estimation results become statistically weaker than in the larger samples, due in part to the
smaller sample size. Qualitatively, however, our previous results are confirmed, especially with
respect to the gender and marital status, immigration, age and education. Overall, these findings
strongly support the job referral hypothesis and make the reverse causation argument unlikely.
Finally, we take a more detailed look at the effect of match quality on labor market
outcomes in Table 9. The objective here is to focus on individuals who were more likely to be
searching for a job and thus more likely to receive, rather than provide, referrals. In panel 1, we
report estimates using the sub-sample of individuals that have been in residence at least two
years, adding a dummy variable for whether the individual was not employed for the full previous
year. We report the coefficient estimates both for our measure of match quality and for the
interaction term of match quality with the ‘not-employed-for-full-previous-year’ dummy. In this
case, the measure of match quality is based on the parameter estimates for the specification
reported in the second set of columns in Table 8. The results are quite striking: match quality per
se does not have a significant impact on any outcome for the individuals who were employed for
the full previous year (presumably because they were unlikely to have been unemployed last year
and did not need a referral), whereas it has strongly positive and significant effects for the
individuals who were not employed for the full year, and thus more likely to benefit from
referrals.
The second panel in Table 9 reports results of an analogous specification where the
sample is limited to those in residence at least two years and not employed for the full previous
year and match quality is based on the estimated coefficients of the specification reported in the
third set of columns in Table 8. Despite the sharp reduction in the sample size, the results for
labor force participation and employment correspond well with those reported for individuals not
employed for the full previous year in the specification reported in the first panel. A one standard
deviation increase in match quality is associated with a statistically significant increase in labor
force participation of 2.4 percent and employment of 1.9 percent for those individuals not
employed for the full previous year. In this way, the labor market outcome effects appear to be
important for precisely the group that one would think was mostly likely to have received the
referrals.
7 CONCLUSION
This paper aims at detecting and measuring the importance of neighborhood referrals on
labor market outcomes by using a novel data set and identification strategy. Using Census data
that detail the exact block of residence and workplace for a large sample of prime-age workers in
30