Nguyen Luu Bao Doan - Evidence of the impacts of urban sprawl on social capital

By summing the coefficient for LNCOMPACT present in the race, employment status, income-brackets, and child-status components with that present in the county mean social-capital componen[r]

Introduction

Sprawl has become one of the most discussed topics that draws the attention of scholars from different fields such as planning, public health, economics, and sociol-ogy Certain practices and planning policies have been based on the perception of the effects of sprawl If urban sprawl has negative impacts, planners should implement strategies to curb leapfrog suburbanization across the country Otherwise, if urban sprawl in fact does not have negative impacts, the urban growth versus suburban growth debate should be reexamined thoroughly

However, empirical studies of urban sprawl provide opposing evidence of the effects of sprawl with respect to social capital Findings from those studies may give some answers to our question, but due to a lack of social capital and sprawl data, those answers are not complete As there is a growing need to build communities that promote social and health welfare, planners need more rigorous studies to help them reach agreement on how to achieve it (Freeman, 2001) The purpose of this study is to reexamine the relationship between sprawl and social capital by improving the method used in previous studies and taking advantage of the new measures of urban sprawl and social capital This analysis, in the context of American metropolitan areas, teases out impacts of person characteristics and place characteristics on individual social capital by adopting hierarchical or multilevel modeling techniques

Background

Urban sprawl, defined as noncontiguous commercial and residential developments with low density and large separation, has been at the center of a debate among planning practitioners and scholars in recent years (Burchell et al, 1998; 2002; Ewing, 1997; Ewing et al, 2002; Galster et al, 2001; Weitz and Moore, 1998) Sprawl opponents believe that owing to sprawl's connection to greater travel distances and increased automobile dependence, it is responsible for the reduction in quality of life on several aspects They have argued that sprawl led to an increase in vehicle-miles traveled and traffic congestion (Bento et al, 2005; Burchell et al, 1998; 2002; Cervero and Wu, 1998; Ewing, 1994), energy consumption (Anderson et al, 1996; Ewing, 1997), social segregation (Burchell et al, 1998; Ewing, 1997), and physical inactivity and obesity Evidence of the impacts of urban sprawl on social capital

Doan Nguyen

School of Architecture, Planning, and Preservation, University of Maryland, College Park, MD 20742, USA; e-mail: dnguyen@ursp.umd.edu

Received 19 September 2008; in revised form 12 August 2009; published online 17 May 2010

(Cervero and Duncan, 2003; Ewing et al, 2003), and the depletion of land and air quality (Anderson et al, 1996; Burchell et al, 1998; 2002)

A number of researchers have contended that sprawl has negative impacts on our physical resources as well as on our social resources, or social capital According to Putnam (Putnam, 2000; Putnam et al, 1993), social networks have value, and the network or connection among individuals provides members of the network with actual and potential benefits He blamed sprawl and suburbanization among other major causes such as time pressure, technological advances (television and the Inter-net), and generational change for the decline of social capital in the US As people live farther away from each other, sprawl weakens linkages between neighbors (Burchell et al, 1998) Jacobs (1961) suggested that compact urban areas were likely to promote social interaction and building compact communities with high density should be one of the important planning guidelines New Urbanists have adopted this view and sought to create high-density and walkable communities as an ideal approach to develop suburban areas (Freeman, 2001; Hayward, 1998)

A number of empirical studies indicate that compact and walkable neighborhoods promote some form of social capital In his 2001 study, Freeman successfully estab-lished a negative relationship between the level of car usage and the level of social ties in neighborhoods by using a binomial logistic regression model Leyden (2003) found that high neighborhood walkability resulted in residents' higher likelihood of political participation, of knowing their neighbors, and of social interaction However, both studies fail to address the problem of dependence among people residing in the same neighborhood: they are not random, independent observations This limitation results in coefficient biasness in both studies Meanwhile, a number of scholars argue that compactness does not alleviate problems caused by sprawl nor is it better than sprawl in addressing issues such as improving social interaction and reducing commute time (Audirac et al, 1990; Glaeser and Kahn, 2003; Gordon and Richardson, 1997; 2000; Hayward, 1998; Kahn, 2000; 2001; 2006) Gordon and Richardson (1997) argued that changes in land-use policy to favor compactness would violate the principles of the free market and would be deemed unnecessary They rejected that urban sprawl would lead to increased automobile usage and traffic congestion

Along this line, scholars criticizing compact development have shown empirical evidence to demonstrate that urban sprawl does not undermine social capital By using DDB Needham Lifestyle Survey data, Glaeser and Gottlieb (2006) found that central city residence decreases four types of social-capital activity such as attending church, working on a community project, contacting a public official, and being a registered voter In a study using the Social Capital Benchmark Survey data (The Roper Center, 2005), Brueckner and Largey (2006) regressed individuals' social-interaction variables on census-tract density among other variables; they found that high density was negatively related to all friendship and group-involvement variables Even though those scholars used different approaches and different datasets, neither of those studies addressed the issues of the hierarchical nature of their data and both used population density as a proxy for urban form Nevertheless, their findings apparently support the notion that compactness, not sprawl, is the culprit of the decline of social capital that Putnam (2000) lamented in his book

fail to account for the effect of poor street connectivity, which also distinguishes compactness from sprawl

Most empirical studies that look at the effects of sprawl on social capital examine some of its factors such as trust and neighborhood ties (Brueckner and Largey, 2006; Freeman, 2001; Glaeser and Gottlieb, 2006; Leyden, 2003) A few studies in this group are likely to overlook nonneighborhood social ties As Guest and Wierzbicki (1999) indicated, social interactions with other nonlocal groups of people could substitute for neighborhood social ties In his book, Putnam suggested that the range of activities embodying social capital includes civic participation, religious participation, and polit-ical participation More importantly, those studies that use regression analysis of individual social capital and a sprawl variable measured at some geographical scale not take into account the hierarchical nature of the data and thus might produce biased estimates

This study improves the common limitation of previous studies by using hierarch-ical modeling to address the problem of dependence among people residing in the same neighborhoods and by using Ewing's sprawl index to tease out spatial impacts on a variety of social-capital factors

As a result of increasing sprawl, space between residential and commercial devel-opment increases, forcing people to drive more and to divert their time away from social-interaction activities This study tests the hypothesis that urban sprawl plays certain roles in the relationship between individual characteristics and social capital Some scholars suggested that there is a positive relationship between income and certain aspects of social capital such as social interaction or voting habits (Brueckner and Largey, 2006; Glaeser and Gottlieb, 2006) In the light of this, urban sprawl can amplify the effects of income on social capital such that in more sprawled areas, the effects of income on social capital are greater as income increases due to the fewer choices of transportation for the lower-income groups There is a possible interaction between urban form and race such that urban sprawl lessens the effects of different racial groups on social capital since longer commutes reduce interaction opportunities for all racial groups

Data

Social-capital variables came from the 2000 Social Capital Community Benchmark survey data (The Roper Center, 2005) The survey was conducted by the Saguaro Seminar at Harvard University's Kennedy School of Government between July and November 2000.(1)The effective sample size for the analysis is 22191.

The survey provides nine social-capital variables to capture civic engagement, political participation, and social interaction on the basis of a person's answers to the survey questionnaire Those variables include social trust, diversity of friendship, the number of group involvements, informal social interaction, organized-group inter-action, faith-based social capital, giving and volunteering, nonelectoral participation, and electoral politics (see table for a detailed description of dependent variables)

The survey data also provide individual weight and socioeconomic and demo-graphic information of respondents The weight accounts for the population distribution in the sample and for the odds of selection for any household in the sample Some or all of those person characteristics have been included in prior studies on social capital by Brueckner and Largey (2006), Campbell and Lee (1992), Freeman (2001), Glaeser and

(1)The Roper Center, University of Connecticut manages and disseminates both restricted and

Gottlieb (2006), Glaeser et al (2000), Guest and Wierzbicki (1999), Iyer et al (2005), and Leyden (2003)

The restricted version of the Social Capital Benchmark Survey data includes counties and census tracts where the respondent resided, which allows for the deriva-tion of three variables to capture mean household income, the percentage of college graduates, and racial diversity at the census-tract level Freeman (2001) indicated that the level of poverty in a neighborhood negatively affects the social capital of its residents Thus, the census-tract mean household income captures any effect of income on social capital In addition, since living in a neighborhood with higher education Table Definition and measures of social-capital variables

Variable Definition

SOCTRUST Social trust: combines answers to questions of levels at which respondents trust different groups in society (people in the neighborhood, work colleagues, people at church or places of worship, people working in stores, local police) The index is standardized with respect to the national norms

DIVRSITY Diversity of friendship: counts the number of different types of personal friends the respondents has from the 11 possible types

GRPINVLV Number of formal group involvements: counts the number of different nonreligious groups the respondent has been a member of from 18 possible types

FAITHBASED Faith-based social capital: combines answers to questions that ask participants whether they are a member of a local church or other religious community, how often they attend religious services, whether they have taken part in any activity with other people at their church or place of worship other than attending services, and whether they have any affiliation with nonchurch religious organizations Their levels of contributing and volunteering were recorded to calculate the index The index was computed as the mean of the standardized variables obtained from the answers

SCHMOOZ Informal social interaction index: mean of standardized responses to the question asking the respondent to supply the estimate of the number of times he or she has undertaken certain social activities in the past 12 months Those activities include times of playing cards with others, visiting relatives or having them visit, of having friends over, socializing with coworkers outside of work, and socializing with friend in public places

ORGINTER Organized group interactions: mean of the scores standardized against the national normal of a 3-item question It asks how many times in the past 12 months the respondent has attended (1) any public meetings in which there was a discussion of town or school affairs, (2) a club meeting, and (3) a celebration, parade, or an event in his or her community

CHARITY Giving and volunteering: combines reversed polarity versions of volunteering for different types of organizations: arts, health-related, neighborhood, religious, youth groups, and those which help the poor or elderly; the total number of times volunteered, and contributions to secular charities and religious causes

PROTEST Nonelectoral political participation: signing petitions, attending political meetings or rallies, joining in any demonstrations, protests, boycotts, or marches; also, involvement in local reform efforts, membership of political groups, ethnic, national, or civil-rights groups, or labor unions ELECPOL Electoral politics: combines past voting, voter registration, interest in

attainment might raise one's opportunities to interact and network, the percentage of college graduates is another important census-tract variable

Putnam (2000) suggested that racial diversity could negatively affect the bridging social capital, which is needed to build solidarity among members of a community In their experimental analysis, Glaeser et al (2000) gave evidence of association between lower levels of trustworthiness and high racial diversity Kahn and Costa (2002) documented empirical evidence from fifteen studies to show lower social capital in more heterogeneous communities by calculating the Simpson's index of diversity Table Definition and descriptive statistics of independent and dependent variables

Variable name Description Mean Min Max

Level descriptive statistics (sample size: 22191)

WEIGHT Sampling weight

BLACK ˆ for Black respondent ASIAN ˆ for Asian respondent HISPN ˆ for Hispanic respondent

AGE Age of respondent

GENDER ˆ for male respondent

OWN ˆ for homeowner

MARITAL ˆ for married respondent

SMCOLL ˆ for respondent with some college COLGD ˆ for respondent with college degree LIVCOM < 1year ˆ if lived in community < year LIVCOM1-5years ˆ if lived in community ± years LIVCOM6-10years ˆ if lived in community ± 10 years EMP ˆ if respondent employed

INCOME0 ˆ for undisclosed income IN30-50K ˆ for annual household income

30 ± 50 K

IN50-75K ˆ for annual household income 50 ± 75 K

INover75K ˆ for annual household income over 75 K

KIDS0 ˆ for undisclosed children situation KIDS1 ˆ for having children 17 years of age

or younger in household SOCTRUST Social trust

DIVRSITY Diversity of friendship

GRPINVLV Number of group involvements FAITHBASED Faith-based social capital SCHMOOZ Informal social interaction ORGINTER Organized group interaction CHARITY Giving and volunteering

PROTEST Nonelectoral political participation ELECPOL Electoral politics

Level descriptive statistics (sample size: 6436)

INCOMETR Mean household income in census trust 34 325 150 001 COLGRAD Percentage of college graduates in

census tract 15.6 69.16

RACDIVER Racial-diversity index in census tract 0.2 0.81 Level descriptive statistics (sample size: 259)

RURAL Percentage of county residents in rural

areas 36.63 100

LNCOMPACT Natural log of county-sprawl index 4.57 4.21 5.86

(Keylock, 2005) for racial and birthplace diversity Similarly, in this study the Simpson's index of diversity was used to measure the heterogeneity in racial composition in a census tract For each census tract, the index is constructed as follows Census-tract racial diversity

D ˆ ÿXS

i ˆ

n

i

N 2

,

where n represents the number of individuals of race i in a census tract (i ˆ Black, Asian, White, or Hispanic), S is the number of races, and N is the total population sample size in the census tract The index has value from to and represents the probability that two individuals, randomly selected from a specific tract, are from different racial groups Therefore, as the value of D approaches one, diversity increases The demographic data came from the 1990 Population Census for White, Black, Hispanic, and Asian (http://www.census.gov/population/www/documentation/ twps0056/twps0056.html)

At the county level, the county sprawl index that Ewing et al (2003) developed in their study of urban sprawl impacts on health was used The index is composed of residential density and street accessibility More compact urban form implies higher gross and net population density and a higher percentage of people living in high densities Street accessibility is defined in terms of the length, and size of blocks (in square miles) The length of each side of a block and its size in a more compact urban neighborhood should be smaller than those in a less compact suburban area with less connected cul-de-sacs and fewer alternative routes The component was transformed to a scale with a mean value of 100 and standard deviation of 25.(2)This study uses only a subset of the 259 counties and statistically equivalent entities such as independent cities The higher the value of the index, the less sprawled or more `compact' an urban county is At the county level, two or more variables including the percentage of county population living in rural areas and the county population size are deployed to control for possible effects of county sizes and of rural versus urban and suburban settings on the social capital of county residents Putnam (2000) suggested that urban and suburban settings are less conducive to social-interaction activities This possibility will be tested in this study Table presents descriptive statistics for all variables Method

The hierarchical-modeling technique is adopted to examine the possible impacts of urban form on social capital Conceptually, the hierarchical model differs from multi-variate regression models in several aspects, but most importantly, the former has more than one error term in its equation to account for errors at different levels of analysis

Hierarchical modeling takes into account the dependence between observations In this study, Social Capital Community Benchmark survey participants are clustered in census tracts in different counties When using the ordinary least squares (OLS) models, the researcher is likely to ignore the fact that people residing in the same census tract or in the same county may be related, which violates OLS assumptions that the observations are independent (Gelman and Hill, 2007; Raudenbush and Bryk, 2002; Snijders and Bosker, 1999), leading to standard errors that are too small As a result, the researcher is more likely to reject the null hypothesis than in the case where hierarchical modeling is used In addition, by using hierarchical modeling, the researcher can test a realistic hypothesis that there exists some cross-level interaction

between personal characteristics and place characteristics (Luke, 2004; Raudenbush and Bryk, 2002).(3)The technique allows the researcher to disentangle and examine the complex effects of the environmental factors on an individual's outcome

In this study, the hierarchical nature of the dataset and the original research hypotheses have rendered the use of hierarchical modeling necessary To test the validity of 3-level hierarchical models, a fully unconditional model which is similar to ANOVA is conducted This fully unconditional model partitions total variance in the dependent variables into three components of variance: variance among different surveyed individuals within census tracts, variance among census tracts within counties, and variance among different counties

Using HLMß for Windows version to estimate the model, it was possible to show that variance among individuals living in the same census tracts account for most variance in the social-capital data (table 3) Even though most variability is within counties, statistically significant variance at census-tract and county levels requires the inclusion of tract-level and county-level predictors in the analysis

The model is specified as a random intercept model with nonrandomly varying slopes In level 2, the slopes have been assumed to be fixed and only the intercept has the random-effect component Variables of interest at level (census tract) include racial diversity in 1990 (RACDIVER), median household income in 1989 (INCOMETR), and the percentage of college graduates in 1990 (COLGRAD) The data are from 1990 population census To reduce complex computation due to the presence of cross-level interaction terms, it is also assumed that variables at the tract level only affect the intercepts of the set of regressions at the individual level In level 3, county population size (LNPOP), sprawl (LNCOMPACT), and the percentage of the county population living in rural areas (RURAL) are assumed to affect the intercept of the model This is equivalent to saying that models to predict social capital have intercepts varying across different counties depending on the county population size, the degree of sprawl, and the degree of urban or suburban settings of the county In addition, it is also assumed that sprawl has differential impacts on the relationship between an individual's race and income and his or her social capital because of his or her housing preferences (Rong, 2006) While it is harder to predict the direction of the relationship of race and urban sprawl with respect to social capital a priori, it is possible that high-income people increase their social capital when living in more compact counties, assuming that less sprawl means less social interaction Similarly, individuals living with children could gain more social capital by living in more compact counties, other things being constant

(3)See Raudenbush and Bryk (2002) for a complete theoretical discussion of multilevel modeling.

Table Variance analysis of 3-level hierarchical data Variance (%)

level level level

Charity (giving) 98.6 0.5 0.9

Diversity of friendship 98.4 0.7 0.9

Electoral politics 95.2 1.4 3.4

Faith-based social capital 94.9 0.7 4.4 Informal social interaction 98.8 0.4 0.8 Number of group involvements 99.3 0.4 0.3 Organized-group interaction 99.1 0.6 0.3 Nonelectoral political participation 96.5 0.9 2.5

The model specification is as follows Level (person level):

Yijk ˆ p0 jk ‡ p1 jk…BLACK†ijk ‡ p2 jk…ASIAN†ijk‡ p3 jk…HISP†ijk‡ p4 jk…EMP†ijk

‡ p5 jk…IN30-50K†ijk‡ p6 jk…IN50-75K†ijk‡ p7 jk…INover75K†ijk‡ p8 jk…KIDS1†ijk

‡ p9 jk…AGE†ijk‡ p10 jk…GENDER†ijk ‡ p11 jk…OWN†ijk‡ p12 jk…MARITAL†ijk

‡ p13 jk…SMCOLL†ijk‡ p14 jk…COLGD†ijk ‡ p15 jk…LIVCOM < 1year†ijk

‡ p16 jk…LIVCOM1-5years†ijk ‡ p17 jk…LIVCOM6-10years†ijk

‡ p18 jk…INCOME0†ijk‡ p19 jk…KIDS0†ijk‡ eijk

Level (census-tract level):

p0 jk ˆ b00 k‡ b01 k…INCOMETR†jk‡ b02 k…COLGRAD†jk ‡ b03 k…RACDIVER†jk ‡ r0 jk,

p1 jk ˆ b10 k,

p19 jk ˆ b190 k

Level (county level):

b00 k ˆ g000‡ g001…LNCOMPACT†k‡ g002…LNPOP†k‡ g003…RURAL†k‡ u00 k,

b10 k ˆ g100‡ g101…LNCOMPACT†k,

b80 k ˆ g800‡ g801…LNCOMPACT†k,

b90 k ˆ g900,

b190 k ˆ g1900 ,

where

Yijk is the social-capital-factor measure of person i in census tract j in county k;

[g000‡ g001RURAL ‡ g002LNCOMPACT ‡ g003LNPOP ‡ g010INCOMETR

‡ g020COLGRAD ‡ g030RACDIVER] is county mean social capital;

eijk is the random effect varying across different individuals;

rjk is the random effect of the intercept across different census tracts in county k;

uk is the random component varying across different counties

The social-capital survey's individual weight for each observation or surveyed person at level is applied for nine equations corresponding to nine factors of social capital There are no weights assigned to census tracts and counties Because LNCOMPACT is highly correlated with RURAL (r ˆ ÿ0:75), LNCOMPACT is centered on its grand mean at level to avoid multicollinearity Tests of normality are performed and the standard errors of coefficient estimates are robust standard errors

Findings Urban sprawl

children seventeen years old or younger, and whose household income in 1999 was less than $30 000 The degree of compactness is negatively related to informal social interaction (ÿ0:3), to faith-based social capital (ÿ0:41), and to giving and volunteering (ÿ1:13) Table Hierarchical model results (See table for definitions of variables.)

Fixed effect Diversity of Informal social Organized-group

friendship interaction interaction

coefficient t-statistics coefficient t-statistics coefficient t-statistics County mean social capital

Base 5.27* 43.0 0.73* 20.9 ÿ0.10* ÿ2.8

RURAL 4.0  10ÿ3 1.3 ÿ3.0  10ÿ4 ÿ0.5 ÿ7.0  10ÿ4 ÿ1.3

LNCOMPACT ÿ0.22 ÿ0.5 ÿ0.30* ÿ5.0 0.04 0.7

LNPOP 0.06 1.4 4.0  10ÿ4 0.0 ÿ0.01 ÿ1.8

INCOMETR ÿ8.0  10ÿ6* ÿ3.1 ÿ2.0  10ÿ6* ÿ3.3 ÿ3.0  10ÿ6* ÿ5.4

COLGRAD 0.01* 4.2 3.0  10ÿ4 0.4 2.0  10ÿ3 2.2

RACDIVER 0.68* 4.5 ÿ0.11** ÿ2.2 ÿ0.04 ÿ1.0

Racial differentiation for BLACK

Base ÿ0.28* ÿ3.4 ÿ0.14* ÿ7.2 ÿ0.01 ÿ0.2

LNCOMPACT 0.82** 2.2 ÿ0.02 ÿ0.4 ÿ0.27* ÿ2.7

Racial differentiation for ASIAN

Base ÿ1.11* ÿ5.1 ÿ0.21* ÿ3.5 ÿ0.20* ÿ3.4

LNCOMPACT ÿ0.79 ÿ1.6 ÿ4.0  10ÿ3 0.0 0.12 0.6

Racial differentiation for HISPN

Base ÿ0.91* ÿ6.5 ÿ0.29* ÿ9.2 ÿ0.06 ÿ1.8

LNCOMPACT 0.57 1.6 0.09 0.8 ÿ0.15 ÿ1.5

Employment differentiation (EMP)

Base 0.39* 7.4 ÿ0.12* ÿ8.0 ÿ0.01 ÿ0.9

LNCOMPACT 0.24 1.2 0.12 1.8 0.07 1.7

Income differentiation for IN30-50K

Base 0.51* 7.6 0.06* 4.2 0.09* 4.9

LNCOMPACT 0.37 1.8 0.10** 2.0 ÿ0.07 ÿ1.2

Income differentiation for IN50-75K

Base 0.67* 7.9 0.08* 3.9 0.12* 5.4

LNCOMPACT 0.27 0.9 0.19* 2.9 ÿ0.07 ÿ0.9

Income differentiation for INover75K

Base 0.92* 12.8 0.14* 7.8 0.21* 7.1

LNCOMPACT 0.29 1.4 0.15* 2.6 ÿ0.18 ÿ1.4

Child differentiation (KIDS1)

Base 0.001 0.0 ÿ0.02 ÿ1.4 0.12* 8.7

LNCOMPACT 0.13 0.7 0.05 1.1 ÿ0.02 ÿ0.4

AGE ÿ0.01* ÿ4.0 ÿ0.01* ÿ28.6 ÿ3.0  10ÿ3* ÿ5.7

GENDER ÿ0.03 ÿ0.6 ÿ0.05* ÿ4.4 ÿ0.02** ÿ1.9

OWN 0.09 1.2 ÿ0.04* ÿ2.3 0.02 1.8

MARITAL 0.10** 2.3 ÿ0.14* ÿ9.9 ÿ0.03* ÿ2.5

SMCOLL 0.96* 15.5 0.08* 4.6 0.15* 11.0

COLGD 1.08* 19.5 ÿ2.0  10ÿ3 ÿ0.1 0.29* 17.2

LIVCOM<1year ÿ0.24* ÿ3:0 ÿ0.09* ÿ4.1 ÿ0.12* ÿ5.4

LIVCOM1-5years ÿ0.31* ÿ5.3 ÿ0.09* ÿ5.6 ÿ0.11* ÿ8.1

LIVCOM6-10years ÿ0.13** ÿ2.3 ÿ0.04* ÿ2.4 ÿ0.03** ÿ2.1

INCOME0 0.33* 4.5 0.06* 2.9 0.07* 4.3

KIDS0 0.77 1.3 ÿ0.10 ÿ1.3 ÿ0.13 ÿ1.1

However, it is positively related to electoral politics (0.38) and nonelectoral political participation (0.79) These findings confirm recent findings of Glaeser and Gottlieb (2006) concerning faith-based social capital and social interaction, and findings by Brueckner Table Hierarchical model results (See table for definitions of variabless.)

Fixed effect Number of Faith-based Social trust

group involvements social capital

coefficient t-statistics coefficient t-statistics coefficient t-statistics County mean social capital

Base 0.70* 4.9 ÿ0.61* ÿ17.4 ÿ0.45* ÿ12.3

RURAL ÿ2.0  10ÿ3 ÿ0.8 ÿ3.0  10ÿ3** ÿ2.3 ÿ9.0  10ÿ4 ÿ1.2

LNCOMPACT 0.22 1.2 ÿ0.41* ÿ2.6 ÿ0.19 ÿ1.7

LNPOP ÿ0.01 ÿ0.5 ÿ0.03 ÿ1.6 ÿ0.02 ÿ2.0

INCOMETR ÿ7.0  10ÿ6* ÿ2.8 4.0  10ÿ6* 3.7 2.0  10ÿ6* 2.6

COLGRAD 9.0  10ÿ3* 2.9 ÿ3.0  10ÿ3* ÿ2.7 4.0  10ÿ3* 7.4

RACDIVER ÿ0.06 ÿ0.4 8.0  10ÿ4 0.0 ÿ0.19* ÿ4.4

Racial differentiation for BLACK

Base 0.81* 8.2 0.23* 9.0 ÿ0.45* ÿ20.1

LNCOMPACT ÿ0.77 ÿ1.7 ÿ0.14 ÿ1.0 0.13 1.4

Racial differentiation for ASIAN

Base ÿ0.33 ÿ1.7 ÿ0.07 ÿ0.7 ÿ0.25* ÿ4.2

LNCOMPACT ÿ0.04 ÿ0.1 0.11 0.8 0.32 1.6

Racial differentiation for HISPN

Base ÿ0.17 ÿ1.6 0.02 0.6 ÿ0.40* ÿ13.4

LNCOMPACT ÿ0.27 ÿ1.1 0.23 1.8 0.26** 2.0

Employment differentiation (EMP)

Base 0.10** 2.1 ÿ0.06* ÿ4.2 0.02 2.0

LNCOMPACT ÿ0.01 ÿ0.1 0.01 0.2 0.03 1.1

Income differentiation for IN30-50K

Base 0.47* 7.3 0.11* 7.4 0.09* 4.2

LNCOMPACT ÿ0.15 ÿ0.6 ÿ0.16 ÿ1.5 0.01 0.1

Income differentiation for IN50-75K

Base 0.63* 8.0 0.12* 7.0 0.11* 5.8

LNCOMPACT ÿ0.38 ÿ1.0 ÿ0.16* ÿ2.6 ÿ0.06 ÿ1.0

Income differentiation for INover75K

Base 1.08* 12.2 0.16* 8.5 0.11* 4.7

LNCOMPACT ÿ0.61** ÿ2.0 ÿ0.25* ÿ3.4 ÿ0.04 ÿ0.9

Child differentiation (KIDS1)

Base 0.38* 8.4 0.10* 8.7 ÿ0.02** ÿ1.9

LNCOMPACT 0.14 0.6 0.17* 4.4 0.01 0.2

AGE 0.02* 7.5 0.01* 10.2 7.0  10ÿ3* 13.6

GENDER 0.15* 3.1 ÿ0.11* ÿ9.7 ÿ0.07* ÿ6.3

OWN 0.19* 3.1 0.09* 6.1 0.08* 5.4

MARITAL ÿ0.06 ÿ1.2 0.14* 12.0 0.08* 6.6

SMCOLL 0.93* 17.4 0.18* 14.3 0.13* 9.8

COLGD 1.68* 22.1 0.28* 13.7 0.27* 22.1

LIVCOM<1year ÿ0.57* ÿ7.1 ÿ0.16* ÿ6.2 ÿ0.02 ÿ1.1

LIVCOM1-5years ÿ0.34* ÿ5.5 ÿ0.12* ÿ8.5 ÿ0.03* ÿ2.7

LIVCOM6-10years 0.00 0.0 ÿ0.06* ÿ3.3 ÿ0.03** ÿ2.1

INCOME0 0.29* 5.1 0.10* 5.1 0.04 1.7

KIDS0 0.11 0.2 0.08 0.5 0.07 0.5

and Largey (2006) concerning social interaction that more compact residence leads to less social capital However, similar to Leyden's results (2003), the findings provide evidence that compact urban form supports political participation as suggested by Putnam (2000) Table Hierarchical model results (See table for definitions of variables.)

Fixed effect Giving and volunteering Electoral politics Nonelectoral political participation

coefficient t-statistics coefficient t-statistics

coefficient t-statistics County mean social capital

Base 1.82* 9.5 1.00* 17.5 0.40* 5.6

RURAL ÿ0.01** ÿ2.2 ÿ3.0  10ÿ3 ÿ2.0 3.0  10ÿ3 2.2

LNCOMPACT ÿ1.13** ÿ2.6 0.38** 2.0 0.79* 3.5

LNPOP ÿ0.06 ÿ1.0 ÿ0.09* ÿ3.2 1.0  10ÿ3 0.0

INCOMETR 1.0  10ÿ6 0.3 ÿ4.0  10ÿ6* ÿ3.2 ÿ7.0  10ÿ6* ÿ8.3

COLGRAD 0.01** 2.5 0.01* 8.5 0.01* 5.1

RACDIVER 0.21 1.1 ÿ0.10 ÿ1.7 0.06 0.7

Racial differentiation for BLACK

Base 0.36* 2.7 0.02 0.6 0.20* 4.4

LNCOMPACT ÿ0.53 ÿ1.2 ÿ0.07 ÿ0.4 ÿ0.18 ÿ1.1

Racial differentiation for ASIAN

Base ÿ1.23* ÿ3.1 ÿ0.73* ÿ6.7 ÿ0.21 ÿ2.4

LNCOMPACT 1.02 0.8 0.15 0.6 ÿ0.77 ÿ4.0

Racial differentiation for HISPN

Base ÿ0.72* ÿ4.2 ÿ0.58* ÿ8.7 ÿ0.06 ÿ0.9

LNCOMPACT 0.32 0.6 ÿ0.28** ÿ2.1 ÿ0.29 ÿ1.4

Employment differentiation (EMP)

Base 3.0  10ÿ3* 0.0 0.06* 3.4 0.11* 4.5

LNCOMPACT 0.17 0.4 0.27* 4.6 0.15 1.6

Income differentiation for IN30-50K

Base 1.04* 11.4 0.21* 5.6 0.15* 4.7

LNCOMPACT ÿ0.95** ÿ2.0 0.16 1.5 0.25 1.6

Income differentiation for IN50-75K

Base 1.39* 14.1 0.34* 10.8 0.28* 6.0

LNCOMPACT ÿ0.60 ÿ1.4 ÿ0.24** ÿ1.9 0.32 1.2

Income differentiation for INover75K

Base 2.58* 19.3 0.44* 11.4 0.35* 7.2

LNCOMPACT ÿ1.49* ÿ3.0 ÿ0.32 ÿ1.6 0.12 0.9

Child differentiation (KIDS1)

Base 0.57* 9.1 ÿ0.06* ÿ2.9 0.04 1.3

LNCOMPACT 0.56** 2.4 ÿ0.07 ÿ0.7 ÿ0.18 ÿ1.4

AGE 0.02 5.4 0.03* 36.9 2.0  10ÿ4 0.2

GENDER ÿ0.44* ÿ7.6 0.14* 6.7 0.10* 5.2

OWN 0.64* 7.3 0.26* 11.7 0.09* 3.5

MARITAL 0.42* 4.5 0.13* 5.5 ÿ0.06* ÿ2.7

SMCOLL 1.45* 17.6 0.53* 21.0 0.40* 12.4

COLGD 2.51* 25.7 0.88* 34.5 0.75* 18.4

LIVCOM<1year ÿ0.89* ÿ6.3 ÿ0.40* ÿ10.3 ÿ0.25* ÿ6.4

LIVCOM1-5years ÿ0.63* ÿ7.1 ÿ0.27* ÿ12.7 ÿ0.18* ÿ7.4

LIVCOM6-10years ÿ0.13 ÿ1.5 ÿ0.13* ÿ4.1 ÿ0.04 ÿ1.2

INCOME0 ÿ0.32* ÿ2.9 0.12* 3.5 0.05 1.6

KIDS0 ÿ1.04 ÿ1.4 0.30 1.1 ÿ0.12 ÿ0.6

By summing the coefficient for LNCOMPACT present in the race, employment status, income-brackets, and child-status components with that present in the county mean social-capital component in each equation and testing for significance, one can comment on the overall impacts of urban form on different social capital factors for different household groups Table summarizes those findings with statistical significance p < 0:01

For most household types, urban compactness has negative impacts on social interaction, faith-based social capital, and giving and volunteering; meanwhile, it has positive impacts on electoral politics and nonelectoral political participation

The following sections discuss in detail the impacts of sprawl on the relationship between different household types and social capital, with respect to the county average The impacts of sprawl and race

Regarding the effects of race on social capital, the findings indicate that respondents belonging to different racial groups have different degrees of social capital; and urban sprawl can increase the social-capital gaps among different racial groups Compared to white Americans, African American participants take part in more orga-nizations, have higher faith-based social capital, more volunteering, join more nonelectoral political activities such as protests, marches, and demonstrations, and join more political groups Nevertheless, they have less social trust and less informal social interaction The social-capital gap between white Americans and African Americans, however, is only influenced by urban sprawl in the case of the diversity of friendship [ÿ0:28(1) ‡ 0:82LNCOMPACT] and the participation in club meetings, community meetings, and events (ÿ0:27LNCOMPACT) In the former case, any practical levels of compactness will lead to higher numbers of friends from different backgrounds for African Americans and they will increase their network at a faster rate as sprawl decreases In the latter, African Americans participate less in organized-group activities compared with white Americans but more sprawl narrows the gap

Compared with white respondents, Asian respondents have less diverse friendship and social trust, participate less in informal social interaction, and also participate less in volunteering or giving and politics Hispanic respondents share patterns similar to the Asian population As far as the role of urban form is concerned, sprawl does not have amplifying or attenuating effects on the relationship between being Asian and Table Overall effects of LNCOMPACT on social capital for different household types (See table for definitions of variables.)

For For For For For For For For

BLACK ASIAN HISPN employed IN30-50K IN50-75K INover75K KIDS1

Informal social ± ± ± ± ±

interaction

Organized group ± interaction

Faith-based ± ± ± ±

social capital

Giving and ± ± ± ±

volunteering

Electoral politics ‡ ‡

Nonelectoral ‡ ‡ ‡ ‡ ‡ ‡

political participation

social capital However, the Hispanic population has higher social trust in more compact counties [ÿ0:40(1† ‡ 0:26LNCOMPACT] Even in the most sprawled county of Jasper (IN) in the dataset (LNCOMPACT ˆ 4:21), the Hispanic social-trust index exceeds that of the white population by 0.69 However, compact living conditions make the Hispanic population fall further behind the white population in electoral political participation [ÿ0:58(1) ÿ 0:28LNCOMPACT]

The impacts of sprawl and employment status

Being employed is correlated positively with diversity of friendship and the number of group involvements An employed person is also more likely to participate in civic activities such as giving and volunteering, and has high political participation, compared with the group consisting of students, retirees, people who are between jobs, and the unemployed; however, the faith-based social capital of an employed person and his or her degree of informal social interaction are less Urban com-pactness only has amplifying effects that increase the gap between the employed and nonemployed groups of people with respect to electoral-politics participation [0:06(1) ‡ 0:27LNCOMPACT]

The impacts of sprawl and income

The relationship between income and social capital is positive for most social-capital factors The role of urban sprawl is only statistically significant for some factors of social capital and can either reduce or widen the gap between income groups In the case of informal social interaction (playing cards with others, visiting relatives or having them visit, having friends over, socializing with coworkers outside of work, and socializing with friends in public places), a higher degree of compactness expands the social-capital gap between different income groups [0:06(1) ‡ 0:1LNCOMPACT for $30 ^ 50 K; 0:08(1† ‡ 0:19LNCOMPACT for $50 ^ 75 K; and 0:14(1) ‡ 0:15LNCOMPACT for above $75 K] Meanwhile, the negative signs of LNCOMPACT for different income groups suggest that more compact development helps reduce the gap in income that affects social-capital factors including the number of group involvements, faith-based social capital, giving and volunteering, and electoral politics The opposite signs of the base and of LNCOMPACT also suggest that the magnitude of LNCOMPACT can reverse the relationship between social capital and income For example, after controlling for other factors, the difference between the levels of participation in electoral politics of a person whose household income of $50 ^ 75 K and of a person whose household income is under $30 K is 0:34(1) ÿ 0:24LNCOMPACT In any counties as sprawled as Jasper (IN), the difference is ÿ0:67, meaning the person of the lower household income participates more in electoral politics This pattern applies to more income groups with respect to faith-based social capital The effect of living in counties that have sprawl similar to that in Jasper for a person whose household income is below $30 K is 0.55 units higher compared with a person whose income is $50 ^ 75 K, and 0.89 units higher compared with a person whose income is above $75 K (table 5) Such effects of urban form can also be detected for household income over $75 K and $30 ^ 50 K in the case of giving and volunteering activities, and for household income over $75 K in the case of the number of group involvements

The impacts of sprawl and child presence

Similar to the relationship between income and informal social interaction, living in less sprawled or more compact counties adds to the difference between those who live with children of 17 years of age or younger and those who not, when faith-based social capital and giving and volunteering are concerned (tables and 6) However, the presence of children in the same household is negatively related to the respondent's social trust and electoral-political participation

Findings of other variables and controls

Table summarizes the direction of relationships between individual social-capital factors and other independent variables at the person and place levels Among indi-vidual characteristics, education attainment and length of residence in the community have significantly positive association with all nine factors of social capital As an individual's levels of education and length of residence in the community increase, he or she has more social interactions with friends and neighbors informally and in group activities, more religious participation, and more civic and political participa-tion In particular, the relationship between residence tenure across all factors of social capital may suggest that endogeneity is not the issue If people with high levels of social capital move to neighborhoods composed of people with similar levels of social capital, there should be no statistically significant difference in social capital between those who have lived in the neighborhood for under five years and those who have lived there for longer, at least in social-capital factors that are mostly determined outside the residential neighborhood, such as informal social interaction and electoral-political participation

Owning a house is related positively to the number of group involvements, faith-based social capital, social trust, and community and political participation However, homeownership is related negatively to informal social interaction and not signi-ficant for the remaining social-interaction variables: diversity of friendship and organized-group interaction

As one becomes older, a person increases his or her civic and political participa-tion as indicated by a positive relaparticipa-tionship between AGE and the number of group involvements, faith-based social capital, giving and volunteering, and electoral politics However, his or her intensity of social interaction, as indicated by informal social interaction and organized-group interaction, declines This result is consistent with Putnam's observation of the difference between the number of organizations that a person belongs to and the frequency at which that person participates in his or her member activities (Putnam, 2000)

Males are more likely to belong to a variety of groups and to participate in political activities; however, females are more likely to be involved in more social-interaction activities and in volunteering activities Females also have higher levels of social trust and of faith-based social capital

Marriage significantly reduces the intensity of informal and organized social-interaction activities This is probably because married couples tend to spend more time together, leaving them less time to devote to visiting friends and clubbing Also, married respondents are less likely to participate in nonelectoral political activities such as signing petitions, joining protests, and boycotting, and likely to be involved in labor unions and ethnic, nationality, or civil rights groups

D

Ngu

ye

n

N

:/ps

files/

epb3

70

4w

/

friendship social

interaction groupinteraction groupinvolvements socialcapital volunteering politics politicalparticipation

AGE ± ± ± ‡ ‡ ‡ ‡ ‡

Being male ± ± ‡ ± ± ± ‡ ‡

Owning house ± ‡ ‡ ‡ ‡ ‡ ‡

Being married ‡ ± ± ‡ ‡ ‡ ‡ ±

Education attainment ‡ ‡ ‡ ‡ ‡ ‡ ‡ ‡ ‡

Length of residence ‡ ‡ ‡ ‡ ‡ ‡ ‡ ‡ ‡

Being Black

(versus White) on sprawldepends ± on sprawldepends ‡ ‡ ± ‡ ‡

Being Asian

(versus White) ± ± ± ± ± ± ±

Being Hispanic

(versus White) ± ± on sprawldepends ± ±

Being employed ‡ ± ‡ ± ‡ ‡ ‡

Presence of children 17 years or younger in household

‡ ‡ ‡ ± ‡ ±

Household income ‡ ‡ ‡ depends

on sprawl on sprawldepends ‡ on sprawldepends on sprawldepends ‡ Not disclosing income

(versus income < 30 K)

‡ ‡ ‡ ‡ ‡ ± ‡

Not disclosing child information Tract mean household

income ± ± ± ± ‡ ‡ ± ±

Tract percentage of

college graduates ‡ ‡ ‡ ± ‡ ‡ ‡ ‡

Tract racial diversity ‡ ± ±

Percentage of rural

population in county ± ± ± ‡

At the census-tract level, mean household income is negatively related to most individual social-capital factors except for faith-based social capital and social trust, which have a positive correlation with the mean household income in the census tract The relationships between giving and volunteering and mean household income, how-ever, is statistically insignificant Another census-tract variable, the percentage of college graduates, is positively related to most individual social-capital factors except for faith-based social capital, with which it has a negative relationship This percentage appears to be irrelevant with respect to informal social interaction The remaining census-tract variable, racial diversity, is positively related to diversity of friendship but negatively related to informal social interaction and social trust In other words, if one lives in an area with high racial diversity, one tends to have more friends from a variety of backgrounds, but to interact less with friends and have less trust in different groups in society This result supports the argument about diversity and social capital by different scholars, most notably Glaeser et al (2000), Kahn and Costa (2002), and Putnam (2000) At the county level, county population size is shown to have statis-tically significant negative impacts on a person's participation in electoral politics Meanwhile, the percentage of rural population living in the county is negatively related to faith-based social capital and giving and volunteering, but it is positively related to nonelectoral political participation

Conclusion

By controlling for place characteristics, the analysis provides important statistical information on the association between different socioeconomic and demographic characteristics and different aspects of social capital Compact living, as characterized by high population density and street accessibility at the county level, is found to be unfavorable to social interaction, faith-based social capital, and giving and volunteer-ing However, the analysis shows that compact living is positively related to political participation such as voting, involvement in political groups and local reforms, and interest in national affairs

It also sheds light on the role of urban form as a possible confounder of the association between socioeconomic and demographic characteristics and social cap-italöthere exist differential impacts of sprawl on the relationship between some factors of social capital and race, income, employment status, and child status Social inter-action may not always take place in more compact areas as some authors have contended, but compact living can amplify the positive effects of income on social interaction Compact living can compensate for or widen the social-capital gap intrinsic to race and income However, planners should be aware that most variation in individual social capital is explained by personal characteristics and less at the neighborhood or county level

Regardless of those limitations, this study has contributed to the debate on urban sprawl and provided evidence to support conclusions different from earlier studies The findings strongly suggest that focusing on urban form, at least at the county level, may not be an ideal approach to improving social capital, which may disappoint New Urbanists and other compact-living enthusiasts It is because social capital is not a simple factor and social-capital determinants such as a person's education may be more important

Finally, the study raises new questions about land-use planning and social equality such as whether more equal opportunities for certain groups in society can be achieved via land-use-planning policies due to the complex interaction between urban form and the relationships between individual characteristics and social capital

References

Anderson W P, Kanaroglou P S, Miller E, 1996, ``Urban form, energy and the environment: a review of issues, evidence and policies'' Urban Studies 33 ^ 35

Audirac I, Shermyan A H, Smith M T, 1990,``Ideal urban form and visions of the Good Life Florida growth management dilemma'' Journal of the American Planning Association 56 470 ^ 482 Bento A M, Cropper M L, Mobarak A M, Vinha K, 2005, ``The effects of urban spatial structure

on travel demand in the United States'' Review of Economics and Statistics 87 466 ^ 478 Brueckner J, Largey A, 2006, ``Social interaction and sprawl'', WP 060707, Department of

Economics, University of CaliforniaöIrvine, Irvine, CA

Burchell R W, Lowenstein G, Dolphin W R, 1998 Costs of SprawlöRevisited (National Academy Press, Washington, DC)

Burchell R W, Lowenstein G, Dolphin W R, Gally C C, 2002 Costs of Sprawl Transportation Research Board, National Research Council, http://onlinepubs.trb.org/onlinepubs/tcrp/ tcrp rpt 74-a.pdf

Campbell K, Lee B, 1992, ``Sources of personal neighbor networks: social integration, need, or time? '' Social Forces 70 1077 ^ 1100

Cervero R, Duncan M, 2003, ``Walking, bicycling, and urban landscapes: evidence from the San Francisco Bay area''American Journal of Public Health 93 1478 ^ 1483

Cervero R, Wu K, 1998, ``Subcentering and community: evidence from the San Francisco Bay area'' Urban Studies 35 1059 ^ 1076

Ewing R, 1994, ``Characteristics, causes, and effects of sprawl: a literature review'' Environmental and Urban 21 ^ 15

Ewing R, 1997, ``Is Los Angeles-type sprawl desirable? '' Journal of the American Planning Association 63 107 ^ 125

Ewing R, Pendall R, Chen D, 2002, ``Measuring sprawl and its impact Smartgrowth America'', http://www.smartgrowthamerica.org/sprawlindex/sprawlindex.html

Ewing R, Schmid T, Killingsworth R, Zlot A, Raudenbush S, 2003, ``Relationship between urban sprawl and physical activity, obesity, and morbidity'' American Journal of Health Promotion 18 47 ^ 58

Freeman L, 2001, ``The effects of sprawl on neighborhood social ties: an explanatory analysis'' Journal of American Planning Association 67 69 ^ 77

Galster G, Hanson R, Ratcliffe M R, Wolman H, Coleman S, Freihage J, 2001, ``Wrestling sprawl to the ground: defining and measuring an elusive concept'' Housing Policy Debate 12 681 ^ 717 Gelman A, Hill J, 2007 Data Analysis Using Regression and Multilevel/Hierarchical Models

(Cambridge University Press, New York)

Glaeser E, Gottlieb J, 2006, ``Urban resurgence and the consumer city'', DP 2109, Harvard Institute of Economic Research, http://www.economics.harvard.edu/hier/2006papers/HIER2109.pdf Glaeser E, Kahn M, 2003, ``Sprawl and urban growth'', DP 2004, Harvard Institute of Economic

Research, http://www.post.economics.harvard.edu/hier/2003papers/2003list.html

Glaeser E L, Kahn M E, Rappaport J, 2000,``Measuring trust'' The Quarterly Journal of Economics 15 811 ^ 846

Gordon P, Richardson H, 1997, ``Are compact cities a desirable planning goal? '' Journal of the American Planning Association 63 95 ^ 107

Gordon P, Richardson H, 2000, ``Defending urban sprawl'' Public Interest 138 65 ^ 72

Hayward S, 1998, ``Legends of the sprawl'' Policy Review 91 26 ^ 32

Iyer S, Kitson M, Toh B, 2005, ``Social capital, economic growth and regional development'' Regional Studies 39 1015 ^ 1040

Jacobs J, 1961 The Death and Life of Great American Cities (Random House, Toronto) Kahn M, 2000, ``The environmental impact of suburbanization'' Journal of Policy Analysis and

Management 19 569 ^ 586

Kahn M, 2001, ``Does sprawl reduce the black/white housing consumption gap? '' Housing Policy Debate 12 77 ^ 86

Kahn M, 2006, ``The quality of life in sprawled versus compact cities'', paper prepared for the OECD ECMT Regional Round Table 137, Berkeley, CA, http://www.OECD.org/publishing Kahn M, Costa D, 2002, ``Civic engagement and community heterogeneity: an economist's

perspective'', Social Science Research Network, http://www.ssrn.com/abstract=311839 Keylock C, 2005,``Simpson diversity and the Shannon-Wienner index as special cases of generalized

entropy'' Oikos 109 203 ^ 207

Leyden K, 2003,``Social capital and built environment: the importance of walkable neighborhood'' American Journal of Public Health 93 1546 ^ 1551

Luke D, 2004 Multilevel Modeling (Sage, Thousand Oaks, CA)

Putnam R, 2000 Bowling Alone: The Collapse and Revival of American Community (Simon and Schuster, New York)

Putnam R D, Leonardi R, Nanetti R Y, 1993 Making Democracy Work (Princeton University Press, Princeton, NJ)

Raudenbush S W, Bryk A S, 2002 Hierarchical Linear Models: Applications and Data Analysis Methods 2nd edition (Sage, Newbury Park, CA)

Rong F, 2006 Impact of Urban Sprawl on US Residential Energy Use PhD thesis, University of Maryland, College Park, MD

Snijders T, Bosker R, 1999 Multilevel Analysis: An Introduction to Basic and Advanced Multilevel Modeling (Sage, London)

The Roper Center, 2005 Social Capital Benchmark Survey Restricted Data 2000 The Roper Center for Public Opinion Research Data, University of Connecticut, Storrs, CT, http://www.ropercenter.uconn.edu/

Weitz J, MooreT, 1998,``Development inside urban growth boundaries: Oregon's empirical evidence of contiguous urban form'' Journal of the American Planning Association 64 424 ^ 440