Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, noncommercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=resl.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a notforprofit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
The Review of Economic Studies Ltd. is collaborating with JSTOR to digitize, preserve and extend access to
The Review of Economic Studies.
http://www.jstor.org
Review of Economic Studies (2001) 68, 261295 00346527/01/00120261$02.00
? 2001 The Review of Economic Studies Limited
Social Interactions,Local
Spillovers and Unemployment
GIORGIO TOPA
New York University
First version received May 1997;final version accepted July 2000 (Eds.)
I analyse a model that explicitly incorporates local interactions and allows agents to
exchange information about job openings within their social networks. Agents are more likely to
be employed if their social contacts are also employed. The model generates a stationary distri
bution of unemployment that exhibits positive spatial correlations. I estimate the model via an
indirect inference procedure, using Census Tract data for Chicago. I find a significantly positive
amount of social interactions across neighbouring tracts. The local spillovers are stronger for areas
with less educated workers and higher fractions of minorities. Furthermore, they are shaped by
ethnic dividing lines and neighbourhood boundaries.
1. INTRODUCTION
Over the past decade economists have increasingly recognized the importance of non
market interactions in a variety of contexts, such as joblessness, crime and other social
pathologies, peer influences in education, social learning and the diffusion of innovations,
localization choices by households and firms, growth and income inequality. One common
feature in these studies is the assumption that agents' choices and payoffs are affected by
other agents' actions not just indirectly through markets, but also directly through imi
tation, learning, social pressure, information sharing, or other nonmarket externalities.
It is also assumed that agents interact locally, with a set of neighbours defined by an
economic or social distance metric.
Benabou (1993, 1996) and Durlauf (1996a, b) incorporate local interactions in the
human capital accumulation process into endogenous growth models that exhibit neigh
bourhood stratification and persistent and widening income inequality. In the field of
economic geography, Audretsch and Feldman (1996) and Rauch (1993) argue that local
knowledge spillovers produce agglomeration economies and hence affect the location
decisions by firms. There exists also a rich theoretical literature that stresses the role
of interactions in models of herds and information cascades (see Banerjee (1992) and
Bikhchandani et al. (1992)), or in models of social learning (Bala and Goyal (1998), Gale
and Rosenthal (1999), Morris (1997)).
On the empirical side, a vast and growing literature has attempted to provide a stat
istical estimate of the magnitude of local interactions and neighbourhood effects.1 Glaeser
et al. (1996) explain the very high variance of crime rates across U.S. cities through a
model in which agents' propensity to engage in crime is influenced by neighbours' choices:
in so doing, they provide estimates for the range of social interactions. Case and Katz
(1991) explore the impact of neighbourhood effects on several behavioural outcomes, such
1. Jencks and Mayer (1990), loannides and Datcher (1999) and Brock and Durlauf (1999) give excellent
surveys of the empirical literature.
261
262 REVIEW OF ECONOMIC STUDIES
as criminal activity, drug and alcohol use, childbearing out of wedlock, schooling, church
attendance. Ludwig et al. (1999) use the Moving To Opportunity programme as a natural
experiment to evaluate the magnitude of neighbourhood effects. Bertrand et al. (1999) find
that local social networks have a significant impact on individual welfare participation. In
studies concerning education, there is a long tradition starting with the Coleman report
(Coleman et al. (1966)) of studying possible peer influences on educational outcomes:
Hanushek et al. (2000) use very detailed data on Texan schools to estimate peer effects in
student achievement, whereas Zax and Rees (1999) use a Wisconsin Longitudinal Study
to estimate the impact of peer influences during school years on subsequent earnings.
Sociologists have also argued, long before economists, that "one's neighbours mat
ter" in defining one's opportunities and constraints (see Burt (1992)). Individuals are not
considered as isolated entities but rather as being part of networks of friends, relatives,
neighbours, colleagues, that jointly provide cultural norms, economic opportunities, infor
mation flows, social sanctions and so on. Wilson (1987) argues that adults in a community
influence young people by providing role models in terms of the value of education, steady
employment and stable families. Coleman (1988) considers social networks as a source of
"social capital" since they provide valuable information, lower transaction costs, allow
monitoring and enforcement of socially optimal outcomes.
The main objective of this paper is to formulate and estimate a model of local interac
tions in the labour market. In particular, I assume that agents exchange information with
their social contacts about job opportunities and that hiring may occur through informal
channels. The model generates a stationary spatial distribution of unemployment that can
be compared to the empirical unemployment distribution over a set of contiguous
locations in order to estimate the model parameters. I can then test for the existence of
local information spillovers and provide an empirical measure of their magnitude. The
main innovation with respect to the existing empirical literature is to provide a more
structural approach to the estimation of local interaction effects.
The importance of informal channels in finding jobs has been documented, among
others, by Corcoran, Datcher and Duncan (1980) and Granovetter (1995). Both studies
report that more than 50% of all new jobs are found through friends, relatives, neighbours
or occupational contacts rather than through formal means. This is especially true for low
skill jobs, for less educated workers and for black workers. From a theoretical standpoint,
informal hiring channels may coexist in equilibrium with a formal labour market because
of information asymmetries: Montgomery (1991) analyses a model in which employers
cannot perfectly observe the quality of prospective employees and solve the adverse selec
tion problem by relying on referrals from their highability workers. The basic assumption
is that there exists assortative matching in agents' social networks, so that highability
workers are more likely to refer individuals like themselves.
In my model, agents reside in locations that are linked through an explicit network
structure. Each agent, while employed, can transmit information about job openings to
her unemployed contacts; the same agent, while unemployed, can receive useful tips about
jobs from her employed contacts. Riskaverse individuals find it in their best interest to
engage in such information exchanges in order to (partially) insure themselves against
possible future unemployment shocks. These mutual insurance arrangements can be
sustained even in a limited commitment environment.
Unemployment in the model evolves according to a Markov process over the set of
locations. At any point in time, employed individuals may become unemployed with some
exogenous probability, whereas unemployed individuals may find a job with probability
TOPA SOCIAL INTERACTIONS 263
 5179 fK1
rate,1980
Unemployment
05:::1_n ]:.:: :'.''1.e:19: 2 0 2 4 Miles
7.912.1
12.1177
17766.7
FIGURE 1
Map of unemployment rate, 1980
264 REVIEW OF ECONOMICSTUDIES
Unemploymentrate 1990
0=5iio56 2 0 2 4 Miles
569
9 9
139234
234100
FIGURE 2
The unit of observationin the structuralmodel is a Census tract, and the basic
assumptionof the model is that residentsof one tract exchangeinformationlocally with
residentsof the adjacenttracts.This local interactionimpliesthat the employmentrate in
one tract is affectedpositivelyby the employmentrate in the neighbouringtracts. The
transitionsinto and out of unemploymentalso dependon a set of observabletractcharac
teristics.Allowing for tract heterogeneityis imnportant in order to addressthe issue of
positive sorting of individualsacross locations. In fact, agents may sort into different
neighbourhoodson the basis of their neighbours'characteristicsor because they have
similar preferencesover differentconsumptionbundles (see, for example, Becker and
Murphy(1994)).
Such sorting may induce positive spatial correlationof unemploymenteven in the
absence of any local informationspillovers.I try to distinguishthe social interaction
TOPA SOCIAL INTERACTIONS 265
....,...i.~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~.
...
rate,90SO
Unemnployment
[I]66*719 2 0 2 4 Miles
1 90.6
0 63.2
3.27.8
7.872.9
FIGURE 3
Map of unemploymentrate, 199080
chisquare criterion for the indirect estimation.2 In the auxiliary estimation, I also control
for unobserved tractspecific fixed effects that do not vary over time.
The empirical results support the model with builtin local interactions and reject the
hypothesis that the observed spatial patterns in unemployment can be explained by
observed tract characteristics alone. The estimated local spillovers are significantly posi
tive: on average, an increase in the employment rate of neighbouring tracts by one stan
dard deviation brings about an increase in expected employment of 06 and 1 3 percentage
points in 1980 and 1990, respectively.3 Interestingly, spillovers are stronger in tracts with
lower education levels and with higher fractions of nonwhites. This is consistent with
other direct evidence on informal hiring.4
Furthermore, the local interaction channel is found to be weaker across adjacent
tracts that have very different ethnic compositions, as well as across local community
boundaries that have been identified by residents. This result lends support to the hypoth
esis that the observed spatial patterns in unemployment are indeed generated by social
interactions as opposed to other possible intertract linkages. Finally, an important caveat
is represented by the finding that the introduction of spatially correlated shocks, to mimic
the possible presence of correlated unobservables, greatly reduces the size of the infor
mation spillover. However, the model with correlated shocks does not fit the data as well
as the baseline model.
The theoretical and empirical framework developed in this paper may be applied
more generally, to any of the settings described above in which local interaction effects
may be present. The wider goal of this paper then is to provide some tools to test for the
existence of interaction effects, to estimate their magnitude, and to simulate the effects of
proposed policy experiments in their presence.
The remainder of the paper is organized as follows. Section 2 presents the structural
model and its general properties. Section 3 describes the indirect inference methodology
and the auxiliary model, and discusses important identification issues. The data used in
the estimation and the simulations are described in Section 4. Section 5 reports the empiri
cal results of the indirect inference estimation of the structural model, and provides some
numerical estimates of the magnitude of the spillover effects. Section 6 concludes.
the context of longterm wage contracts, and by Ligon, Thomas and Worrall (1999) in
the context of informal credit in developing economies).
The specific way in which I model the information interaction is through a discrete
time Markov process on a set of locations, that is similar to a contact process (this was
first studied by Harris (1974) in the context of interacting particle systems). In the first
part of this section I present the model and describe its basic properties. The individual
units in this setup are Census tracts, since the empirical analysis will be conducted at this
level of aggregation. In the second part, I discuss the use of physical distance to determine
who is "close" to whom, and I present some evidence that justifies carrying out the analy
sis at the tract level. In addition, I discuss possible microfoundations of the model at the
level of individual agents, and show that this disaggregated framework is observationally
similar to the former one.
and will stay at the present state with probability (1 Pd). On the other hand, the prob
ability of going from zero employment to the next higher employment rate depends on
the tract's characteristics, but also on the information Iit that residents of tract i may
receive from their employed social contacts in the neighbouring areas
where yN 1/N Yit.7 Again, tract i may stay at a zero employment rate next period
with probability (1 Pu). In the estimation that follows f() is simply the identity function,
so the information variable Iit is equal to the average employment rate of the neighbours
yN. As for the functions A(), a(), and y( ), they are assumed to be linear in their
arguments: e.g. 2(X) .X0+ = l iXl.
Finally, if the state of tract i at time t is in the interior of the unit interval, then it
switches to either an upward mode or a downward mode with probability 0.5, and in each
case the same transition probability as in equations (1) or (2) applies. Therefore, we can
write:
7. With a slight abuse of notation, I use Ni to indicate both the set of neighbours of tract i, and the
number of elements in this set.
TOPA SOCIAL INTERACTIONS 269
sort.8 The absence of a time subscript on the X covariates amounts to assuming that the
process through which agents lose and find jobs takes place at a higher frequency than
the process that rules individuals' locational choices and hence the spatial distribution of
these characteristics across tracts.
One final note concerns the nature of the random shocks that are used to simulate
the model. The system is started off at some initial configuration yoe X< Every period, a
vector (o, of shocks is drawn from a uniform distribution on [0, 1]. Then the state of each
location in the model is updated according to the realization of the shock wi, using the
transition probabilities detailed above.9 These exogenous shocks to employment can be
i.i.d. in space over the map of the city, or they can be generated according to some
autocorrelation structure. The latter option is exploited in the estimation to allow for the
possibility of correlated unobservables driving the spatial correlation patterns observed in
the data.
At+ = Q It,
where Q is the (Wx W) transition matrix, whose entries qrs denote the transition prob
abilities from state r to state s. These transition probabilities can in principle be calculated
from the conditional transition probabilities (1)(5). A stationary distributionof the pro
cess is a vector v such that v = Q v. It is straightforward to show that a unique stationary
distribution v(X) exists, for any given choice of tract characteristics X.'0
It is very difficult to prove analytical results on the properties of the stationary distri
bution. However, if one considers the case where locations are homogeneous in terms of
their characteristics, the structural model is very similar to a discrete time contact process
defined over a finite integer lattice. The behaviour of the contact process has been exten
sively studied in the literature on interacting particle systems." Two results are particu
larly interesting from my perspective. According to the first, the states of any two sites on
the map exhibit nonnegative spatial correlations. The second result states that this corre
lation is bounded above by a quantity that decays exponentially with the distance between
locations.
Simulations of the structural model show that indeed as one lets the system evolve
for several periods, high or low employment clusters appear on the map of locations. If
tracts are homogeneous, these clusters are equally likely to take place anywhere on the
map. Tract heterogeneity implies that lowemployment clusters are more likely to occur
in certain areas of the map, since tract characteristics affect the transition probabilities
between different employment levels.
In my simulation and estimation exercise I take as given the spatial distribution of the
X covariates (determined by the sorting process), and I run the local interaction process
conditional on that, until it converges to the stationary distribution. The structural param
eters of the y ( ), a(Q), and X(.) functions are estimated off the stationary distribution.
The case in which all the spatial correlation is driven by sorting rather than by local
interactions corresponds to the case where )L(*) is identically zero for all values of the X
characteristics, so the model delivers a very straightforward way to distinguish the two
effects.
12. Conley and Topa (2000) define four different metrics, physical distance, travel time, ethnic, and occu
pational distance, and examine the spatial patterns of unemployment with respect to each metric.
TOPA SOCIAL INTERACTIONS 271
between pairs of agents who live less than 1 mile apart; this percentage rises to 64% for
contacts between agents less than 5 miles apart. In a Detroit study that used a 1975 survey
of about 1,200 residents, Connerly (1985) states that 41% of respondents had at least one
third of their Detroit friends residing within one mile. More relevant to this paper, Hunter
(1974) reports that out of roughly 800 Chicago residents interviewed during 196768,
about 49% said that the majority of their friends resided in the same local community.
Therefore, identifying social contacts with agents who live physically nearby seems like a
reasonable approximation for one of the dimensions of social networks.
A second question concerns the appropriate scale of analysis. Perhaps the most
relevant social interactions occur within Census tracts, thus making the assumption that
residents of one tract exchange information with residents of neighbouring tracts not very
meaningful. However, in the city of Chicago, Census tracts represent fairly small units of
analysis. Pairwise distances between adjacent tract centroids are less than one mile for all
but a handful of tracts. Typically, tracts contain five or six street blocks. The average
population (16 years and older) in a Chicago Census tract was 2700 in 1980 and 2500 in
1990.
More importantly, Census tracts are a subdivision of larger units called Community
Areas. There are 75 such areas in Chicago, and each area contains on average 12 tracts.
These areas, such as Hyde Park, Woodlawn, Lincoln Park, Englewood, have names that
Chicago residents readily associate with neighbourhoods. In fact, Community Area
boundaries were drawn in the 1920's by a group of Chicago sociologists, such as Ernest
Burgess and Robert Park, to represent communities with a distinctive identity. The main
criteria used to establish these Areas were: "(1) the settlement, growth, and history of the
area; (2) local identification with the area; (3) the local trade area; (4) distribution of
membership of local institutions; and (5) natural and artificial barriers" such as rivers,
railroad lines, large roads, parks.'3
Even though neighbourhood boundaries change over time, these Community Areas
still represent, in many cases, meaningful and cohesive neighbourhoods. Hunter (1974),
in his survey of Chicago residents, asked respondents to name the boundaries of their
neighbourhoods. He identified roughly 200 neighbourhoods, which in all cases represent
units larger than Census tracts. In many cases, the boundaries of these neighbourhoods
coincide with the original Community Area boundaries.'4 Therefore, it makes sense to
assume that there exist meaningful intertract social interactions, and to take Census tracts
as the unit of analysis.
Finally, I would like to focus on the issue of aggregation. The structural model could
be set up at the individual level, where sites represent individual agents, and the set of
neighbours is defined as the set of contacts within a neighbourhood of radius r. The
transition probabilities would then be defined at the single agent's level, determining the
transitions into and out of work. Then the model could be aggregated to the tract level
for estimation purposes.
It is hard to come up with an individual level model that, when aggregated, generates
transition rules as in (1)(5). However, one can compare the behaviour of such a model,
in terms of its spatial properties, with the structural model presented here. Conley and
13. From the Local CommunityFact Book, Erbeet al. (1984).A certaingeographicarea was considered
a CommunityArea if it had "a historyof its own as a community,a name, an awarenesson the part of its
inhabitantsof common interests,and a set of local businessesand organizationsoriented to the local
community".
14. In the estimationthat follows,I exploittheseboundariesto try to distinguishspilloversdue to social
interactionsfrom otherpossiblesourcesof spatialcorrelation.
272 REVIEW OF ECONOMIC STUDIES
Topa (2001) analyse a model in which artificial agents are placed at the centroid of each
tract, and interact both with agents in their same tract, and with agents in tracts whose
centroids are less than one mile away, on the physical map.15 The transition probabilities
are defined at the agent's level, and are very similar to the ones described here. The
contagion term At() is a declining function of physical distance, so the interaction is
stronger between residents of the same tract than between residents of two separate tracts.
The spatial properties of these two alternative models can be compared by looking
at the AutoCorrelation Functions (ACFs) and the spectral densities generated by the
spatial distribution of employment in model simulations, following a procedure described
03 ,  l
Tractlevel
   Individual level
025
02 
015
01
005
005
0 2 4 6 8 10 12
Physical distance (adjacenttracts)
FIGURE 4
ACF comparison: tract level vs. individual level model
in Section 3.2. Figures 4 and 5 show that one can find parameter values such that the
ACFs and the spectral densities are quite similar in the two models, thus implying that
the models can be made to deliver similar spatial correlation patterns.16 Therefore, I con
sider the structural model presented in this section as a viable approximation to a more
disaggregated model in which local interactions are defined at the level of individual
agents.
3. ESTIMATION STRATEGY
The objective of the estimation strategy is to estimate the structural parameters of the
model presented in Section 2.1, in order to test for the existence of spillovers generated
15. The numberof agentsin each tractis proportionalto the actualpopulationin Chicagoin 1980.One
artificialagentcorrespondsto about 20 peoplein real life.
16. However,the parameterestimatesand the spillovermagnitudesmay be differentin the two setups.
Conleyand Topa (2001)examinethis issue in more detail.
TOPA SOCIAL INTERACTIONS 273
035 , , I, , ,
Tractlevel
   Individuallevel
03

t 025

d02
015 /
4 3 2 1 0 1 2 3 4
Frequency
FIGURE 5
Spectrum comparison: tract level vs. individual level model
by the local interaction process. The crosssectional data used in the estimation are
assumed to be a realization of the unique stationary distribution of the model. I focus on
the stationary distribution in order to be able to disregard the potential impact of different
initial conditions on the spatial moments used in the estimation.
In principle, one can write the likelihood function for the stationary distribution of
the model starting from the set of conditional probabilities of the state at one location,
given the contemporaneousstate at neighbouring locations (the key result here is the fac
torization theorem discussed in Cressie (1993, p. 412)). However, it is extremely hard to
write the likelihood in this case, because the construction involves pinning down a nor
malizing constant (itself a function of the model parameters) which is not available in
closed form.
Therefore, following an indirect inference strategy, I look for an auxiliary statistical
model that can fit the unemployment data well, but is much simpler to estimate than the
structural model. Such a model need not necessarily nest the structural model and may in
fact even be misspecified. In particular I look for a statistical model that best approximates
the spatial properties of the stationary distribution out of the local interaction model.
Choosing an auxiliary model is similar to selecting moments in Simulated Method of
Moments to be matched between the data and simulations of the structural model, in
order to deliver estimates of the structural parameters.
The idea behind indirect inference is quite intuitive. Let 0E Oc93P be a vector of p
parameters that fully characterize the structural model.'7 Let 00 denote the true value,
assuming that the structural model is correctly specified. Likewise, let pe AcC91q be the
17. These are the parameters of the y( ), a(), and 2() functions in the conditional transition probabilities
(1)(5).
274 REVIEW OF ECONOMICSTUDIES
vector of q parameters of the auxiliary model, with q' p. First, one estimates the auxiliary
model using the actual data. These parameter estimates p depend on the true value 00.
Then one simulates the structural model for different values of 0 in the parameter space
0. For each choice of 0, one can estimate the auxiliary model using the outcome of the
simulations (in this case, a simulated unemployment realization over the map of the city).
This estimation yields parameter estimates P(0) that depend on the specific value of 0
used for that simulation. The parameters of interest 0 are estimated by minimizing the
distance between i(00) and p(O), according to some metric to be specified. In the remain
der of this section, I first of all present the indirect inference procedure and describe the
auxiliary model; secondly, I discuss identification issues.
The auxiliary model is possibly misspecified, since I assume that the structural model is
a,s
the true model. One can show that p, 4 p as n * oo, where p r(0O). The function
r: e  is defined as follows:
r(0) = arg min Joo(G,0, p). (7)
pEM
where Joo(G,0, p) = lim,OO J(Yn 9Xn Ip), and G is the distribution of the random shocks
that determine the stochastic process of y. The limit criterion Joo(G,0o, p), evaluated at
the true value 00, is assumed to have a unique minimum at p.
Turning now to the simulations, for each value of Oe 0EI draw H simulated realiza
tions of y out of the stationary distribution of the structural model, PYh(X, 0), h
1,.. , H. I then perform the auxiliary estimation on the simulated outcomes, for each
value of 0 and for given x,: this yields
pn~(0)= arg minJn (Yn 0),xn,p) (8)
peS
on = argmmi[n
L _ H 1h =1 p(o)] 
Qn[Pn HI
Hh= lPI(0) 9
Oee
As is the case for a standard GMM procedure, the optimal weighting matrix Qn in the
quadratic criterion (9) is Q = V1, where V, is the covariance matrix of the auxiliary
TOPA SOCIAL INTERACTIONS 275
where the superscript Nd refers to the average level of unemployment in the tracts at
distance d from tract i.19
I use a criterion, based on spectral decompositions, to compare the spatial properties
of a contact process and of a SAR model, and to choose the order of the SAR that best
fits the contact process. Let the contact process be the "true" model, and let the SAR be
an approximating model to the truth. For a given order D of the Spatial AutoRegression,
I estimate via maximum likelihood the parameters of the SAR that best fit the true
model.20 I can then compare the fit across different possible approximating models (i.e.
SAR models of different orders). The result is that the goodness of fit improves as one
adds higher order terms to the SAR: Figure 6 reports the spectral density of a typical
crosssectional distribution generated by the contact process and compares it to that of
several SARs. In most cases, a SAR of order 6 or higher fits the spectra generated by a
contact process remarkably well: therefore, I use a SAR(6) as the basis of my auxiliary
model.
The statistical model in (10) needs to incorporate a set of M covariates xi in order to
control for tract characteristics that may both affect the probability of finding and losing
jobs, and be dimensions along which agents sort into locations. The structural model of
Section 2.1 also lets the local interaction term 24() be itself a function of local character
istics. To capture this feature in the auxiliary model, I add interaction terms between y_1
18. Consistencyof the indirectinferenceestimatordoes not dependon the specificchoiceof the auxiliary
model. However,the closenessof fit betweenthe structuraland the auxiliarymodels affects efficiency:see
Tauchen(1996).
19. This is definedas yNd= Wi(d) y, where Wi(d) is the ith row of a weightingmatrix W(d) constructed
as follows. Wi(d)givespositiveequalweightsto all tractsthat are at distanced from i. If a tractsharesan edge
with i on the physicalmap, d= 1; if a tract is adjacentto one of these immediateneighboursof i (but is not
adjacentto i), d = 2, and so on. The weightsadd up to unity.
20. See Hansenand Sargent(1993).The detailsof the constructionare availablefrom the authorupon
request.
276 REVIEW OF ECONOMIC STUDIES
04 0.4
4
G
2 0 2 4
~~~~~~~~~~~0
4 2 0 2 4
0.4  0.4
0.2 0.2
4 2 0 2 4 2 0 2 4
FIGURE 6
and a subset of the covariates. Let xi be a Jx 1 vector of regressors, with J< M. Let
= [y,..., y6 ] Equation (10) can be rewritten as
yi = (.i) + (xi) ,B+ i, i= 1... ,n. (1
The following assumptions are made concerning the covariance structure of the error
terms: E(8i) = 0, E(,c2) = ui, E(8cie) = 0, Vi?j.. In other words, I assume no crosscorre
lation between different units, but I allow for heteroskedasticity. Another observation
concerning the error terms is in order. The data used in the estimation are taken from
two Census years, 1980 and 1990, thus creating a short panel. Therefore, all the variables
in (11) should be indexed by a time subscript. However, the short panel feature allows me
to incorporate an unobserved, timeinvariant component into the error term:
Eit= pi + uit. The term rpiattempts to capture features of a specific location (unobservable
to the econometrician), that may attract or turn away agents with certain characteristics
that may be in turn correlated with the ability to find jobs. Insofar as these unobserved
characteristics are really timeinvariant over the decade under consideration, I can elimin
ate them by firstdifferencing the data. I follow this approach in the estimation, so the
auxiliary model (11) is to be taken in its firstdifference specification from now on, unless
otherwise noted.21 Hence, all time subscripts are suppressed.
In equation (11), the yN covariates are correlated with the error term E?, since they
are endogenous. Therefore, I use instruments for these variables: one obvious choice is to
use the exogenous variables in the neighbouring tracts to i. In particular, I use obser
vations in neighbours up to a distance 3 from tract i: so the instruments are
21. Therefore, y' really denotes Yit Yit1; xi denotes xit  xit; and the error term actually denotes the
difference in the timevarying part of the original error term, uit uit 1 .
TOPA SOCIAL INTERACTIONS 277
Xi[N ... .N3] The complete set of instruments can then be defined as
Zi[ [ N] . These variables are assumed to be uncorrelated with the error terms
E(zjEj)= 0. (12)
Equation (12) provides moment conditions that I can use to estimate the auxiliary model
of equation (11) via a GMM procedure.
The use of IV is not strictly necessary here. The auxiliary model is only a statistical
model that summarizes certain properties of interest of the structural model, and can be
misspecified. Thus, an equally valid procedure would be to estimate (11) via ordinary least
squares. This procedure would yield biased estimates for 4 and ,B,but it would not affect
the consistency of Vn'. However, using IV can be interpreted as adding several spatial
crosscorrelations of y and x to the auxiliary model (as it is, the model only includes
autocovariances of y). This is roughly equivalent, in a timeseries context, to adding cross
moments involving leads and lags of y and x.
Finally, I augment the auxiliary model bv a certain number of raw spatial moments
of the unemployment variable. The rationale for doing so is to add flexibility to the
auxiliary model, in order to be better able to detect signs of misspecification of the struc
tural model. As Tauchen (1996) shows, the X2 test of the overidentifying restrictions is
more likely to fail to detect misspecification if the auxiliary model is not flexible enough.
In addition to the mean and variance, I consider the spatial covariances up to distance
D, since the structural model delivers implications in terms of the spatial distribution of
unemployment: these are the pairwise autocovariances for pairs of tracts at given distance
d, whereas the 4 parameters in the SAR measure the autocovariance between unemploy
ment in tract i and the average unemployment in the neighbouring tracts. These additional
auxiliary parameters can be estimated together with those of the SAR using the same
GMM framework. Let v, be the vector of moments to be estimated
where Cd = Cov (yi yY), d= 1, ... , D. Here yd indicates the unemployment rate in tracts
that are at distance d from tract i. So the complete vector of auxiliary parameters is p
[? yr] . I also need to define
Zi  Ei(0, 0)
Yi 
2 _ 2 2
g(4s,p)YiY'12C (13)
I yD
_2C
where [Y jiN i2
i ... yf xN . The GMM estimator pn6MM is obtainedvia minim
ization of a quadratic criterion Jn(p) based on g(41, p), with the standard properties, and
the optimal choice of a weighting matrix is SA?', where S is the covariance matrix of the
limitingdistributionof 1/ S In?= I g(4i5 p).22
22. Conley (1999) providesa robust estimatorfor the covariancematrix S that does not rely on any
specificassumptionson the structureof the error terms, but only uses informationon economic distances
betweenlocations.Use of this robustestimatordoes not leadto significantlydifferentresultsthanmorestandard
estimators. Therefore, in what follows, I use Sn = 1/n n= g(4i, pn) *g(4, pn)
278 REVIEW OF ECONOMIC STUDIES
3.3. Identification
Identification issues are crucial in this empirical exercise. This paper argues that local
spillovers, generated by local social interactions, may explain part of the spatial patterns
of urban unemployment observed in the data. The estimation methodology basically
works by matching empirical spatial moments with simulated moments out of the local
interaction model, to estimate the structural parameters of the model and thus give a
measure of the magnitude, if any, of the local spillovers. However, there may be other
factors that give rise to the same patterns of positive spatial correlation in unemployment,
but may be indistinguishable from the social interaction channel that I focus on.
Formally, this discussion can be cast in terms of Manski (1993)'s analysis of identifi
cation issues in the context of endogenous social effects, such as peer influences, neigh
bourhood effects, social interactions, contagion, and the like. Restricting oneself for
simplicity to linear models, one can posit the following population relationship
where y is a scalar outcome variable, x and ? are attributes that directly affect the unit's
outcome (observed and unobserved,respectively), and A are attributes that define the indi
vidual unit's "reference group". In the present context, y is unemployment, x are tract
characteristics, and A is the distance metric that specifies which tracts are neighbours of
each tract i. Each term in (14) represents a separate effect: in addition to the direct effect
,Bof observable characteristics on outcome, 4 expresses an "endogenous social effect" that
the mean outcome in the reference group has on the individual unit's outcome, whereas y
represents an "exogenous or contextual effect" of the mean observable attributes in the
neighbouring tracts on tract i's unemployment rate. Furthermore, the error term E may
depend in part on unobservable characteristics that are themselves spatially correlated.
In the context of the linear model (14), Manski (1993) shows that the endogenous
social effects 4 can be identified if one assumes that there are neither exogenous effects
(y = 0) nor correlated effects (Eei?j = 0). I make these assumptions both in the structural
and in the auxiliary model. Specifically, in the structural model the conditional transition
probabilities (1)(5) only depend on tract i's own characteristics and on the neighbours'
employment rate in the previous period, while the neighbours'characteristics do not play
a role. In addition, the evolution of the system is ruled by random draws a) from a vector
of uniform distributions that are i.i.d. in time and in space by construction.
In the special case where exogenous and correlated effects are absent, 4 and ,B are
identified if the conditional expectation E(x IA) varies nonlinearly with A, and
Var (xIA)> 0 (Proposition 2 in Manski (1993)). In my model, x are tract characteristics,
whereas A is the physical distance metric. Therefore, Manski's identification conditions
are satisfied. In practice, I estimate E(x IA1) nonparametrically as a local average
I /Ni xj1 1, where Ni is the set of neighbours at distance 1 from i.
Furthermore, it is worth noting that even if neighbours' characteristics did play a
role in the evolution of the system, identification would still be ensured by the nonlinearit
ies implied by the stationary distribution of the structural model, as well as by the asym
metric way in which tract characteristics and information enter those conditional
transition probabilities (this point has been made by Brock and Durlauf (1999, p. 3133)).
Identification conditions for the structural parameters in ( ), ac( ), and y( ) can also
be expressed in terms of the indirect inference methodology. Intuitively, one wants to rule
out the possibility that the chisquare criterion (9) may be minimized by more than one
set of parameter values. Formally, this requirement amounts to assuming that the limit
TOPA SOCIAL INTERACTIONS 279
criterion JCO(G, 9o, p) has a unique minimum at p, that the binding function r( ) in (7) is
onetoone and that ar/aO is of full column rank. The rank condition can be locally
tested and is in fact satisfied at the point estimates reported in Section 5.23
Finally, with regard to the identification of the auxiliary model, I assume that the
error terms are uncorrelated in the crosssection, E(Ei?j) = 0, Vi ?j, and are uncorrelated
with the neighbouring tracts' characteristics: E(x7E?) = 0. The latter condition concerns
the validity of the instruments used in the auxiliary estimation. Both conditions can be
easily tested.
The assumptions made here, regarding the absence of exogenous or correlated effects,
may be considered too strong. For example, suppose criminal activity in one location
spills over to neighbouring locations, because of physical contiguity. An increase in crime
may induce certain types of residents to move out, who are more likely to be employed
than people who stay. Alternatively, the rise in crime may have an adverse effect on local
businesses and employers, forcing them to leave. Thus the crime level in the neighbouring
tracts would impact unemployment within those tracts, and have a contextual effect on
the unemployment rate in tract i (y? 0), generating positive spatial correlation in unem
ployment that is not due to social interactions. Another example is the location of schools.
Suppose high school quality in tract j induces agents with high ability or motivation to
locate in tract j as well as in neighbouring tract i. This may affect attributes such as
graduation rates or school dropout rates in both tracts, while at the same time reducing
unemployment in both tracts. Again one would observe a nonzero exogenous effect (y? 0)
in that attributes of neighbour j would be associated with outcomes of tract i . Finally, as
an example of correlated effects, one can think of positive sorting inducing positive corre
lation of certain unobserved attributes across neighbouring areas, that may directly affect
employment outcomes.
These are potentially serious violations of the identification assumptions. Therefore,
I follow three alternative strategies in the estimation of the structural model, to try to
distinguish spillovers that are due to social interactions across neighbouring tracts from
other possible intertract influences. The first two approaches exploit indirect information
that may be available on the dimensions along which social networks are constructed,
while the third addresses the issue of generic correlated effects across tracts.
There is considerable evidence in sociology on the extent of assortative matching in
agents' social networks. In particular Marsden (1987,1988), using data from the 1985
General Social Survey, shows that network homogeneity with respect to race and ethnicity
is very high.24Quite simply put, individuals are more likely to interact with people of the
same race or ethnicity than with members of different groups. The idea then is to divide
the set Ni of tracts physically adjacent to i into two subsets: those that are ethnically
"close" to tract i, and those that have a very different ethnic composition. Let me call the
latter subset EDNi.25 If social networks follow racial and ethnic lines, then one can expect
23. This is done by running a very long simulation of the structural model at the estimated value 0 H, and
numerically evaluating the matrix of partial derivatives ap/IO at the optimum OH. I can then test whether this
matrix has full rank.
24. For example, the likelihood of observing a social contact between two black persons is 4 2 times higher
than that generated by pure random matching, given the relative proportions of the different ethnic and racial
categories in the population.
25. Operationally, I use the same procedure as in Conley and Topa (2000) to calculate pairwise ethnic
distances between tracts. I consider nine different racial and ethnic groups to construct this metric. I then define
a set EDNi of "Ethnically Distant Neighbours" of i as those tracts in Ni whose ethnic distance from i is above
a certain threshold: 85 for 1980, 91 for 1990 (the maximum possible ethnic distance is 100v i). These thresholds
roughly correspond to the 55th percentile of the distribution of pairwise ethnic distances.
280 REVIEW OF ECONOMIC STUDIES
= a2(X,) + X(X,3 
PU N Yjt + XEDN(Xi) ED =
SYkl
 kt (15)
N1 EDN,
The otherupwardtransitionprobability(5) is modifiedaccordingly.If local spilloversare
mostly generatedby social exchanges,one would expect the local interactionchannelto
be weakerwith tract i's ethnicallydistant neighboursthan with its other neighboursin
Ni. On the otherhand, otherintertractinfluencessuch as crimeshouldnot be so affected
by racialand ethnicdistance:in the data, the spatialcorrelationof crimeacrossadjacent
tractsis 075 for areas that are ethnicallyclose, and 079 for ethnicallydistant tracts. In
termsof equation(15), this hypothesistranslatesinto XEDN(') < 0, and can be testedusing
the KCtest statistic.
The second approach relies on Hunter (1974)'s study of local communities in
Chicago. As I mentioned earlier,he identifiesneighbourhoodboundarieson the map
of Chicago that in many cases correspondto CommunityArea boundaries.Since local
neighbourhoodsare loci of social exchange,one would again expectsocial interactionsto
be weaker between tracts that are separatedby one of these boundaries.The way to
implementthis in the modelis similarto the ethnicdistancecase. For each tract i, I define
a subset NCAi of its adjacentneighboursas those tracts that are not in the same Com
munityArea as i and belong to a differentneighbourhoodaccordingto Hunter'sbound
aries. The transitionprobability(2) is modified in a manner similar to (15), with an
additionalinteractionterm XNCA (). The hypothesisis then 2NCA( () < 0.
As for the possibilityof correlatedeffects,I experimentwith the assumptionthat the
random shocks 0it are uncorrelatedin space, to see whether it affects the estimation
resultsin any way. In particular,I simulatethe structuralmodel introducinga certain
amount of spatial autocorrelationin the random draws woit, and look at whether the
introductionof this alternativesourceof spatialcorrelationin the model affectsthe par
ameterestimates,especiallythose of the social interactionterm 2( ).
TABLE 1
Summary statistics of all variables
1980 1990 199080
Variables Mean Std. Dev. Mean Std. Dev. Mean Std. Dev.
Unemployment rate 11 71 8 16 1493 12 06 3 22 8 95
Segregation index 87 20 13 03 9182 11 52 4 62 1432
Nonwhite persons (%/o) 50 55 3963 57 91 37 14 7.36 1459
Hispanic persons (%l) 7 19 1292 19 53 26 36 1234 17 24
Highschool grad. or more (%) 53 65 1694 62 53 17 20 8 88 1023
College grad. or more (%l) 1086 12 91 1562 17 37 4 77 8 62
Crime index 22 59 1551 19 39 13 50 3 20 8 04
Females (%) 52 91 6 63 51 81 5 29 1 10 5 68
Persons 1824 y.o. (%) 1935 670 1594 706 341 6 19
Persons 024 y.o. (%) 41 74 1203 3801 11 97 3 73 7 74
Persons per household (average) 2 77 0 62 2 82 0 67 005 040
Vacant housing units (%) 736 5 99 1048 8 18 3 12 8 14
Median gross rent 227 42 55 22 252 07 73 22 24 65 51 77
Average housing value 10 05 13 52 26 18 2230 16 13 19 60
Employed in prfs/mngr jobs (%) 9.59 891 1277 1155 318 712
Out of labour forcemales (%) 30 29 13 22 3048 1403 0.19 1208
Out of labour forcefemales(%) 51 01 1256 45 82 13 73 5 19 1187
Same county workers (%) 83 89 12 02 94 60 5 46 1071 1241
Median travel time to work 2726 6 38 27 99 6 52 073 5 96
See the Appendix for a definition of all the variables and units.
covariates. Summary statistics for all variables are presented in Table 1 and precise defi
nitions are given in the Appendix.
First of all, I consider a set of sorting variables, i.e., variables that may affect the
decisions by different types of individuals to locate in a given area. These include average
housing values in the Census tract, median gross rents, the fraction of vacant housing
units in the area, a crime index, the fraction of persons with managerial or professional
jobs, the percentage of nonwhite persons, the percentage of Hispanic persons, a segre
gation index, and the number of persons per household.
Secondly, I consider variables that may be linked more directly to the probability of
being employed. These include the percentage of persons with at least a high school dip
loma, the fraction of persons with at least a college degree, the age composition in the
tract (to proxy for potential experience), the fraction of females 16 years and older, and
the percentage of males and females out of the labour force in the tract.
Finally, there is a relatively large literature in urban and labour economics that dis
cusses the spatial mismatch hypothesis. This literature aims at explaining the high unem
ployment levels in mostly black, inner city neighbourhoods by local labour market
conditions. The basic idea is that during the 1970's and 1980's many jobs (especially low
skill ones in the service industry) have moved from central city areas to the suburbs. In
addition, the contention is that there is low residential mobility and a certain degree of
housing segregation for innercity blacks. For example, it may be very costly for a black
household to relocate to the suburbs in a mostly white neighbourhood, where the social
capital provided by a black community would be missing. Several authors, such as Holzer
(1991) and Ihlanfeldt and Sjoquist (1990,1991) have analysed this issue empirically. By
now, there is a certain consensus that physical proximity to jobs explains a portion of
black/white unemployment differences, for instance. Therefore, I include the median com
muting time to work for workers who live in each tract, and the fraction of residents who
go to work in the same county.
282 REVIEW OF ECONOMIC STUDIES
ynw= 8anw
Yed= i50ed,
Ycr = _3ac X
28. The initial condition does not really matter, since there exists a unique stationary distribution.
29. The auxiliary estimation actually uses firstdifferenced data. To make the simulations y consistent with
this, I use the following strategy. For each value of 0, I first run the simulation fixing the X characteristics at
their 1980 levels. This yields a simulated unemployment realization for 1980, Y(0)80. I then repeat the simulation
for the same 0 (and the same sequence of shocks) using the 1990 values of X. This delivers Y(0)go. I then
compute Y(0)go80Y(0)go Y(0)80. This is the simulated counterpart to the dependent variable of the auxiliary
regression on the data.
TOPA SOCIAL INTERACTIONS 283
where 3 > 0. All these manipulations imply that the actual vector of structural parameters
to be estimatedis 0 =_Ronw Xed Xcr a0 anw aed acr 3] . This vector is augmentedby
EFDN or XNCA in the model specifications that allow the intercept of the local interaction
term to differ for ethnically distant neighbours or for neighbours in different Community
Areas and different "Hunter neighbourhoods".
I use a simulated annealing algorithm to minimize the indirect inference criterion (9)
with respect to the parameters 0. This algorithm is explicitly designed to find a global
minimum of functions that may present multiple optima and discontinuities. Therefore,
it is particularly well suited for the problem at hand. Very briefly, simulated annealing
performs a random search over the parameter space, and accepts not only downhill moves
but also uphill moves. The probability of accepting an upward move depends positively
on a "temperature" parameter that decreases as the search progresses. At the beginning
of the search, the algorithm is allowed to make large upward moves, and thus searches
over the whole parameter space. As the temperature drops, the algorithm concentrates on
more promising regions, but the random nature of the search still allows it to escape local
minima. Simulated annealing has proven to be very effective in this particular problem.30
5. EMPIRICAL RESULTS
As a preliminary step, I report in Table 2 the results of the estimation of the auxiliary
model, performed on the actual unemployment data in first differences. It is worth noting
that, even after controlling for tract characteristics as well as timeinvariant tractspecific
fixed effects, the unemployment rate of neighbouring tracts has a positive and significant
parameter estimate, at least for tracts up to a distance 2 from tract i. Even though the
auxiliary parameters are not the actual parameters of interest, they still give an indication
that unemployment is characterized by positive spatial correlations: this in turn is consist
ent with the model of local interactions.
Figure 7 also reports the nonparametric ACF estimates for the dependent variable
and the fitted residuals of the auxiliary regression. The solid line refers to the actual
unemployment rate in first differences. The dashed line refers to the fitted residuals,
ii (Yi) 'GMM(x1) GMM. Finally,the intermediateline showsthe ACF of a differ
ent set of fitted residuals, calculated excluding from the covariates the direct effect of
unemployment in the neighbouring tracts: ij=yi  (xi) GMM. The circles, diamonds and
asterisks highlight the portions of the ACFs where one can reject the null hypothesis of
spatial independence, at the 5% significance level.31
It is interesting to note that both the raw unemployment variable and the fitted
residuals 'i display a positive and statistically significant amount of spatial correlation,
that declines monotonically with distance. On the other hand, the fitted residuals Ei do
not exhibit any positive spatial correlation and are close to being white noise (in space),
except for a small amount of negative autocorrelation at distances 1 and 2. Therefore, it
seems that the assumption of no crosscorrelation of the residuals across individual units
is a reasonable one.
Tables 3 and 4 present the results of the indirect inference estimation of the structural
parameters. The model is estimated under four different specifications. The first is the
baseline model, with conditional transition probabilities as in equations (1)(5). The
30. For a moredetaileddescriptionof the algorithm,see Goffe (1996)and Goffe et al. (1994).I am very
gratefulto Bill Goffe for sharinghis MATLABsimulatedannealingroutineswith me.
31. See Conleyand Topa (2000)for the detailsof the ACF estimationand the bootstraptest of spatial
independence.
284 REVIEW OF ECONOMIC STUDIES
TABLE 2
Auxiliary regression on the data
Dependent variable:
Unemployment rate, 199080
Variable name rhogmm S.E. Included?
Unemployment rate (nbs1) 0 6959 0.1637 Yes
Unemployment rate (nbs2) 04351 0.1444 Yes
Unemployment rate (nbs3) 0.3236 0 1760 Yes
Unemployment rate (nbs4) 0 0891 0 2634 Yes
Unemployment rate (nbs5) 0 3877 0 3880 Yes
Unemployment rate (nbs6) 0 2317 0 3160 Yes
Unemployment rate (nbs1) x
Nonwhite persons (%o) 0 0013 0 0133 Yes
Unemployment rate (nbs1) x
College grad. or more (%) 0 0462 0 0166 Yes
Unemployment rate (nbs1) x
Crime index 0 0136 0 0139 Yes
Constant 05943 0 8129 No
Segregation index 0 0151 0 0240 No
Nonwhite persons (%) 0.0111 0 0538 Yes
Hispanic persons (%) 00524 00258 No
Highschool grad. or more (%/6) 0 0336 0 0377 No
College grad. or more (%/6) 0 0269 0 0671 Yes
Crime index 0 0420 0 0681 Yes
Females (%) 0 1780 0.0798 No
Persons 1824 y.o. (%) 0 1732 00749 No
Persons 024 y.o. (%) 0 1073 0 0927 No
Persons per household (average) 0 2168 01313 No
Vacant housing units (%o) 0.0758 0 0497 No
Median gross rent 0 0022 00078 No
Average housing value 0 0100 0 0141 No
Employed in prfs/mngr jobs (/o) 02467 0 0967 No
Out of labour forcemales (%) 00628 0.0340 No
Out of labour forcefemales (%o) 0 0610 0.0443 No
Same county workers (%o) 0 0235 0 0307 No
Median travel time to work 0 1615 0 0738 No
Estimated moments of the
Dependent variable
rhogmm S.E. Included?
Mean 3 0466 02577 Yes
Variance 48 0445 8 0315 Yes
Spatial Cov(l) 170045 3 5405 Yes
Spatial Cov(2) 118436 2 3545 Yes
Spatial Cov(3) 8.4629 1 8718 Yes
Spatial Cov(4) 6 3632 1 4791 Yes
Spatial Cov(5) 5 7835 1 3146 Yes
Spatial Cov(6) 3 9162 1.2448 Yes
second takes into account the possibility that local interactions may be weaker across
tracts that are ethnically very distant, as in equation (15): in practice, only the intercept
of the 2(4) term is allowed to vary, but the additional parameter )ODN is estimated separ
ately for 1980 and 1990. The third model specification is similar to the previous one, but
considers tracts that belong to different Community Areas and different neighbourhoods
TOPA SOCIAL INTERACTIONS 285
  Resid GMM
05 0 5% signif.
   Resid (X only)
0 5% signif.
04 Raw unempl.
* 5% signif.
03
Q002
0.1
0~~~~~~~~~0
10
01 ~C or
0 21
0 2 4 6 8 10
Physical distance (adjacenttracts)
FIGURE 7
32. The auxiliary parameters p based on simulations performed at the optimal value nHare very close to
the auxiliary parameters p from the data. These comparisons are reported in a table available from the author
upon request.
286 REVIEW OF ECONOMIC STUDIES
TABLE 3
Structuralparameter estimates
Parameter Baseline Ethnicities Comm. areas lambda 0 Corr. shocks lambda 0
lambda_0 02495 0.2057 0 1966 0 3001
(00104) (0.0103) (0.0211) (0.0254)
lambda_EDN(80) 00024
(0.0017)
lambda_EDN(90) 0 2094
(0.0226)
lambda_NCA 0 0159
(0 0046)
lambda_ed 01049 0 1558 0 0922 0 1691
(0 0109) (0 0175) (0 0182) (0 0312)
lambda_nw 0 0910 0 1101 0 2718 01657
(0 0107) (0 0090) (0 0328) (0 0157)
lambda_cr 0 2592 0 3262 0 3583 0 3510
(0 0257) (0 0175) (0 0165) (0 0470)
alpha_0 0 3919 0 3724 0 3719 0 6505 0 7528 0 2313
(0 0170) (0 0139) (0 0225) (0 0158) (0 0352) (0 0064)
alpha_ed 0 0778 00814 00926 0 0737 0 1662 0 1319
(0 0046) (00068) (00095) (0 0054) (0 0274) (0 0174)
alpha_nw 01248 0 1017 0 1234 0 1109 00600 0 0614
(00040) (00052) (0 0064) (00049) (00054) (0 0031)
alpha_cr 0 0401 0 0289 0 0392 02473 0 2009 00866
(0 0048) (0 0016) (0 0029) (00096) (00238) (0 0063)
delta 0 6483 0 6687 0 6784 0 4169 0 5754 0 3128
(0 0171) (0 0113) (0 0118) (0 0066) (0 0274) (00045)
Ktest 12 0666 6 0698 10 3123 20 0642 18 4197 18 7334
pvalue (0 3586) (0 7329) (0 4135) (0 1695) (0 0723) (0 2261)
Kctest
(Ho: lambda =0) 7 9976 13 9944 9 7519 0 3137
pvalue (0 0917) (0 0297) (0.0826) (0 9889)
Kctest
(Ho: lambda_EDN= 0) 5 9967
pvalue (0 0499)
Kctest
(Ho: lambda_NCA = 0) 1 7542
pvalue (0 1853)
Furthermore, I can characterize how the information spillovers vary with tract character
istics: the interaction term X( ) has a positive intercept, is decreasing in the education
level and in the crime rate, and is increasing in the percentage of minorities in the tract.
Thus the spillover effects are stronger for areas with lower education levels and with
a higher percentage of nonwhite residents. This is consistent with empirical findings
reported by Corcoran, Datcher and Duncan (1980) and by Granovetter (1995). This litera
ture focuses on the nature of informal hiring channels and is based on detailed data on
employees' work history (in particular, how they were hired). The authors find that infor
mal contacts used to acquire jobs are more important for younger workers, lowskilled
jobs, less educated workers and minorities. An alternative explanation for the reported
signs of X,A, and Xed could be related to the choice of physical distance as a proxy for
social distance, in the structural model. Since social networks of poorer, less educated
agents tend to be more geographically concentrated (see Fischer (1982)), it is quite intuit
ive that the local interaction effect is stronger for tracts with these characteristics, since I
TOPA SOCIAL INTERACTIONS 287
am focusing on the geographic component of social networks. Finally, the sign of X,, may
indicate that local networks and institutional loci of social interactions such as neighbour
hood organizations tend to break down in the presence of high crime rates in the neigh
bourhood, perhaps because crime raises the cost of maintaining local social ties.
Columns 2 and 3 in Table 3 report the parameter estimates for the model specifica
tions that take into account ethnic distance and neighbourhood boundaries. Again, the
overall ic, specification test does not reject the structural model in either case, with very
high pvalues. The ethnic distance version fits the data particularly well. As for the joint
test of the null hypothesis that the X parameters are all equal to zero, one can reject the
null even more strongly than in the baseline model (the pvalues are now 0 03 and 0.08).
The estimates of the X and a parameters are similar to the baseline model.
The interesting result is that the signs of the additional parameters )EDN and 2?NCA
go in the direction expected according to the theory: the local interaction channel is indeed
weaker across tracts with very different ethnic compositions (at least for 1990), and for
tracts that are separated by neighbourhood boundaries. The first result is especially strong,
since one can rejectthe null hypothesisthat XODN = 0 at the 5%level.
Therefore, there is some evidence that intertract influences follow ethnic lines and
neighbourhood boundaries that were identified by local residents. Since social networks
exhibit strong assortative matching with regard to ethnicity and are thought to be stronger
within, rather than across neighbourhoods, this evidence lends support to the idea that
the local spillovers detected here are indeed caused by social interactions rather than by
other factors that may generate positive spatial correlations.
The last two columns of Table 3 introduce a note of caution. Although the point
estimate of the 2(4) intercept is quite positive and significant, one can no longer reject the
null hypothesis that the )( ) term is identically equal to zero. However, the model with
correlated shocks fails to pass the overall Kc,specification test, and in general provides a
worse fit of the data than the model specifications with i.i.d. shocks. Figures 8 and 9
contain the simulated maps of unemployment in the ethnic distance specification with
i.i.d. shocks and in the correlated shock specification, respectively: these can be compared
to the actual data, in Figure 3. The signs of the parameter estimates remain unchanged,
with the notable exception of )L, which reverses sign: thus the interaction effect would
be weaker for tracts with higher fractions of minorities. It may be that the degree of
autocorrelation introduced in the shocks used to simulate the model is too high and simply
washes out everything else. This issue certainly warrants further examination.
In Table 4, I use the parameter estimates of the structural model to perform several
thought experiments, aimed at evaluating the magnitude of the spillovers generated by
local interactions. I consider different types of tracts: the first is an artificial average tract,
in the sense that it has the citywide averages of unemployment, education, presence of
nonwhites, and crime. The other four tracts are real ones, taken from four different
neighbourhoods:33 Woodlawn and Grand Boulevard are characterized by high unemploy
ment, high poverty rates, relatively low education levels and relatively high crime rates,
and are mostly nonwhite. On the other hand, Brighton Park and Lake View have low
unemployment, low poverty rates, low crime, high education levels, and are mostly white.
The contrast between Grand Boulevard and Lake View is especially strong, whereas the
differences between Woodlawn and Brighton Park are not as marked.34
rate,9080:SIM(EDN)
Unemnployment 2 0 2 4 Miles
Lii66.71.9
3.27.8
7.872.9
FIGURE 8
Map of simulatedunempl.rate, 199080 (EDN)
rate,9080:SIM(COR)
Unemnployment 2 0 2 4 Miles
3278
78729
FiGuRE 9
Map of simulatedunempl.rate, 199080 (COR)
TABLE 4
Magnitude of spillovers
Baseline Ethnicities Comm. areas Corr. shocks
Experiment 1980 1990 1980 1990 1980 1990 1980 1990
Average tract:
Raise Info. by 1 s.d. 0 63 1 26 045 114 0 68 1 38 0 11 0 39
Raise Educ. by 1 s.d. 0 81 0 80 048 047 1 30 1 27 1 28 1 29
Lower nw by 1 s.d. 469 422 361 317 121 100 385 360
Lower Crime by 1 s.d. 2 74 2 33 3 65 3 00 3 34 2 85 407 3 54
GrandBoulevard:
Raise Info. by 1 s.d. 0.91 1.53 0 66 1 27 1.11 1 84 0 24 0 05
Same Info. as Lake View 2 86 4 77 2.10 3 99 3 68 5 81 0 79 0 17
Raise Educ. by 1 s.d. 0 86 0 89 049 0 62 1 28 1 28 1.50 1 44
Lower nw by 1 s.d. 5 05 444 3 74 3 29 1 14 1 40 4.68 3 58
Lower Crime by 1 s.d. 3 08 227 3 86 2 68 3 34 2 55 493 3 55
Lake View:
Lower Info. by 1 s.d. 0 57 0 77 0 36 0 56 0 57 0 69 0 15 0 41
Same Info. as Grand Blvd. 1 89 2 68 1 17 1 91 2 17 2 58 0 48 1 39
Raise Educ. by 1 s.d. 0 85 0 84 0 50 0 56 1.39 1.39 1 29 1 18
Lower nw by 1 s.d. 4 85 4 15 3 81 3 30 1 30 1.39 3 83 295
Lower Crime by 1 s.d. 2 78 205 3 86 282 3 54 2 85 406 293
Woodlawn:
Raise Info. by 1 s.d. 078 1 27 0.59 1 37 091 1.42 0 11 024
Same Info. as Brghtn. Pk. 0 61 1.19 0 45 2 55 0 61 1 29 0 08 0 22
Raise Educ. by 1 s.d. 072 067 035 041 111 098 128 110
Lower nw by 1 s.d. 442 348 324 276 085 099 430 288
Lower Crime by 1 s.d. 2 85 1 88 3 64 2 62 3 01 2 03 4 50 2 85
Brighton Park:
Lower Info. by 1 s.d. 0 51 1 21 0 30 1 22 0 42 1 27 0 25 0 64
Same Info. as Woodlawn 0 39 1 13 0 22 2 70 0 21 1 11 0 19 0 60
Raise Educ. by 1 s.d. 0 89 0 83 0.59 0.51 1.51 1 37 1 26 1.24
Lower nw by 1 s.d. 485 423 389 331 1 52 1.09 3 51 333
Lower Crime by 1 s.d. 2 58 221 3 58 3 06 3.76 3 06 3 74 3 28
Each cell reports the expected change in employment, in percentage points, following each experiment.
information, the size of the spillover effect for Lake View is roughly half the size of the
effect for Grand Boulevard (in absolute value). Thus a tract in Lake View would not
suffer from a decrease in information levels as much as a tract in Grand Boulevard would
gain from an increase in information of the same size.
Finally, giving the tract in Grand Boulevard the same information as the tract in
Lake View has a large effect on unemployment: the expected unemployment rate would
decrease by almost three percentage points in 1980, and by more than four points in 1990.
Again, the reverse experiment on Lake View would raise unemployment by a smaller
amount, because of the different tract characteristics that affect the strength of the local
interactions.
The magnitudes of these effects remain roughly unchanged across the different
model specifications with i.i.d. shocks: the size of the local interaction spillover is
largest under the specification with neighbourhood boundaries (except for the Wood
lawn/Brighton Park exchange). With correlated shocks, the spillover effect is greatly
reduced for the average tract, and reverses sign in some cases, especially those regarding
Grand Boulevard. This counterintuitive result may be due to the sign of 2,. under this
specification.
TOPA SOCIAL INTERACTIONS 291
See
Poor
the CrimeCollege Median
Nonwhite
Households
Highschool
indexgrad.
or on households
Unemployment
Appendix grad.
persons household
for or (%o) rate
more
a (%) welfare
(%) (%) income
more
(%)
definition
of
all
the
99 2 30 54 4 25 Tract Grand
442
61 83 44 1943
09 41 82
variables 3801
Blvd.
and
units.18
3
34 65 0 15 17 3 Tract Lake
23 2706
58 76 23 27 42
612View
Summary
1980
31 91 46 14 5 23 Tract
67 statistics
52 68 08 4811
51 13
4202 for
Woodlawn
TABLE
5
selected
19 52 12 17 Tract
6 11.8 7 8 51 Brighton
736
53 072
5801 Census
Pk.
tracts
40 99 41 15 68 2 Tract
Grand
5
327
46 88 88 87 3546
3801
Blvd.
10 58 0 8 22 Lake
Tract
7209
12 8846
05 39 04 569
612View
1990
12 87 17 55 7 43 7 33 Tract
9
89 83 03 93 06 52 18
4202
Woodlawn
29 5 14 17 9 Tract
131
21 49 518529 05 31
023 Brighton
5801
Pk.
292 REVIEW OF ECONOMIC STUDIES
6. CONCLUSION
This paper analyses a model of local social interactions in the context of urban unemploy
ment, and provides empirical estimates of the size of the spillovers generated by local
interactions. The underlying idea is that agents exchange information within their social
networks about job opportunities. There exists a local positive feedback in that agents are
more likely to transmit useful information about jobs if they themselves are employed.
This feedback generates positive spatial correlations of unemployment.
Insofar as the dimensions along which social networks are constructed are at least
partially observed by the econometrician, the information exchange process generates
observable implications that can be used to detect the existence of local interactions. In
this paper, I argue that physical distance is an important determinant of social ties, and
use the observed spatial patterns in unemployment according to this metric to estimate
the structural model. I also argue that Census tracts constitute a meaningful unit of analy
sis, since they still provide a fine enough grid to look at interactions within
neighbourhoods.
The results of the estimation support the hypothesis that there exist local social inter
actions across tracts. At least in the case of i.i.d. shocks, one can reject the hypothesis
that the interaction term is identically equal to zero, and the shape of the interaction term
is consistent with independent evidence on informal hiring channels: in particular, the
spillovers are stronger for tracts with lower education levels, higher fractions of minorities,
and lower crime rates. A thought experiment also reveals an interesting asymmetry: if one
raises the level of information available to a disadvantaged tract by one standard deviation
and lowers the information flow to a welloff tract by the same amount, the decrease in
expected unemployment in the former area is twice as large as the increase in unemploy
ment in the latter. This is due to the different "initial conditions" in the two areas, in
terms of education levels, racial composition and crime rates.
In addition, the local spillover effect is weaker between tracts with very different
ethnic compositions, and between tracts that belong to different neighbourhoods. Since
one expects social ties to be stronger between ethnically homogeneous groups and within
neighbourhoods, this result seems to be consistent with a social interaction interpretation
of the observed spatial patterns in unemployment, rather than with other possible inter
tract linkages.36
Further research should be done to clarify certain remaining issues. The model speci
fication with correlated shocks cannot reject the null hypothesis that the interaction term
is zero and generates a very small local spillover, but fits the data more poorly than the
other model specifications. It would be worthwhile to experiment further with different
forms and degrees of autocorrelation. One may also try to think of alternative ways to
distinguish social interactions from other effects (e.g. crime spillovers or school spillovers)
by using police precinct or school district boundaries. Finally, the structural model has
implications for the evolution of the system over time that could be tested using panel
data with several time periods. These timeseries properties would provide an additional
set of moments to be matched, and may help clarify some of the forces that determine
the rise and fall of neighbourhoods.
36. As for the exact mechanismsthroughwhich social interactionsoperate,the data are not suitableto
determinewhethertheyworkthroughdirectreferralsto employers,pureinformationexchangesaboutjob open
ings, or role model effectsand peerinfluencesexertedby one's employedsocial contacts.Individuallevel data
may be more appropriatefor this kind of analysis.
TOPA SOCIAL INTERACTIONS 293
APPENDIX
The following are the definitions of the variables used in the paper:
Unemploymentrate: the percentage of unemployed persons over the civilian labour force (16 years and
older). Unemployed persons are individuals who were neither "at work" nor "with a job but not at
work" during the Census reference week, and who were actively looking for work during the last four
weeks prior to the reference week;
Segregation index: the Euclidean distance of a specific tract from the citywide mix of ethnicities: if
(a, b, h, g,1,v) are the citywide proportions of Asian and Pacific islanders, Blacks, Hispanics, Others,
and Whites (respectively) over the whole population, then the index for tract i is defined as:
segri_
1(ai _
# +(bi _b)2+ (hifi)2 + (oi _ o)2 + (Wi _
w)2;
Acknowledgements. I would especially like to thank Jose Scheinkman, Lars Hansen and Gary Becker for
their encouragement and guidance. Comments and suggestions from Roland Benabou, Tim Conley, Steven
Durlauf, Chris Flinn, Jim Heckman, Mike Keane, Pat Kehoe, Steve Levitt, Charles Manski, Derek Neal, Tano
Santos, Wilbert van der Klaauw, Ken Wolpin, the editor, two anonymous referees and participants at workshops
at Chicago, NYU, Boston University, Duke, U.Penn., NBER, SITE, and Toulouse were very helpful. The
simulated annealing Matlab codes were kindly provided by Bill Goffe. Thanks to the ICPSR for providing the
294 REVIEW OF ECONOMIC STUDIES
Bloch and Bloch Chicago homicide data. Financial support from the C.V. Starr Center at NYU is gratefully
acknowledged. All remaining errors are of course mine.
REFERENCES
AUDRETSCH, D. B. and FELDMAN, M. P. (1996), "R&D Spillovers and the Geography of Innovation and
Production", American Economic Review, 86, 630640.
BALA, V. and GOYAL, S. (1998), "Learning from Neighbours", Review of Economic Studies, 65, 595622.
BANERJEE, A. V. (1992), "A Simple Model of Herd Behavior", Quarterly Journal of Economics, 107, 797
817.
BECKER, G. S. and MURPHY, K. M. (1994), "The Sorting of Individuals into Categories when Tastes and
Productivity Depend on the Composition of Members" (Unpublished manuscript, University of Chicago).
BENABOU, R. (1993), "Workings of a City: Location, Education, and Production", QuarterlyJournal of Econ
omics, 108, 619652.
BENABOU, R. (1996), "Equity and Efficiency in Human Capital Investment: The Local Connection", Review
of Economic Studies, 63, 237264.
BERTRAND, M., LUTTMER, E. F. P. and MULLAINATHAN, S. (1999), "Network Effects and Welfare
Cultures" (Unpublished manuscript, University of Chicago).
BIKHCHANDANI, S., HIRSHLEIFER, D. and WELCH, I. (1992), "A Theory of Fads, Fashion, Custom,
and Cultural Change as Informational Cascades", Journal of Political Economy, 100, 9921026.
BROCK, W. A. and DURLAUF, S. N. (1999), "InteractionsBased Models", in Heckman, J. and Leamer, E.
(eds.), Handbook of Econometrics (Amsterdam: North Holland).
BURT, R. S. (1992), Structural Holes: The Social Structure of Competition, (Cambridge: Harvard University
Press).
CASE, A. C. and KATZ, L. F. (1991), "The Company You Keep: the Effects of Family and Neighborhood on
Disadvantaged Youths" (NBER Working Paper no. 3705).
COLEMAN, J. S. (1988), "Social Capital in the Creation of Human Capital", American Journal of Sociology,
94, S95S120.
COLEMAN, J. S., CAMPBELL, E., HOBSON, J., McPARTLAND, J., MOOD, A., WEINFELD, F. and
YORK, R. (1966) Equality of Educational Opportunity(Washington DC: US Government Printing Office).
CONLEY, T. G. (1999), "GMM Estimation with Cross Sectional Dependence", Journal of Econometrics
(forthcoming).
CONLEY, T. G. and TOPA, G. (2000), "SocioEconomic Distance and Spatial Patterns in Unemployment"
(Mimeo., New York University).
CONLEY, T. G. and TOPA, G. (2001), "Dynamic Behavior of Local Interaction Models" (Work in progress,
New York University).
CONNERLY, C. E. (1985), "The Community Question: An Extension of Wellman and Leighton", Urban
Affairs Quarterly, 20, 537556.
CORCORAN, M., DATCHER, L. and DUNCAN, G. (1980), "Information and Influence Networks in Labor
Markets", in Duncan, G. and Morgan, J. (eds.), Five ThousandAmerican Families. Patterns of Economic
Progress, (Ann Arbor, MI: Institute For Social Research) 7, 137.
CRESSIE, N. A. C. (1993) Statistics for Spatial Data (New York: John Wiley and Sons).
DURLAUF, S. N. (1996a), "A Theory of Persistent Income Inequality", Journal of Economic Growth, 1, 75
93.
DURLAUF, S. N. (1996b), "Neighborhood Feedbacks, Endogenous Stratification, and Income Inequality", in
Barnett, W., Gandolfo, C. and Hillinger, C. (eds.), Dynamic DisequilibriumModelling: Proceedings of the
Ninth InternationalSymposium on Economic Theory and Econometrics (Cambridge: Cambridge University
Press).
ERBE, W. et al. (1984) Local Community Fact Book: Chicago Metropolitan Area (Chicago: Chicago Review
Press).
FISCHER, C. S. (1982) To Dwell Among Friends. Personal Networks in Town and City (Chicago: University of
Chicago Press).
GALE, D. and ROSENTHAL, R. (1999), "Experimentation, Imitation, and Strategic Stability", Journal of
Economic Theory, 84, 140.
GLAESER, E. L., SACERDOTE, B. and SCHEINKMAN, J. A. (1996), "Crime and Social Interactions",
QuarterlyJournal of Economics, 111, 507548.
GOFFE, W. L., FERRIER, G. D. and ROGERS, J. (1994), "Global Optimization of Statistical Functions with
Simulated Annealing", Journal of Econometrics, 60, 6599.
GOFFE, W. L. (1996), "SIMANN: A Global Optimization Algorithm using Simulated Annealing", Studies in
Nonlinear Dynamics and Econometrics, 1, 169176.
GOURIEROUX, C., MONFORT, A. and RENAULT, E. (1993), "Indirect Inference", Journal of Applied
Econometrics, 8, S85S118.
GRANOVETTER, M. S. (1995) Getting a Job: A Study of Contacts and Careers, 2nd edition (Chicago: Univer
sity of Chicago Press).
HANSEN, L. P. and SARGENT, T. J. (1993), "Seasonality and Approximation Errors in Rational Expectations
Models", Journal of Econometrics, 55, 2155.
TOPA SOCIAL INTERACTIONS 295
HANUSHEK, E., KAIN, J. F., MARKMAN, J. and RIVKEN, S. (2000), "Do Peers Affect Student
Achievement?" (Unpublished manuscript, University of Rochester).
HARRIS, T. E. (1974), "Contact Interactions on a Lattice", The Annals of Probability, 2, 969988.
HOLZER, H. J. (1991), "The Spatial Mismatch Hypothesis: What Has The Evidence Shown?", Urban Studies,
28, 105122.
HUNTER, A. (1974) Symbolic Communities: The Persistence and Change of Chicago's Local Communities
(Chicago: University of Chicago Press).
IHLANFELDT, K. R., and SJOQUIST, D. L. (1990), "Job Accessibility and Racial Differences in Youth
Employment Rates", American Economic Review, 80, 267276.
IHLANFELDT, K. R. and SJOQUIST, D. L. (1991), "The Effect of Job Access on Black and White Youth
Employment: a CrossSectional Analysis", Urban Studies, 28, 255265.
IOANNIDES, Y. M. and DATCHER LOURY, L. (1999), "Job Information Networks, Neighborhood Effects
and Inequality" (Unpublished manuscript, Dept. of Economics, Tufts University).
JENCKS, C. and MAYER, S. E. (1990), "The Social Consequences of Growing Up in a Poor Neighborhood",
in Lynn, L. and McGeary, M. (eds.), InnerCity Poverty in the United States (Washington, DC: National
Academy Press), 111186.
LIGGETT, T. M. (1985) Interacting Particle Systems (New York: SpringerVerlag).
LIGON, E., THOMAS, J. P. and WORRALL, T. (1999), "Mutual Insurance with Limited Commitment: The
ory and Evidence from Village Economies" (Unpublished manuscript, UC Berkeley).
LUDWIG, J., DUNCAN, G. J. and HIRSCHFIELD, P. (1999), "Urban Poverty and Juvenile Crime: Evidence
from a Randomized HousingMobility Experiment" (Unpublished manuscript, Georgetown University).
MANSKI, C. F. (1993), "Identification of Endogenous Social Effects: The Reflection Problem", Review of
Economic Studies, 60, 531542.
MARSDEN, P. V. (1987), "Core Discussion Networks of Americans", American Sociological Review, 52, 122
131.
MARSDEN, P. V. (1988), "Homogeneity in Confiding Relations", Social Networks, 10, 5776.
MONTGOMERY, J. D. (1991), "Social Networks and LaborMarket Outcomes: Toward an Economic Analy
sis", American Economic Review, 81, 14081418.
MORRIS, S. (1997), "Contagion" (CARESS Working Paper 9701, University of Pennsylvania).
MUNSHI, K. and MYAUX, J. (1998), "Social Effects in the Demographic Transition: Evidence from Matlab,
Bangladesh" (Unpublished manuscript, University of Pennsylvania).
RAUCH, J. E. (1993), "Does History Matter Only When It Matters Little? The Case of CityIndustry Location",
QuarterlyJournal of Economics, 108, 843867.
TAUCHEN, G. (1996), "New Minimum ChiSquare Methods in Empirical Finance" (Unpublished manuscript,
Duke University).
THOMAS, J. P. and WORRALL, T. (1988), "SelfEnforcing Wage Contracts", Review of Economic Studies,
55, 541554.
WELLMAN, B. (1996), "Are Personal Communities Local? A Dumptarian Reconsideration", Social Networks,
18, 347354.
WILSON, W. J. (1987) The Truly Disadvantaged: The Inner City, the Underclass, and Public Policy (Chicago:
University of Chicago Press).
ZAX, J. S. and REES, D. I. (1999), "Environment, Ability, Effort, and Earnings" (Unpublished manuscript,
University of ColoradoBoulder).