participation for the average individual by 1.6 percentage points and earnings by 3.8 percent in
our preferred specification.
In presenting the results related to neighborhood referrals and labor market outcomes, we
also provide direct evidence on the key identifying assumption underlying of our research design.
In particular, we present evidence that the within-block group correlation in observable neighbor
characteristics does not contribute to the increased propensity of individuals on the same block to
work together. In fact, the analysis implies that based on their observable characteristics
(including education, sex, marital status, race, age, presence of children, immigration status),
pairs on the same block are actually slightly less likely to work together than those on nearby
blocks. Thus, in as much as it is testable on the observables, our research design is robust to
within-block group sorting.
In this way, in addition to providing new evidence on the importance of neighborhood
referrals for labor market outcomes, our analysis also demonstrates the potential strengths of the
general research design that we introduce in this paper. In a manner that deals directly with the
correlation of individual and neighbor characteristics (e.g., due to sorting), this design allows for
the identification of neighborhood effects operating (i) through a specific mechanism, (ii) for a
broad population and a wide variety of subsets of that population, and (iii) for individuals that
have resided in a neighborhood for a variety of tenure lengths. The applicability of this design
extends to the study of neighborhood effects in other contexts (e.g., other metro areas, specific
types of neighborhoods), on specific populations (e.g., youths), and for alternative outcomes (e.g.,
education, teenage fertility, health, welfare participation), provided the researcher can
demonstrate that the within-block group correlation in observable neighbor characteristics does
not contribute significantly to outcomes, thereby ensuring that the key identifying assumption on
unobserved characteristics is at least plausible.
The remainder of the paper is organized as follows. Section 2 sets the paper in the context
of the existing literature. Section 3 describes the data set that we have assembled for the Boston
metropolitan area. Sections 4 and 5 describe our research design and present evidence concerning
the orthogonality of the block-level variation in individual and neighbor characteristics. In these
sections, we also discuss several extensions of our methodology designed to deal with additional
issues related to identification. We report our empirical findings in Section 6 and conclude in
Section 7.