table. This increased propensity to work together for individuals on the same block versus block
group is especially strong for pairs of individuals in which (i) both have children and especially
similar aged young children; (ii) both are married; (iii) both are young; (iv) both are high school
graduates; and (v) both are recent immigrants. The propensity for recent immigrants to work
together is not surprising given the importance of social networks for recent immigrants. Because
immigrants in most visa classes are required by US law to be sponsored by a specific employer, it
is very likely that many recent immigrants receive referrals for both residential location and
employment from a social contact already in the US. Thus, importantly, we view the inclusion of
immigration status in the subsequent analysis as a control for this possibility rather than as an
attempt to identify the causal impact of neighborhood referrals for immigrants. It is also
important to note that all of the results of the analysis presented below are robust to dropping
immigrants from the sample.
Table 1 also makes clear that the propensity that two individuals residing in the same
block work together is not a simple monotonic function of the baseline propensity for individuals
residing in the same block group but not the same block. While pairs of all age combinations
residing in the same block group but not the same block are about equally likely to work together,
pairs of young workers residing on the same block are especially likely to work together.
Similarly, while pairs of workers with children in nearby blocks are about as equally likely to
work together as pairs without children, the corresponding propensity of pairs with children to
work together is more than twice that of those without at the block level.
Baseline Specifications. While Table 1 provides suggestive evidence as to the presence and
nature of a social interaction operating at the very local (block) level, two features of our
regression specifications help clarify this evidence. First, the regressions include block group
fixed effects. This ensures that the estimation of our social interaction effects is based exclusively
on comparisons of block- versus block-group-level propensities to work together within the same
block group. Second, by simultaneously including controls for education, race, age, children,
marital status, and gender in the regression, these regressions isolate the marginal contribution of
each characteristic. Given the strong correlation between marital status and the presence of
children, for example, it is difficult to ascertain which of these is important from the analysis of
Table 1 alone.
Table 4 reports the results of three specifications for both equations (1) and (2). The first
row of each column reports the parameter estimate of the average social interaction effect, α0, for
specification (1), which includes block group fixed effects but no covariates Xij. Column 1 reports
22