposure locations, the available evidence suggests that residents do tend to spend time at places that are within a relatively short distance of their home address. This pattern also pertains to residential social ties, as the likelihood of network ties between residents is highest between nearby residents, and declines following a distance decay function (Hipp and Perrin 2009).

Egocentric residential neighborhoods have been shown to better capture variation in residents’ potential exposures such as crime and high-traffic activity locations compared to residents’ residential census tracts (Allen and Turner 1995; Sastry et al. 2002; Hipp and Boessen 2013; Crawford et al. 2014, 2014; Perchoux et al. 2016). Furthermore, egocentric residential neighborhoods are more diverse in composition than increasingly homogenous census geography, more precisely reflecting exposure to local residential segregation processes (Lee et al. 2008, 2018; Sperling 2012). Taken together, there is conceptual as well as empirical grounds for use of egocentric residential neighborhood measurement strategies over census-based approaches. To date however, no study has empirically investigated the extent to which census tracts, egocentric neighborhoods and residents’ own self-defined residential neighborhoods—the latter being a measurement strategy of the Adolescent Health and Development in Context (AHDC) Study, described below—comparatively align with the range of their routine activities of various types.

Drawing on alternative theoretical approaches to understanding the process by which residents define the boundaries of their residential neighborhoods, we consider the significance of both routine activity patterns and social affiliative explanations of self-defined residential neighborhoods—focusing on racial composition and socioeconomic status—as key inputs to the identification of subjective neighborhoods. First, we assess how residents’ census tracts, egocentric residential neighborhoods, and self-defined residential neighborhoods compare in aligning with the coordinates of a wide range of residents’ self-reported activity location types—such as schools, workplaces, grocery stores, and friends’ houses. Importantly, our hypothesis is not that self-defined neighborhoods will include routine activity locations. Instead, we test the hypothesis that self-defined residential neighborhood boundaries are in closer proximity to the coordinates of a range of activity location types than are the boundaries of residents’ census-based and egocentric residential neighborhoods. This expectation is guided by recent studies considering alignment between self-defined residential neighborhoods and various activity types (Campbell et al. 2009; Spilsbury, Korbin, and Coulton 2009; Crawford et al. 2014; Iossifova 2015; Pratt et al. 2019). For example, Burdick‐Will (2018) finds that parents’ self-defined residential neighborhoods are pulled toward the location of the school their child attends, and Perchoux et al. (2016) find that self-defined neighborhood boundaries are influenced by potential exposure areas as measured by local prevalence of high-traffic destinations. In addition, we assess the hypothesis that egocentric neighborhoods better align with residents’ key locations compared to census tracts.

Second, we explore alternative perspectives on the influence of social affiliative processes in the bounding of self-defined residential neighborhoods, focusing on the well-established tendency toward homophily in social and spatial interaction (McPherson, Smith-Lovin, and Cook 2001; Wang et al. 2018). Residents’ self-defined residential neighborhoods may be bounded in such a way as to maximize the prevalence of comparable social characteristics and minimize connectedness with neighborhoods of differing compositions or routine activity types, producing estimates similar to those based on census geography and illustrating a highly segregated image of the city in the aggregate (Taylor, Gottfredson, and Brower 1984; Campbell et al. 2009; Sperling 2012; Jackson and Benson 2014). In contrast, to the extent that residents bound residential neighborhoods with regard to activity locations or affiliative symbolic ties beyond their residential census tract, we expect compositions of self-defined residential neighborhoods to be more similar to estimates based on egocentric residential neighborhoods, underscoring that residents bound more compositionally diverse residential neighborhoods than is illustrated by oft-utilized census geography (Downs and Stea 1974; Guest and Lee 1984; Lee et al. 2008, 2018; Campbell et al. 2009).

Furthermore, should self-defined neighborhood significantly align with the coordinates of residents’ routine activity locations, compositions of self-defined neighborhoods may be indicative of residents’ broader exposure to social segregation (Krivo et al. 2013; Jones and Pebley 2014; Browning et al. 2017). In the case that self-defined residential neighborhoods do not better align with the coordinates of routine activities but are yet still more diverse than residents’ census tracts, compositions of self-defined neighborhoods are likely more indicative of residents’ perceived affiliative ties (Park and Burgess 1925; Suttles 1972; Campbell et al. 2009). In this instance, the boundaries of self-defined neighborhoods are nevertheless important for understanding sentiments of attachment and behavioral orientations toward local spaces (for example, the willingness to mobilize on behalf of a local space is potentially heavily influenced by whether that space is identified as part of the “neighborhood”).

Our investigations represent, to our knowledge, the first assessments of 1) how residential neighborhood measurement strategies compare in aligning with residents’ self-reported routine activity locations and 2) how self-defined and census-based residential neighborhoods compare in racial and socioeconomic composition, with potentially significant implications for our understanding of residential segregation.

DATA AND MEASURES

Data are from the Adolescent Health and Development in Context (AHDC) study; a longitudinal, representative study of 1,401 youth aged 11 to 17 and their caregivers in Franklin County, OH, USA (Boettner, Browning, and Calder 2019). The AHDC study area is within Interstate 270, the outer belt highway of the Columbus, OH area and is comprised of a diverse set of urban and suburban communities of varying affluence and income levels. We employ data from the in-home entrance surveys administered to residential caregivers (mean age = 45.63, 87% female) which included a question sequence on their neighborhood and personal routine activity locations. Prior to questioning respondents about their neighborhoods, interviewers asked a series of questions about residential stability, including how long the respondent had resided in Franklin County, Ohio and how many times they had moved in the past two years. Caregivers were then asked to report which of the following locations they go to on a regular basis: Work (primary job), school, library, religious place or church, civic organization (e.g. local nonprofit), neighborhood organization (e.g. neighborhood watch), shopping or grocery store, relative’s house, friend’s house, recreation space(s) (e.g. rec center, park, sports fields, fitness club), restaurant, store or other business, and somewhere else. If there were other key routine destinations not included in this list, respondents were asked to report them. If the respondent had more than one of any of these location categories (e.g. two grocery stores they regularly frequent), they were asked to report those as well. Respondent caregivers were then asked to report a self-determined name for each destination, with interviewers able to use a Google Maps application interface to search for establishment names or addresses or drop a pin to indicate the correct location. Employing this interviewer-assisted Google Maps-based “location generator” module, respondents were asked to identify the location of each destination. Respondents were then questioned about what they do at each location and what days and times they typically go there.

Finally, respondents were asked about their own residential neighborhood spaces. To delineate self-defined residential neighborhood boundaries, respondents were presented with their current residential location on Google Maps and asked to provide four major street intersections or landmarks that they “think of as the boundaries of [their] neighborhood.” Residential neighborhood boundary coordinate points were then 1) ordered based on the shortest path connecting the points that returns to the first ordered point (i.e. by solving the so-called traveling salesman problem) and then 2) connected by the most efficient road network path between the ordered points per OpenStreetMap road network data using the “ArcPy” ArcGIS module for Python (Applegate et al. 2006; ESRI 2011; OpenStreetMap contributors 2015). A small buffer was created around the neighborhood boundary line segments in cases where two road network segments did not meet but were very close, to effectively join the line segments into a polygon. We only consider respondents who report at least three neighborhood boundary coordinates and whose neighborhood boundary coordinates yield a coherent polygon based on the road network for the present analyses, thus restricting the sample to 1,152 adult caregivers. See Appendix A for more information on data cleaning and sample selection.

Lastly, Census tract administrative neighborhoods were linked based on respondents’ home addresses.² Following Hipp and Boessen (2013) egocentric individual social environment neighborhoods were constructed as circular areas with a half-mile (804.7 meter) radius surrounding home addresses using the R “sf” package (Pebesma et al. 2019). Self-defined and egocentric residential neighborhood compositions are based on aggregations of 2010 census block (for race) and 2013 block group (for adult poverty) compositions proportional to their areal contribution to the neighborhood area. For example, if an egocentric neighborhood was comprised of two block groups, the first of which accounting for 90% percent of the egocentric neighborhood and with a poverty rate of .25 and the second of which accounting for 10% percent of the egocentric neighborhood and with a poverty rate of .8, the egocentric neighborhood poverty rate = (.25*.9) + (.8*.1) = .305. Notably, we created a binary control measure for whether the home address is in or out of the self-defined neighborhood, as it is relatively common for homes to be outside the neighborhood. The median distance from the home to their nearest neighborhood border is .09 miles (1609 meters)³.

CONTROLS

Years in the current neighborhood were reported by adult caregivers in response to the question “How long have you lived in this neighborhood, in any house or apartment?” after reporting their self-defined neighborhood boundaries. Number of Moves in the last two years” is measured in response to the number of reported address changes by the respondent during the past two years, with provided responses ranging from 0–3. Because only 29 respondents reported more than one move, categories were collapsed to 0 = no move and 1 = any move during the past two years. Home ownership is measured in response to the question “Is this home rented or owned by a member of this household?” with a third option allowing the respondent to report “some other” housing arrangement. Categories were collapsed to 0 = does not own residence and 1 = own residence. Marital status categories included married, cohabiting, single, divorced, separated, and widowed. Separated and widowed were collapsed into an “other” category, with other categories left as is. Household income is based on a self-report of total income from all sources during the past year, with 14 response categories ranging from 1 = under $10,000 to 14 = more than $250,000. Educational attainment is based on the respondent’s report of their highest level of education completed, with response categories ranging from 0 = no formal schooling to 11 = professional degree. Response categories were collapsed into five categories including no secondary education, secondary degree, some college, four-year college degree, and graduate degree. Finally, gender and race-ethnicity are based on self-reports, with racial categories including white, black, of any Hispanic origin, multiracial, and with all other racial category responses collapsed to “some other” race.

Means and proportions for focal study measures are displayed in Table 1. We retain 75 respondents missing on family income, 25 missing on marital status, 20 missing on home ownership, 12 missing on parent education, 4 missing on age, 2 missing on gender, and 1 missing on years at current address using multiple imputation by chained equations procedures with 5 iterations in STATA (StataCorp 2017), bringing our final analytic sample to 1,152 respondents.

Table 1.

Descriptive Statistics.

	Mean	(SD)	Range
Number of locations reported	8.28	(4.00)	0 – 12
Average distance (meters) from location to:
Self-defined neighborhood boundary	4307	(3286)	0 – 34589
Tract boundary	4217	(3259)	0 – 34770
Egocentric neighborhood boundary	4329	(3233)	0 – 33878
Neighborhood characteristics
Neighborhood area (sq. miles)	1.55	(2.41)	0 – 22.21
Proportion Black	0.24	(0.26)	0 – .86
Proportion poverty	0.21	(0.15)	0 – .99
Census tract characteristics
Proportion Black	0.28	(0.29)	0 – .82
Proportion poverty	0.22	(0.17)	0 – .91
Population density	4991	(2265)	410 – 21063
Moved in last two years	0.15	-	0 – 1
Years at current address	12.48	(10.31)	0 – 62
Own residence	0.64	-	0 – 1
Home in neighborhood	0.80	-
Age	45.63	(8.41)	26 – 81
Male	0.13		0 – 1
Marital status
Married	0.58	-	0 – 1
Cohabiting	0.09	-	0 – 1
Single	0.18	-	0 – 1
Other	0.15	-	0 – 1
Race-Ethnicity
White	0.55	-	0 – 1
Black	0.36	-	0 – 1
Hispanic	0.05	-	0 – 1
Multiracial	0.02	-	0 – 1
Other	0.03	-
Household income	6.30	(4.12)	1 – 14
Educational attainment
No secondary degree	0.05	-	0 – 1
Secondary degree	0.13	-	0 – 1
Some college	0.35	-
4-Year degree (ref.)	0.26	-	0 – 1
Graduate degree	0.20	-	0 – 1
N	1,152

Notes: Descriptive statistics based on MICE imputation with 5 iterations for 75 respondents missing on family income, 25 missing on marital status, 20 missing on home ownership, 12 missing on parent education, 4 missing on age, 2 missing on gender, and 1 missing on years at current address.

NEIGHBORHOOD DELINEATIONS AND ROUTINE ACTIVITY LOCATIONS

To compare the extent to which census tracts, egocentric neighborhoods, and self-defined neighborhoods align with routine activity spaces, we first determined the number of each respondents’ reported locations within the boundaries of each neighborhood delineation. While the range of this count is 0 to 12, the average number of locations within any of the neighborhood delineations per respondent does not exceed 1 (x̄ = .86, SD = 1.36). Even at the 90^th percentile of this count, the number of locations within self-defined neighborhoods is 3, and only 2 for egocentric neighborhoods and census tracts. We thus move to comparing the Euclidean distance between caregiver reported locations and the nearest boundary of their self-defined neighborhood, egocentric neighborhood, and census tract (e.g. distance to the respondent’s nearest self-defined neighborhood boundary minus the distance to their nearest tract boundary). In cases where reported locations are inside the respective neighborhood delineation, the distance to the neighborhood delineation boundary equals zero (reflecting that the location is encompassed by the unit), and all pairs of differences in the average distance to the nearest boundary are assumed to be normally distributed.

For illustration, Figure 1 contains an example based on synthetic data illustrating the comparison of the distance (black lines) between a reported routine activity location and the nearest boundary of their self-defined neighborhood (gray), egocentric neighborhood (black circle), and census tract (white). In this example, the reported grocery store is closest to the nearest self-defined neighborhood boundary, followed by the nearest egocentric neighborhood boundary, indicating that the boundaries of these neighborhood measurement strategies are in closer proximity to the grocery store than are the boundaries of the census tract of residence. For the present analyses we focus on location types reported by respondents, including workplaces, schools, libraries, churches, civic and neighborhood organizations, grocery stores, relatives’ and friends’ houses, parks and recreation areas, restaurants, and stores and business. Of the 7,620 unique locations reported by respondents in our analytic sample, 191 outliers are dropped because the distance was larger than 30,000 meters (18.6 miles) to the nearest neighborhood boundary.

Figure 1.

Example resident address (home), census tract (white outline), egocentric neighborhood (black outline), self-defined neighborhood boundary points (1–4), resulting self-defined neighborhood following the road network (gray area), self-reported grocery store location, and the distance from this location to the nearest boundary of each neighborhood delineation (black lines). This is a purely illustrative example and does not represent data for an actual respondent in order to ensure anonymity.

As is evident in Table 1, there is considerable variation in the number of locations reported by residents. Additionally, comparisons between distances to the nearest neighborhood delineation border may vary systematically with individual-level demographic factors or neighborhood characteristics such as size (Coulton et al. 2013). To account for this variation, we estimate differences in the distance to the nearest neighborhood delineation boundary in the context of a three-level linear multilevel model, with location distance differences clustered within respondents and respondents clustered within census tracts of residence (Raudenbush and Bryk 2002). This allows us to estimate average differences in the distance to the nearest respective boundaries for each location type in reference to some excluded location type, and net of covariates. Postestimation average predictions of the distance difference can then be obtained for each location type.

Results for these models are displayed in Table 2, with the first model presenting results for the difference of the distance to the respondent’s nearest self-defined neighborhood boundary minus the distance to their nearest census tract boundary. For example, if a respondent’s given location is 1,000 meters to the nearest boundary of their self-defined residential neighborhood and 1,200 meters to the nearest boundary of their residential census tract, the difference can be calculated as 1,000 – 1,200 = −200 meters. In this example, the given location is 200 meters closer to the respondent’s nearest self-defined residential neighborhood boundary than their nearest census tract boundary. Examination of the standardized neighborhood size coefficient indicates that a one standard deviation increase in the natural log of the size of self-defined neighborhoods is associated with a 500.21 meter reduction in the outcome (p < .001). That is, on average, as neighborhood size increases, locations tend to be further from the nearest census tract boundary than they are from the nearest self-defined neighborhood boundary. Subsequent investigation of this association underscores the major relevance of self-defined neighborhood size to understanding how neighborhoods are bounded with regard to routine locations. Specifically, in predictions not shown, a clear turning point is evident at the 50^th percentile of neighborhood size (1.41 sq. kilometers, .54 sq. miles), where neighborhoods larger than this outperform census tracts, but where neighborhoods smaller than this are outperformed by census tracts. Additionally, a one unit increase in the number of locations reported by the respondent is, on average, associated with a 115.14 meter increase in the outcome (p < .001).

Table 2.

Linear Three-level Multilevel Models of the Differences in Distance to Nearest Neighborhood Boundary.

	Self-defined Neighborhood minus Tract Distance		Egocentric Neighborhood minus Tract Distance
	b	(se)	b	(se)
Individual-level characteristics
z(ln(Neighborhood size))	−500.21***	(73.82)	-
Number of locations reported	115.14***	(21.00)	118.90***	(21.69)
Moved in last two years	−29.43	(207.20)	35.23	(213.87)
Years at current address	−2.58	(7.72)	−5.49	(7.93)
Own residence	23.05	(203.10)	110.51	(207.38)
Home in neighborhood	145.50	(167.30)	-
Age	2.953	(9.61)	1.29	(9.91)
Male	29.21	(212.49)	43.25	(219.94)
Married (ref.)	-		-
Cohabiting	−39.17	(269.53)	108.75	(277.32)
Single	22.67	(219.17)	35.34	(225.77)
Other	31.58	(213.31)	26.12	(219.72)
Race-Ethnicity
White (ref.)	-		-
Black	−116.83	(178.15)	−98.40	(183.03)
Hispanic	−73.37	(331.49)	−142.82	(341.75)
Multiracial	−194.76	(415.88)	−46.19	(418.62)
Other	−216.88	(500.04)	−72.80	(503.36)
Household income	8.48	(23.24)	13.55	(23.83)
Educational attainment
No secondary degree	−389.99	(357.71)	−8.31	(368.11)
Secondary degree	69.39	(246.71)	35.37	(254.57)
Some college	−6.805	(200.71)	98.89	(206.81)
4-Year degree (ref.)	-		-
Graduate degree	−185.88	(197.01)	−88.77	(203.73)
Location-level
Average predicted difference by location type
Workplace (n=711)	2788.45***	(216.8)	−324.32	(223.57)
School (n=1210	−341.22*	(166.78)	111.28	(167.48)
Library (n=321)	−1217.15***	(323.94)	127.07	(335.10)
Church (n=469)	779.74**	(265.01)	−87.93	(273.82)
Civic organization (n=160)	627.51	(456.97)	341.24	(474.16)
Neighborhood organization (n=32)	−2907.46**	(1008.86)	−321.31	(1036.78)
Grocery store (n = 1,501)	−699.18***	(149.35)	195.96	(153.96)
Relative’s house (n=535)	1172.25***	(252.61)	182.76	(262.46)
Friend’s house (n=364)	67.40	(300.85)	−348.68	(311.30)
Park/Recreation area/center (n=314)	−420.23	(323.73)	58.76	(335.37)
Restaurant (n=755)	−716.98**	(210.64)	−133.95	(217.78)
Store/business (n=832)	−18.20	(201.35)	389.99	(207.69)
Other location (n=227)	700.30	(384.08)	508.42	(396.49)
Constant	223.75	(442.50)	−689.17	(441.15)
Census Tract N	184		184
Respondent N	1,152		1,152
Location N	7,431		7,431

Standard errors in parentheses

^*p < .05

^**p < .01

^***p < .001 (two-tailed).

Notes: z(ln(neighborhood size)), moved in last two years, years at current address, residence ownership, home inside or outside self-defined neighborhood, age, gender, and family income are centered. 191 locations with a difference in the distance larger than 30,000 meters were dropped to reduce the influence of outliers.

Turning to location-specific predictions, on average, schools (b = −341.22, p < .05), libraries (b = −1217.15, p < .001), neighborhood organizations (b = −2907.46, p < .01), grocery stores (b = −699.18, p < .001), and restaurants (b = −716.98, p < .01) tend to be closer to respondents’ nearest self-defined neighborhood boundary than their nearest census tract boundary. In contrast, workplaces (b = 2788.45, p < .001), churches (b = 779.74, p < .01), and relatives houses (b = 1172.25, p < .001) tend to be closer to the respondents’ nearest tract boundaries than their nearest neighborhood boundary. Null differences in the distance to the nearest boundary between census tracts and self-defined neighborhoods are observed for respondents’ civic organizations, friends’ houses, parks and recreation areas, stores and business, and ‘other’ unclassified locations.

The second model set presents differences in the average location distance to the respondents’ nearest egocentric neighborhood boundary minus the location distance to their nearest census tract boundary. Examination of the location predictions offers no evidence of differences in the proximity of census tract or egocentric neighborhood boundaries to the coordinates of self-reported activity locations.

Taken together, none of the neighborhood delineations considered align with activities locations particularly well. Most relevant to our hypotheses, though we find that the boundaries of self-defined neighborhoods are on average in closer proximity to some routine activity location types—such as schools and grocery stores—than are census tract boundaries, census tracts outperform self-defined neighborhoods in this respect regarding other key routine activity types such as workplaces and churches, and there are null differences for other activities such as civic organizations, parks/recreation centers, and non-grocery stores and businesses. Further, the extent to which self-defined neighborhood boundaries are in closer proximity to locations is largely contingent on the size of self-defined neighborhoods—a topic for which empirical investigation remains limited (Coulton et al. 2013)—highlighting that self-defined residential neighborhoods are unreliably bounded in the direction of residents’ routine activities. These findings cast doubt on the notion that residents’ self-defined residential neighborhood boundaries align with activity locations, but point to the potential significance of affiliative processes in understanding how urban residents bound their residential neighborhoods (Guest and Lee 1984; Campbell et al. 2009; Iossifova 2015; Burdick‐Will 2018; Pratt et al. 2019).

NEIGHBORHOOD COMPOSITIONS

We now turn to evaluating how the compositions of self-defined residential neighborhoods diverge from the compositions of census tracts—the oft-utilized unit of measurement in calculations of segregation indices (Sperling 2012). As such, the dependent variable is the difference between a given compositional feature (in our case, proportion poverty or proportion black) of the self-defined neighborhood and that of the reference census tract of residence. For example, if a respondents’ self-defined residential neighborhood has a poverty rate of .35, and their census tract has a poverty rate of .45, the difference can be calculated as .35 - .45 = −.1. Thus, the negative difference (less than 0) would mean that the self-defined neighborhood has a 10-percentage point lower poverty rate than that of their census tract of residence. Conversely, a positive difference of .1 would represent a self-defined neighborhood that has a 10-percentage point higher poverty rate than that of their census tract.

All presented composition analyses are based on two-level multilevel linear models (respondents clustered within census tracts) of differences between compositional characteristics net of individual and contextual controls. The first and second columns of Table 3 display summaries of the coefficients for these models. In the first model presented, the dependent variable is the difference of residents’ self-defined neighborhood portion poverty minus census tract of residence proportion poverty. Census tract proportion poverty is negatively associated with the expected difference between self-defined neighborhood proportion poverty and census tract proportion poverty (b = −.304, p < .001). This relationship is displayed graphically in the top panel of Figure 2. Specifically, when all other variables are held at their means, residents of census tracts low in poverty bound self-defined neighborhoods which are relatively higher in proportion poverty, and residents of census tracts high in poverty bound neighborhoods which are relatively lower in proportion poverty.

Table 3.

Linear two-level Multilevel Models of the Differences in Composition Characteristics and Z-score of Self-Defined Neighborhood Compositions.

	Self-defined Neighborhood minus Tract %Poverty		Self-defined Neighborhood minus Tract %Black		Z-score Self-Defined Neighborhood %Poverty		Z-score Self-Defined Neighborhood %Black
	b	(se)	b	(se)	b	(se)	b	(se)
Census tract-level
Census tract characteristics
Proportion Black	0.025	(0.020)	−0.238***	(0.030)	-		-
Proportion poverty	−0.304***	(0.033)	−0.035	(0.050)	-		-
z(ln(Population density))	−0.001	(0.004)	0.001	(0.006)	-		-
Individual-level characteristics
Local area compositional characteristics
Proportion Black	-		-		-		.924***	(0.208)
Proportion poverty	-		-		1.08***	(0.273)	-
z(ln(Neighborhood size))	−0.004	(0.003)	−0.031***	(0.004)	0.016	(0.029)	0.029	(0.033)
Moved in last two years	0.004	(0.008)	0.004	(0.010)	−0.011	(0.074)	0.032	(0.087)
Years at current address	0.000	(0.000)	0.000	(0.000)	−0.007*	(0.003)	−0.001	(0.003)
Own residence	−0.006	(0.008)	0.000	(0.010)	0.043	(0.074)	−0.146	(0.086)
Home in neighborhood	−0.001	(0.006)	0.025**	(0.008)	−0.054	(0.061)	0.074	(0.070)
Age	−0.000	(0.000)	−0.000	(0.000)	0.007*	(0.003)	0.008*	(0.004)
Male	−0.001	(0.008)	−0.016	(0.010)	0.123	(0.077)	−0.119	(0.089)
Married (ref.)	-		-		-
Cohabiting	−0.015	(0.010)	−0.014	(0.012)	−0.202*	(0.094)	−0.197	(0.109)
Single	0.014	(0.008)	0.005	(0.010)	−0.053	(0.079)	−0.064	(0.092)
Other	0.004	(0.008)	0.011	(0.010)	−0.001	(0.079)	−0.188*	(0.091)
Race-Ethnicity
White (ref.)	-		-		-		-
Black	0.020**	(0.008)	0.054***	(0.010)	0.192**	(0.077)	0.238**	(0.092)
Hispanic	0.011	(0.012)	0.024	(0.016)	0.132	(0.125)	−0.112	(0.144)
Multiracial	0.020	(0.019)	0.013	(0.025)	0.091	(0.190)	−0.123	(0.223)
Other	0.037*	(0.015)	0.005	(0.020)	−0.057	(0.149)	0.021	(0.175)
Household income	−0.002	(0.001)	−0.001	(0.001)	0.015	(0.009)	0.008	(0.010)
Educational attainment
No secondary degree	0.032*	(0.013)	0.025	(0.017)	0.044	(0.129)	−0.159	(0.151)
Secondary degree	0.034***	(0.009)	−0.004	(0.012)	0.114	(0.092)	−0.179	(0.107)
Some college	0.018*	(0.008)	0.009	(0.010)	0.101	(0.076)	0.011	(0.088)
4-Year degree (ref.)	-		-		-		-
Graduate degree	0.004	(0.008)	−0.013	(0.010)	−0.062	(0.074)	−0.095	(0.087)
Constant	0.031**	(0.010)	0.001	(0.015)	−0.400***	(0.101)	−0.370***	(0.106)
Census Tract N	184		184		184		184
Respondent N	1,152		1,152		1,152		1,152

Standard errors in parentheses

^*p < .05

^**p < .01

^***p < .001 (two-tailed).

Notes: z(ln(neighborhood size)), moved in last two years, years at current address, residence ownership, home inside or outside self-defined neighborhood, age, gender, and family income are centered.

Figure 2.

Postestimation Predictions for the Respective Difference between the Self-defined Neighborhood and Census Tract Compositional Characteristic at Levels of the Respective Census Tract Compositional Characteristic.

The second column presents models of the difference of residents’ neighborhood proportion black minus census tract proportion black, with results largely matching the patterns illustrated for proportion poverty. Census tract proportion black (b = −.238, p < .001) is negatively associated with the expected difference between self-defined proportion black and census tract proportion black. This relationship is presented graphically in the second panel of Figure 2.

Taken together, we find that self-defined neighborhoods tend to be more racially and socioeconomically diverse than largely homogenous census tracts, indicating that oft-utilized census geography overestimates how residents themselves perceive and experience residential segregation patterns (Sperling 2012). Considered alongside our finding that residents’ self-defined neighborhood boundaries are unreliably in closer proximity to their routine activity locations than are the boundaries of their census tract, this compositional pattern is unlikely to be explained by compositional diversity in residents’ broader activity spaces (Jones and Pebley 2014; Tana et al. 2016). This suggests that our compositional results explicitly pertain to how residents conceive of home residential neighborhood environments.

Even so, though census geography is highly relevant to residential segregation research, we contend that the substantial homogeneity within census tracts may provide an incomplete test of the expectation of homophilous self-defined neighborhoods. Indeed, contemporary neighborhood research underscores the importance of considering proximate areas beyond the immediate area to understanding how residents experience their local environments (Lee et al. 2008; Hipp and Boessen 2013). We thus present an additional set of analyses in which we consider where the composition of residents’ self-defined neighborhoods falls within the distribution of the composition of the local environment. To do so, we first generated twenty randomly placed egocentric neighborhoods around the respondent’s home, all of which encompass the home, and all of which are equal in area to that of their self-defined neighborhood. To ensure that each of these twenty egocentric neighborhoods encompasses the home, an initial egocentric neighborhood equal in area to the respondent’s self-defined neighborhood was first generated, and centroids of twenty ‘random’ egocentric neighborhoods were generated within this area. Like with self-defined neighborhoods, a proportion poverty and proportion black composition was calculated for each of these twenty egocentric neighborhoods. From these, a local mean and standard deviation was then calculated. Finally, a z-score was calculated for each respondent’s self-defined neighborhood within the distribution of the twenty egocentric neighborhoods. For example, a respondent’s self-defined neighborhood proportion poverty z-score = (self-defined neighborhood proportion poverty – mean proportion poverty of twenty egocentric neighborhoods) / the standard deviation of the proportion poverty of the twenty egocentric neighborhoods. This z-score measures how extreme a respondent’s self-defined neighborhood is in composition with respect to the composition of the local area. A score of zero indicates that the respondent’s self-defined neighborhood poverty composition is equal to the mean proportion poverty in the local area; positive scores indicate that self-defined neighborhoods are relatively higher in proportion poverty than the local area, and negative scores indicate that self-defined neighborhoods are relatively lower in proportion poverty than the local area.

Results for analyses of these compositional z-scores are presented in the third and fourth columns of Table 3. Instead of using the respective census tract compositional characteristic, we use the mean proportion of the respective compositional characteristic as a predictor in order to see how extreme a respondent’s self-defined neighborhood is in composition with respect to the average composition of the local area. The first model indicates that local area proportion poverty is positively associated (b = 1.08, p < .001) with the z-score of one’s self-defined neighborhood. This relationship is expressed graphically in the top panel of Figure 3. A similar pattern is evident regarding neighborhood portion black (b = .924, p < .001), and is illustrated graphically in the bottom panel of Figure 3. Considering results of our compositional analyses in sum, we find that residents bound considerably less segregated neighborhoods than is suggested by census geography, but residents still have a preference toward compositional homophily when considering the more general area surrounding the immediate home environment.

Figure 3.

Postestimation Predictions for the Respective Self-defined Neighborhood Compositional Z-score at levels of the Respective Size-adjusted Local Area Compositional Characteristic.

SENSITIVITY ANALYSES

Given the significance of self-defined neighborhood size to our analyses of location distances, we conducted additional neighborhood composition analyses examining the extent to which our results are dependent on self-defined neighborhood size. Specifically, in the first model of Table 3 we included an interaction between tract proportion poverty and neighborhood size, and in the second model we included an interaction between tract proportion Black and neighborhood size. In both of these sensitivity models (not shown) the interactions were statistically significant. Subsequent examination of the average marginal effects of the respective tract compositional characteristics indicates nulls relationships between the 1^st and 5^th percentile of neighborhood size, but statistically significant and progressively increasing negative slopes after the 5^th percentile of neighborhood size. These results indicate that our reported compositional patterns referencing census tracts hold for all but the smallest self-defined residential neighborhoods.

We similarly examine interactions between local area compositional characteristics and neighborhood size in supplemental analyses of those presented in the third and fourth columns of Table 3. We turn first to results for z-score neighborhood proportion poverty. The interaction between local area proportion poverty and neighborhood size was statistically significant, and examination of the average marginal effects of the interaction indicates a null relationship between local area proportion poverty and z-score neighborhood proportion poverty for neighborhoods below the 10^th percentile of neighborhood size, but a statistically significant and progressively increasing negative slopes after the 10^th percentile of neighborhood size. A similar pattern is evident regarding the interaction between local area proportion Black and self-defined neighborhood size for z-score neighborhood proportion Black. Average marginal effects of the interaction indicate a null relationship between local area proportion Black and z-score neighborhood proportion Black for neighborhoods below the 5^th percentile of neighborhood size, but a statistically significant and progressively increasing negative slopes after the 5^th percentile of neighborhood size. Taken together, these additional analyses corroborate that the average neighborhood compositional results presented in Table 3 hold for all but the smallest self-defined residential neighborhoods.

DISCUSSION

Multidisciplinary investigation of how urban residents experience their neighborhood environments spans multiple centuries and has spurred a dynamic literature (Du Bois 1899; Wirth 1938; Sampson et al. 1997; Chetty et al. 2016). The convenience of census data, however, has resulted in reflexive use of census geography to proxy neighborhoods, with limited attention to the correspondence between administrative units and the lived experience of urban residents. In response, contemporary urban researchers have called for incorporation of urban residents’ activity patterns into neighborhood research (Kwan 2009; Sperling 2012; Matthews and Yang 2013; Browning and Soller 2014; Jones and Pebley 2014; Wang et al. 2018). A variety of alternative activity-informed neighborhood measurement approaches have been proposed, but mutual consideration of residents’ own residential neighborhood boundary perceptions and activity location types is largely absent from the literature (Sastry et al. 2002; Lee et al. 2008; Hipp and Boessen 2013; Jones and Pebley 2014; Hasanzadeh et al. 2017; Burdick‐Will 2018; Pratt et al. 2019). This exclusion is important given that neighborhood perceptions have long been recognized as central to how residents’ relate to one another and ultimately mobilize in defense of local spaces (Park and Burgess 1925; Suttles 1972; Hunter 1974; Gieryn 2000; Sampson 2012).

In this paper, we first assessed whether the boundaries of self-defined residential neighborhoods—as constructed by the AHDC design—are in closer proximity to the coordinates of a range of activity location types compared to the boundaries of residents’ census tracts and egocentric neighborhoods. We find that the boundaries of egocentric and census-based neighborhoods are in similar proximity to residents’ activity locations. More importantly, however, we offer at best ambiguous evidence that residents’ own self-defined residential neighborhoods are in closer proximity to the coordinates of routine activity types compared to census tracts. On the one hand, residents’ self-defined residential neighborhood boundaries tend to be relatively closer to the coordinates of their grocery stores, schools, libraries, restaurants, and neighborhood organizations than are the boundaries of their census tract. Still, the coordinates of workplaces, churches, and relative’s houses on average tend to be closer to the nearest census tract boundary, and no differences between residents’ self-defined residential neighborhoods and census tracts are observed for the coordinates of civic organizations, friends’ houses, or ‘other’ stores and businesses. Moreover, the extent to which self-defined residential neighborhoods comparatively align with the coordinates of activity locations is largely dependent on their size, underscoring that self-defined residential neighborhood boundaries are unreliably based upon residents’ routine activity locations (Coulton et al. 2013). Overall this finding points to a division in how residents conceive of their residential neighborhood and greater activity spaces (Jones and Pebley 2014).

The evidence for a division between residential neighborhood and activity space foregrounds the significance of our compositional analyses, highlighting that the divergence of self-defined residential neighborhood space from census tracts is unlikely to simply be a function of compositional heterogeneity in activity patterns. Furthermore, the limited evidence for alignment between self-defined residential neighborhoods and activity locations suggests that self-defined neighborhood compositions are likely distinct from residents’ broader exposures to social segregation in daily mobility patterns (Krivo et al. 2013; Jones and Pebley 2014; Browning et al. 2017). Thus, in the view of the substantial literature on processes contributing to urban residents’ mental maps, self-defined neighborhoods are likely a product of residents’ social-symbolic spatial ties more so than of the locations of routine activities, and are residentially meaningful in their own right (Park and Burgess 1925; Lynch 1960; Suttles 1972; Downs and Stea 1974; Haney and Knowles 1978; Guest and Lee 1984; Krysan 2002; Campbell et al. 2009). Overall, residents bound significantly more socioeconomically and racially diverse residential neighborhoods than is represented by oft-utilized census geography, offering a less segregated image of the city through the eyes of residents. Specifically, residents of both low-poverty, low-proportion Black tracts as well as high-poverty, high- proportion Black tracts bound relatively more socioeconomically and racially diverse neighborhoods. Though this pattern underscores the major relevance of residents’ own conceptions of their neighborhoods to our understanding of residential segregation patterns, we contend that a comparison with overly homogenous census geography likely represents an incomplete test of our homophilous neighborhoods hypothesis (Sperling 2012). Thus, in our final analyses presented we consider the position of one’s own residential neighborhood composition within the distribution of the composition of the local, self-defined neighborhood size-adjusted area. These analyses reveal that, though residents’ self-defined residential neighborhoods are considerably more diverse than their administrative census tracts, residents’ still bound neighborhoods with a preference toward homophily.

Our hypotheses and analyses focus solely on demographic compositions of neighborhoods, but we expect future comparative considerations drawing on residents’ self-defined neighborhoods to be further enlightening. For example, not considered here is how residents’ neighborhoods are bounded with respect to the local prevalence of crime, the quality of local institutions, and emergent social processes such as social disorganization (Haney and Knowles 1978; Guest and Lee 1984). Though qualitative evidence indicates that residents consider such processes in bounding their neighborhoods, potential insights from aggregated patterns are largely absent from the literature (Coulton et al. 2001; Campbell et al. 2009; Iossifova 2015). Indeed, though the literature on ecometrics has advanced our understanding of the spatial distribution of social processes, considering precisely the areas that survey participants are reporting on may be especially informative (Raudenbush and Sampson 1999; Mujahid et al. 2008; O’Brien, Sampson, and Winship 2015).

To this end, further investigation of the drivers of size of residents’ self-defined neighborhoods is also likely to be important for future research (Coulton et al. 2013). Indeed, though our sensitivity analyses indicate that conclusions regarding neighborhood compositions are not due to variation in self-defined neighborhood size, size is central to our conclusions regarding self-defined residential neighborhoods and activity locations. The significance of self-defined neighborhood size ultimately overshadows our results regarding activity location types, but we still expect future studies focused on relationships between specific activity location types and self-defined neighborhood boundary placement to be illuminating (e.g. Burdick-Will 2018).

The literature on neighborhood boundary perceptions assumes that self-reported neighborhoods of all sizes are largely bounded purposefully, reflecting residents’ urban experiences (Lynch 1960; Downs and Stea 1974; Guest and Lee 1984; Pratt et al. 2019). Still, we recognize the potential for our results regarding neighborhood composition, especially, to be due to randomness in residents’ reported neighborhood boundaries and a ‘regression toward the city-level compositional mean.’ We thus encourage additional research on the origin and extent of measurement error in self-defined neighborhood boundary identification, especially those employing longitudinal designs. At minimum, future studies collecting data on residents’ self-defined neighborhoods should employ longitudinal designs to determine consistency of respondents’ self-defined residential neighborhood spaces over the course of even short periods of time. Finally, while the present study is focused explicitly on residential or home self-defined neighborhoods, it is important to acknowledge that recent studies of self-defined neighborhoods indicate residents are readily able to bound multiple and non-contiguous self-defined neighborhoods (Wiehe et al. 2008; Spilsbury et al. 2009; Kwan 2012; Pratt et al. 2019). Accordingly, we urge future studies investigating patterns and perceptions of racial and socioeconomic segregation, especially, to attempt to capture these potentially significant phenomena.

CONCLUSION

This project contributes to the existing neighborhood literature in three key ways. First, we find limited evidence that the boundaries of self-defined residential neighborhoods tend to be in closer proximity to the coordinates of self-reported activity location types compared to census tract and egocentric neighborhood boundaries. This points to the significance of social-symbolic affiliative processes to neighborhood boundary perceptions and highlights the distinctness of residents’ conceptions of residential neighborhoods from their activity spaces. Second, we find that when residents themselves are given the opportunity to define their residential neighborhood space, they collectively bound a considerably less segregated city than is illustrated by census geography. When considering the more general local area, however, compositions of residents’ self-defined neighborhood still reflect a preference toward homophily. In sum, these findings 1) underscore the contributions of residents’ neighborhood boundary perceptions to our understanding of the socially constructed nature of residential segregation, and 2) motivate the potential for investigations of emergent urban social processes to be enriched by incorporating residents’ definitions of their neighborhoods.

Appendix A. Data Cleaning and Sample Construction

In order to assess the quality and completeness of the neighborhood boundary data collected during the caregiver survey, we started by constructing a convex hull polygon formed by the four boundary points named by the respondents. Along with inspecting the text name compared to the corresponding latitude/longitude saved by the Google Maps API for congruence, we also flagged cases with convex hull polygons that indicated potential errors in the intersection identification process. As a result of this assessment, we flagged 11% of points to be cleaned for the following reasons: boundary streets given instead of intersections (6%), intersection text written without corresponding latitude/longitude due to interviewer error (4%), and misspellings and/or incomplete road names (1%). Our cleaning process corrected 61% of the flagged points by creating intersections from the boundary streets, geocoding text that was missing latitude and longitude, and correcting typographical errors.

We excluded 48 of 1,402 respondents from the polygon generation process who had less than 3 unique intersection points, largely due to non-response on all (N=25) or part of the questions that generate neighborhood boundaries (N=15), and caregivers who reported duplicated intersection points (N=8).

From the 1,354 caregivers with 3 or more self-defined boundary intersections 72 did not result in a closed polygon. Reasons for this include boundary road names that did not intersect, points dropped by a pin on the Google Maps that were not near enough to a road network intersection to converge into a polygon when joined with other line segments, and flagged points that were not able to be cleaned.

Of the 1,282 polygons created, caregivers were excluded from analysis when the entire polygon resulting from the road network path was a small sliver of near-zero area (N=83) or a straight line (N=45), due to a combination of points being very close to one another, all on the same road, or to the network path drawn between points that overlapped back on the previous route. We exclude one caregiver who used suburban city names as the neighborhood boundary points, which resulted in a self-defined neighborhood polygon that was 182 square miles in area and included a majority of the city of Columbus metropolitan boundaries, and one caregiver whose self-reported locations were all more than 30,000 meters away from the self-defined neighborhood boundary. As a result, the final analytic sample includes 1,152 caregivers. Sensitivity analyses including all 1,282 residential neighborhood polygons yields substantively identical conclusion, however.