Académique Documents
Professionnel Documents
Culture Documents
www.elsevier.com/locate/jue
Abstract
Various authors, most notably Putnam [Putnam, R.D., 2000. Bowling Alone. Simon and Schuster, New York], have argued that
low-density living reduces social capital and thus social interaction, and this argument has been used to buttress criticisms of urban
sprawl. If low densities in fact reduce social interaction, then an externality arises, validating Putnams critique. In choosing their
own lot sizes, consumers would fail to consider the loss of interaction benefits for their neighbors when lot size is increased.
Lot sizes would then be inefficiently large, and cities excessively spread out. The paper tests the premise of this argument (the
existence of a positive link between interaction and density) using data from the Social Capital Benchmark Survey. In the empirical
work, social interaction measures for individual survey respondents are regressed on census-tract density and a host of household
characteristics, using an instrumental-variable approach to control for the potential endogeneity of density.
2007 Elsevier Inc. All rights reserved.
1. Introduction
* Corresponding author.
1 For a good overview of these arguments, see the 12-article symposium published in the Fall 1998 issue of the Brookings Review.
19
20
(3)
21
(4)
(n 1)f U I
,
U c
nq 2
(5)
where the asterisks on f and the marginal utilities indicate that these expressions are evaluated at the social optimum. Recalling from above that the expression in (5),
absent the n 1 factor, gives one consumers loss from
lower interaction when his lot size increases, it follows
that the tax serves to charge the consumer for the equivalent losses suffered by his n 1 neighbors. With the
8 If individual land consumption levels were chosen by a large-scale
developer rather than by atomistic consumers, the density externality would presumably be internalized and this efficiency eliminated.
However, even though such developers play an increasing role in US
cities, much of the existing housing stock was built in a more atomistic fashion, leaving room for the externality to operate.
22
constituted a nationally representative sample, the remaining 26,000 came from 40 communities in 29
states where a local sponsor was available to support
a survey. Community surveys differ in the proportion
of their samples drawn from urbanized areas and in the
number of urbanized areas each includes.9
As a result of the geographic focus of the community surveys, combining the national and community
data leads to an unbalanced sample. 88% of the observations come from 40 urbanized areas, and the remaining
12% from 268 others. While this imbalance in the sample need not lead to biased estimates, the data structure
does require a cluster correction in computing coefficient standard errors, which accounts for the possibility
that the individual error terms are correlated within urbanized areas (see below).
The analysis focuses on variables narrowly measuring social interaction, not including broader indicators
of social capital. The variables, whose full definitions
are given in Table 1 along with summary statistics, fall
into two sets. The first set consists of variables measuring the extent of the respondents neighborhood contacts and friendships. #NEITALK measures how often
the respondent socializes with neighbors; NEITALK is
a dummy variable indicating whether this interaction
occurs at least once a week; CONFIDE counts the number of people the respondent can confide in; FRIENDS
counts the number of close friends; SOCPUBLIC measures the frequency of socializing with friends in a
public place; SOCHOME measures the frequency with
which friends are invited to the respondents home.
The second set of variables measures the respondents group involvement. NEICOOP is a dummy indicating cooperation with neighbors to get something
fixed or improved; HOBBYCLUB is a dummy indicating membership in a hobby-oriented club; CLUBMTGS
counts the frequency of attendance at any club meetings
over the previous twelve months; # GROUPS counts the
number of formal non-church groups to which the member belongs.
The various social-interaction measures are used as
dependent variables in separate regressions. In addition
to tract-level population density, which is measured in
9 Some communities focus almost entirely on one urbanized area,
whereas others are more evenly distributed across a few. For example, all urbanized-area observations in the Community Foundation for
Atlanta sample are from Atlanta itself, while the Foundation for the
Carolinas draws 45% of its urban observations from Charlotte, 19%
from Gastonia, 17% from Concord, and 13% from Rock Hill. In total,
the community surveys select respondents from 110 urbanized areas,
but the observations are heavily weighted towards 40 of those areas.
23
Table 1
Variable definitions
Variable
Definition
Min
Max
5.079
0.538
3.551
3.326
16.767
22.049
0
0
60
60
0.315
0.261
6.008
3.158
0
0
0
0
1
1
53
18
0.411
44.584
0
18
1
99
0.129
0.622
0.249
0.569
0.258
0.475
0.148
0.257
0.201
0.114
0.126
0.327
0.362
0.147
0.088
0.028
0.044
0.059
0.161
0.949
0.084
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
1
1
14
8
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
8.008
7.845
6.188
1.529
6.747
3.281
12.258
8.863
7.615
9.7917
0.0579
56.033
0.066
0.096
0.247
0.060
0.206
Dependent Variables
#NEITALK
How often respondent talks with or visits immediate neighbors: 1 = never,
2 = once a year or less, 3 = several times a year, 4 = once a month, 5 = several
times a month, 6=several times a week, 7 = just about every day
NEITALK
= 1 if respondent talks or visits immediate neighbors at least once a week, 0
otherwise
CONFIDE
Number of people respondent can confide in: 1 = nobody, 2 = one, 3 = two,
4 = three or more
FRIENDS
Number of close friends respondent has: 1 = none, 2 = one or two, 3 = three to
five, 4 = six to ten, 5 = more than ten
SOCPUBLIC
Number of times per year respondent hangs out with friends in a public place
SOCHOME
Number of times per year respondent visits with friends at home
NEICOOP
HOBBYCLUB
CLUBMTGS
#GROUPS
Mean
0
1
0
1
0
1
0
1
0
1
(continued on next page)
24
Table 1 (continued)
Variable
Definition
Mean
EASOCENT
WESOCENT
MOUNTAIN
MSA Controls
MURDER
DISSIMILARITY
Min
Max
0.044
0.030
0.078
0
0
0
1
1
1
5.606
0.596
0
0.237
20.4
0.840
25
Table 2
First stage regression
MALE
AGE
AGE2
MARRD/PARTN
KIDS6
KIDS6_17
INC2
INC3
INC4
INC5
INC6
SOMECOLL
COLLGRAD
BLACK
HISPANIC
ASIAN
UNEMP
HOMEMAKER
RETIRED
CITIZEN
LIVING5
NEWENGL
MIDATLAN
EANOCENT
WENOCENT
SOUTHATL
EASOCENT
WESOCENT
MOUNTAIN
MURDER
DISSIMILARITY
RUGGED
UA DENSITY
MSA DENSITY
CONSTANT
Coeff.
0.006
0.011
0.000
0.174
0.032
0.026
0.088
0.160
0.280
0.312
0.407
0.090
0.219
0.503
0.390
0.136
0.107
0.032
0.057
0.114
0.006
0.207
0.260
0.328
0.047
0.454
0.345
0.319
0.027
0.040
0.649
0.011
0.781
0.459
0.228
0.29
3.05
1.86
7.71
2.51
2.05
2.90
5.59
7.94
7.49
8.20
3.19
5.40
10.54
4.88
2.37
2.64
0.72
1.30
2.47
0.12
0.47
1.06
1.24
0.27
1.78
1.22
1.21
0.15
3.09
1.18
3.39
3.85
7.40
0.17
grouping of respondents into urbanized areas, the coefficient standard errors in the regression are adjusted using
a cluster correction at the UA level, as are the standard
errors for the interaction equations reported below. The
results show that average urbanized-area and MSA population densities are both strongly significant determinants of individual tract densities, suggesting that use of
both variables as instruments is appropriate. The coefficients are both less than one, indicating that tract densities respond only partially to an increase in the level
of aggregate density. Note also that the urbanized-area
tine, the coefficients and standard errors differ slightly from those in
Table 2.
4. Empirical findings
4.1. First-stage regression
26
and Asian households show less interaction than nonminority respondents. The regional coefficients show
that, relative to the omitted Pacific region, interaction
with neighbors is stronger in the west-north-central part
of the country, which corresponds to the Plains states,
and lower in the south-Atlantic region. Unexpectedly,
the coefficients of the MURDER and DISSIMILARITY variables are both insignificant. Finally, the test for
endogeneity of TRACT DENSITY suggests that this
variable is indeed endogenous, and the overidentification J -test shows that the instruments are valid.
Panel B of Table 3 shows IVPROBIT results using
the binary NEITALK measure as dependent variable.
This variable, which indicates that the respondent interacts with neighbors at least once a week, is a less-precise
measure than #NEITALK. Apparently because of this
difference, the tract-density coefficient, while still negative, is now insignificant. The signs and significance of
the remaining coefficients mostly follow the pattern in
panel A, except that none of the income coefficients is
significant. The endogeneity test rejects exogeneity of
TRACT DENSITY.13
Panel C of the table shows 2SLS results when CONFIDE is the dependent variable. The tract-density coefficient is negative and significant, indicating that respondents living in dense census tracts have fewer confidants. Once again, this result contradicts the hypothesis
of stronger interaction in denser areas. While the effects of income, education, minority status, retirement
and citizenship match those in the #NEITALK equation,
marriage and children now have significantly negative
coefficients. Apparently, marriage and children increase
superficial contacts with neighbors but reduce the formation of strong outside bonds by fostering an inward
orientation. While panel A showed that older respondents interact more with their neighbors, panel C shows
they have fewer confidants.14 Also, residence outside of
the Pacific region mostly reduces the number of confidants, in contrast to the findings for neighbor interaction. Male respondents have fewer confidants than
do females, and unemployment now has a significantly
negative effect. Once again, the coefficients of MURDER and DISSIMILARITY are insignificant. While the
endogeneity test now rejects exogeneity of density only
at the 9 percent level, the overidentification J -test nevertheless indicates validity of the instruments.
13 To test for validity of the instruments, the probit equation was rees-
27
Table 3
Friendship-oriented variables
A
#NEITALK
(2SLS)
TRACT DENSITY
MALE
AGE
AGE2
MARRD/PARTN
KIDS6
KIDS6_17
INC2
INC3
INC4
INC5
INC6
SOMECOLL
COLLGRAD
BLACK
HISPANIC
ASIAN
UNEMP
HOMEMAKER
RETIRED
CITIZEN
LIVING5
NEWENGL
MIDATLAN
EANOCENT
WENOCENT
SOUTHATL
EASOCENT
WESOCENT
MOUNTAIN
MURDER
DISSIMILARITY
CONSTANT
OLS or Probit Coeff. z
TRACT DENSITY
N
Number of clusters
F (d.f.)
P -value
Wald Chi-sq (32)
P -value
Hansen J Statistic
Chi-sq(2) P -value
Endogeneity test of
endog. regressors
Chi-sq(1) P -value
B
NEITALK
(IVPROBIT)
C
CONFIDE
(2SLS)
D
FRIENDS
(2SLS)
Coeff.
Coeff.
Coeff.
0.124
0.054
0.024
0.000
0.091
0.127
0.074
0.021
0.199
0.227
0.241
0.260
0.125
0.081
0.366
0.449
0.280
0.084
0.113
0.144
0.471
0.090
0.227
0.138
0.089
0.149
0.161
0.234
0.096
0.086
0.000
0.401
4.298
3.15
1.61
3.29
1.10
2.50
5.65
3.53
0.32
3.56
3.51
3.44
3.45
3.19
1.89
6.49
6.18
2.98
0.98
1.44
2.20
6.12
1.51
1.87
1.44
1.02
2.32
2.03
2.00
1.02
0.97
0.04
1.42
12.56
0.030
0.026
0.010
0.000
0.074
0.099
0.053
0.035
0.025
0.008
0.008
0.032
0.065
0.031
0.143
0.194
0.157
0.049
0.085
0.108
0.311
0.030
0.079
0.016
0.003
0.134
0.035
0.079
0.049
0.015
0.004
0.012
0.497
1.45
1.11
2.07
0.04
2.86
5.06
4.06
1.13
0.65
0.24
0.19
0.61
2.90
1.07
3.98
3.88
2.93
1.04
1.42
2.45
5.77
0.78
1.03
0.31
0.06
3.99
0.74
1.19
0.91
0.38
0.90
0.08
2.59
0.021 2.24
0.102 9.46
0.009 3.80
0.000 3.17
0.023 1.89
0.036 3.77
0.025 3.43
0.106 3.82
0.173 6.22
0.206 7.16
0.220 8.13
0.272 8.46
0.149 9.44
0.218 14.44
0.188 8.56
0.292 11.23
0.160 3.69
0.055 2.08
0.009 0.35
0.017 0.50
0.120 2.89
0.015 0.79
0.024 1.30
0.108 4.65
0.053 2.15
0.016 0.72
0.123 5.23
0.170 4.34
0.120 3.77
0.007 0.29
0.003 1.30
0.059 0.78
3.710 32.89
0.007
0.43
0.022
14751
302
90.46
0.000
0.183
0.912
6.961
0.008
14751
302
(32, 301)
1735.97
0.000
Coeff.
0.028 2.08
0.061 3.27
0.023 6.38
0.000 6.51
0.044 2.60
0.055 3.46
0.048 4.80
0.130 3.96
0.213 6.70
0.256 7.94
0.271 7.54
0.413 11.32
0.116 4.95
0.219 7.73
0.385 12.73
0.196 5.30
0.160 3.74
0.078 2.13
0.012 0.28
0.121 3.45
0.035 0.88
0.095 3.24
0.117 4.21
0.125 3.52
0.017 0.44
0.054 2.20
0.075 2.17
0.043 0.92
0.012 0.22
0.016 0.39
0.004 1.48
0.089 0.77
3.720 29.67
0.005 0.7
14806
303
(32, 303) 73.74
0.000
E
SOCPUBLIC
(2SLS)
Coeff.
14802
304
(32, 302) 138.96
0.000
8.180
1.941
0.379
2.893
4.237
0.120
3.186
3.622
0.160
3.024
0.004
0.092
0.074
0.082
Coeff.
0.690 2.39
2.343 7.74
1.400 21.76
0.010 15.14
5.278 15.28
1.639 5.88
0.901 5.35
1.202 2.20
2.518 5.03
3.456 5.80
4.446 6.15
5.721 8.24
1.685 4.70
1.095 2.15
2.064 4.65
1.305 1.95
2.322 3.57
0.318 0.36
0.414 0.63
1.489 2.14
1.431 1.83
1.177 2.14
0.579 0.96
1.255 2.02
0.077 0.11
1.560 2.37
2.090 2.77
1.115 1.27
0.147 0.19
0.846 1.35
0.070 1.47
0.716 0.34
59.583 17.82
0.201
F
SOCHOME
(2SLS)
1.62
0.363 1.11
0.116 0.38
1.234 20.64
0.009 13.51
1.933 5.10
0.941 4.47
0.255 0.95
1.186 1.88
1.992 3.42
2.801 4.21
2.836 3.46
5.352 7.23
0.967 2.16
0.568 1.20
3.111 6.48
4.034 5.82
2.943 3.13
0.406 0.50
0.152 0.25
1.821 2.70
4.276 6.46
1.836 3.11
1.021 1.20
0.018 0.02
0.126 0.17
0.616 1.34
1.851 2.76
1.088 1.08
2.211 2.49
1.214 1.85
0.017 0.28
6.464 2.86
59.365 22.32
0.244
14798
304
(32, 303) 189.60
0.000
1.71
(32, 303)
4.950
0.084
1.569
0.210
contradicts the main hypothesis. Most of the other results match those in panel C, but the retirement coefficient is now significantly positive and the sex effect
is reversed, with males having more friends than fe-
28
29
Table 4
Group-involvement variables
A
NEICOOP
(IVPROBIT)
TRACT DENSITY
MALE
AGE
AGE2
MARRD/PARTN
KIDS6
KIDS6_17
INC2
INC3
INC4
INC5
INC6
SOMECOLL
COLLGRAD
BLACK
HISPANIC
ASIAN
UNEMP
HOMEMAKER
RETIRED
CITIZEN
LIVING5
NEWENGL
MIDATLAN
EANOCENT
WENOCENT
SOUTHATL
EASOCENT
WESOCENT
MOUNTAIN
MURDER
DISSIMILARITY
CONSTANT
OLS or Probit Coeff. z
TRACT DENSITY
N
Number of clusters
F (d.f.)
P -value
Wald Chi-sq ( 32)
P -value
Hansen J Statistic
Chi-sq(2) P -value
Endogeneity test of
endog. regressors
Chi-sq(1) P -value
B
HOBBYCLUB
(IVPROBIT)
C
CLUBMTGS
(2SLS)
D
#GROUPS
(2SLS)
Coeff.
Coeff.
Coeff.
Coeff.
0.037
0.036
0.026
0.000
0.070
0.021
0.082
0.018
0.062
0.185
0.261
0.301
0.183
0.192
0.147
0.038
0.079
0.122
0.114
0.018
0.308
0.114
0.172
0.005
0.054
0.148
0.113
0.115
0.148
0.098
0.007
0.237
2.158
1.39
0.99
4.66
3.14
2.49
0.86
4.47
0.32
1.11
3.80
3.81
4.40
5.73
5.27
2.94
0.70
0.61
1.65
1.92
0.29
2.81
1.94
2.73
0.08
0.73
3.06
2.07
0.89
1.70
1.60
0.74
1.07
8.44
0.068
0.136
0.009
0.000
0.084
0.083
0.021
0.014
0.166
0.233
0.301
0.354
0.281
0.407
0.024
0.020
0.054
0.099
0.038
0.161
0.139
0.012
0.117
0.118
0.085
0.059
0.094
0.134
0.040
0.005
0.005
0.369
0.995
3.23
5.34
2.53
2.24
3.85
4.34
1.30
0.32
3.87
5.96
5.37
7.03
10.68
12.32
0.54
0.34
0.86
1.81
0.75
3.48
2.28
0.28
2.07
1.99
1.46
0.91
1.68
1.86
0.58
0.08
1.65
2.93
5.37
0.571
0.631
0.219
0.002
0.897
0.490
0.539
0.488
1.041
1.644
1.291
2.923
2.215
3.763
0.170
0.546
1.269
0.866
0.249
0.859
2.182
0.980
0.655
0.918
0.354
0.313
0.499
1.123
0.301
0.252
0.005
1.181
10.431
2.37
2.94
6.47
5.68
4.67
4.02
3.88
1.79
3.43
5.37
3.72
7.35
8.22
13.74
0.43
1.45
1.89
2.29
0.57
2.23
7.83
2.40
1.57
1.75
0.70
0.73
0.88
1.73
0.47
0.48
0.17
1.00
5.27
0.092
0.034
0.025
0.000
0.062
0.031
0.301
0.252
0.590
0.807
0.973
1.413
0.898
1.694
0.810
0.231
0.250
0.255
0.334
0.084
0.662
0.082
0.167
0.147
0.094
0.065
0.047
0.068
0.019
0.092
0.007
0.232
0.467
1.92
0.65
2.72
0.45
1.48
1.37
11.97
3.54
7.87
12.68
9.99
14.32
17.27
28.06
8.38
2.10
2.03
3.11
3.85
0.80
9.09
1.09
1.35
0.95
0.70
0.46
0.33
0.43
0.12
0.74
0.79
0.77
0.84
0.013
0.88
0.005
0.62
0.016
0.16
0.020
1.15
7314
240
14809
304
798.910
0.000
1046.870
0.000
0.740
0.391
14790
304
67.71
0.000
(32,303)
14823
304
163.40
0.000
9.250
3.933
0.140
2.527
1.692
0.872
2.744
0.002
0.111
0.097
(32, 303)
30
did not reject the overidentifying restrictions. However, the endogeneity test fails to reject exogeneity of TRACT DENSITY by a wide
margin in panel A.
21 In addition, males attend fewer meetings, age and unemployment
now have negative impacts, and long-time residence has a positive
effect. The age effect turns positive beyond 55 years.
22 Respondents who are unemployed or are homemakers also belong
to fewer groups, while sex, marriage, young children, and long-time
residence have no effect. Regional effects are again absent. For panels C and D, the J -test tests are satisfactory, but the endogeneity test
rejects exogeneity of density only at around the 10 percent level.
#NEITALK
SOCPUBLIC
HOBBYCLUB
CLUBMTGS
#GROUPS
Marginal
effect
Change in level
or probability
Mean
0.014
0.690
0.022
0.570
0.092
0.04 (prob)
1.14 (level)
0.06 (prob)
1.54 (level)
0.25 (level)
0.538
16.8
0.252
6.0
3.2
Percent
7%
11%
24%
26%
8%
#NEITALK
NEITALK
CONFIDE
FRIENDS
SOCPUBLIC
SOCHOME
NEICOOP
HOBBYCLUB
CLUBMTGS
#GROUPS
w/ OWNER
Coeff.
Coeff.
Owner
coeff.
0.138
0.044
0.016
0.018
0.614
0.058
0.033
0.056
0.525
0.149
3.39
1.96
1.72
1.31
1.80
0.17
1.21
2.90
2.17
2.25
0.111
0.028
0.020
0.028
0.693
0.195
0.045
0.065
0.494
0.106
2.95
1.22
1.94
1.94
2.04
0.56
1.60
3.32
2.17
1.80
+
+
0
0
0
0
+
0
0
+
31
32
Table 7
Density effects dependent on age and marital status*
Friendship-Oriented Variables
AGE
DENSITY-O
DENSITY-M
DENSITY-Y
MARITAL STATUS
DENSITY-MARRD
DENSITY-UNMARRD
#NEITALK
NEITALK
CONFIDE
FRIENDS
Coeff.
Coeff.
0.097
0.113
0.136
2.29
2.96
3.50
0.095
0.160
2.29
3.41
SOCPUBLIC
SOCHOME
Coeff.
Coeff.
Coeff.
Coeff.
0.007
0.009
0.017
0.77
1.14
2.14
0.019
0.026
0.024
1.48
2.05
1.87
0.006
0.024
0.022
0.43
1.81
1.57
0.595
0.645
0.366
1.83
2.00
1.11
0.336
0.210
0.078
0.96
0.62
0.22
0.003
0.027
0.31
2.78
0.020
0.024
1.98
1.61
0.023
0.034
1.69
1.71
0.426
0.898
1.27
2.17
0.126
0.657
0.41
1.33
Group-Involvement Variables
NEICOOP
Coeff.
AGE
DENSITY-O
DENSITY-M
DENSITY-Y
MARITAL STATUS
DENSITY-MARRD
DENSITY-UNMARRD
HOBBYCLUB
z
CLUBMTGS
#GROUPS
Coeff.
Coeff.
Coeff.
0.013
0.013
0.012
1.38
1.37
1.29
0.025
0.021
0.190
3.54
3.41
2.72
0.573
0.583
0.600
2.27
2.47
2.54
0.135
0.128
0.131
1.99
1.95
2.01
0.022
0.002
2.36
0.16
0.026
0.015
3.44
2.03
0.698
0.392
2.73
1.55
0.113
0.085
1.67
1.29
CHOME equations.
27 The coefficients are not significant in the #GROUPS regression.
Note also that one coefficient is positive and significant in the NEICOOP equation.
33
34
Acknowledgments
We thank Marlon Boarnet, Rainald Borck, Ami
Glazer, David Neumark, David Pines, Stuart Rosenthal,
Ken Small and several referees for helpful comments.
They are not responsible, however, for any shortcomings in the paper. The paper was submitted to the journal
while the first author was editor, but incoming editor
Stuart Rosenthal handled the submission.
References
Aaronson, D., 1998. Using sibling data to estimate the impact of
neighborhoods on childrens educational outcomes. Journal of Human Resources 33, 915946.
Borck, R., 2006. Social agglomeration externalities. Unpublished paper. DIW, Berlin.
Brueckner, J.K., 2000. Urban sprawl: Diagnosis and remedies. International Regional Science Review 23, 160171.
Brueckner, J.K., 2001. Urban sprawl: Lessons from urban economics.
In: Gale, W.G., Pack, J.R. (Eds.), Brookings-Wharton Papers on
Urban Affairs. Brookings Institution, Washington, DC, pp. 6589.
Brueckner, J.K., in press. Urban growth boundaries: An effective
second-best remedy for unpriced traffic congestion? Journal of
Housing Economics.
Burchfield, M., Overman, H.G., Puga, D., Turner, M.A., 2006. Causes
of sprawl: A portrait from space. Quarterly Journal of Economics 121, 587633.
Carlino, G.A., Chatterjee, S., Hunt, R.M. , 2007. Urban density and
the rate of innovation. Journal of Urban Economics 61, 389419.
Ciccione, A., Hall, R.E., 1996. Productivity and the density of economic activity. American Economic Review 86, 5470.
DiPasquale, D., Glaeser, E., 1999. Incentives and social capital: Are
homeowners better citizens? Journal of Urban Economics 45,
354384.
Eid, J., Overman, H.G., Puga, D., Turner, M.A., 2006. Fat city: The relationship between urban sprawl and obesity. Unpublished paper.
London School of Economics.
Evans, W., Oates, W., Schwab, R., 1992. Measuring peer group effects: A study of teenage behavior. Journal of Political Economy 100, 966991.
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,
4757.
Fujita, M., 1989. Urban Economic Theory. Cambridge Univ. Press,
New York.
Glaeser, E.L., Kahn, M.E., 2004. Sprawl and urban growth. In: Henderson, J.V., Thisse, F. (Eds.), Handbook of Urban Economics,
vol. IV. Elsevier, Amsterdam, pp. 24812527.
Glaeser, E.L., Gottlieb, J.D., 2006. Urban resurgence and the consumer city. Urban Studies 43, 12751299.
Lopez, R., 2004. Urban sprawl and the risk for being overweight or
obese. American Journal of Public Health 94, 15741579.
Nechyba, T.J., Walsh, R., 2004. Urban sprawl. Journal of Economic
Perspectives 18 (4), 177200.
Oreopoulos, P., 2003. The long-run consequences of living in a poor
neighborhood. Quarterly Journal of Economics 108, 15331575.
Powdthavee, N., 2005. Identifying causal effects with panel data: The
case of friendship and happiness. Unpublished paper, University
of London.
Putnam, R.D., 2000. Bowling Alone. Simon and Schuster, New York.