1 INTRODUCTION
The relevance of social networks and local interactions for economic outcomes has been
increasingly recognized by economists in a variety of contexts.1 An important strand of this
literature has focused on the detection and measurement of social interactions that operate at the
level of the residential neighborhood.2 The proper identification of such neighborhood effects is
complicated, however, by the non-random sorting of households into neighborhoods and the
likely presence of unobserved individual and neighborhood attributes.3 The resulting correlation
in unobservables among neighbors can lead to serious bias in the estimation of social interaction
among neighbors in the absence of a research design capable of distinguishing social interactions
from these alternative explanations.4
In this paper, we propose a new empirical approach designed to identify neighborhood
effects in observational data by isolating block-level variation in the characteristics of neighbors
within narrowly-defined neighborhoods. In particular, using Census data that detail the block on
which each individual in the Boston metropolitan area resides, we compare outcomes for
neighbors that reside on the same versus nearby blocks, where nearby blocks are defined to be
those in the same Census block group.5 The key identifying assumption underlying this design
(testable on observable attributes) is that there is no block-level correlation in unobserved
attributes within block groups.
We use this approach to study the impact of neighborhood referrals on labor market
outcomes. Rather than focusing on more general forms of neighborhood effects, we exploit the
fact that our restricted Census dataset characterizes the precise location of both an individual’s
place of residence and place of work to study the propensity of neighbors to work together.
1 Some recent examples include crime (Glaeser et al. (1996), Bayer et. al. (2004)); welfare program
participation (Bertrand et al. (2000)); the adoption of new technologies (Conley and Udry (2003), Bandiera
and Rasul (2003), Burke et al. (2004)); peer effects in education (Hoxby (2000), Sacerdote (2001),
Zimmerman (2003), Zax and Rees (2002)); knowledge spillovers and economies of agglomeration (Jaffe et
al. (1993), Audretsch and Feldman (1996), Glaeser et al. (1992)). For a more extensive review of the
literature, both theoretical and empirical, see Brock and Durlauf (2001).
2 Case and Katz (1991) explore the role of neighborhood effects on several behavioral outcomes using a
spatially auto-regressive model. Crane (1991) also looks at neighborhood influences on social pathologies,
focusing on non-linearities and threshold effects. Jencks and Mayer (1990) present a survey of the older
literature on neighborhood effects.
3 See Manski (1993) and Moffitt (2001) for a general discussion of the identification of social interactions
in the presence of correlated unobservables.
4 The recent literature on neighborhood effects has focused on the development and use of research
methodologies designed to distinguish among these potential explanations. We provide a detailed
discussion in Section 2 below.
5 Census block groups are defined by the Census Bureau and contain an average of ten contiguous city
blocks in our sample.