that this set of results further validates our attempt to isolate referral effects from sorting via the
general research design proposed in this paper.
Labor Market Outcome Regressions. We now turn to results of a series of labor market
outcome regressions based on each of the specifications of the work match equation reported in
Tables 4 and 5. As described in Section 5, each regression includes a set of individual and
average neighbor characteristics for each socio-demographic characteristic included in the work
match specification as well as a set of block group fixed effects. The three broad columns of
Table 6 report the effect of a one standard deviation increase in match quality on labor market
outcomes for specifications corresponding to the three columns of Table 4. In this table, we only
report the coefficient estimates associated with match quality for the sake of expositional
clarity.4142 Note also that the number of observations varies across specification due to the
number of observations with imputed dependent variables in each case; we drop such
observations from the analysis.
For the specifications based on the full sample, match quality has a positive and
(statistically and economically) significant impact on all dependent variables under consideration.
Our preferred specification, which drops blocks with fewer than five sampled workers, is reported
in the second broad column. For this specification, a one standard deviation increase in match
quality raises labor force participation by about 1.6 percentage points, average days worked per
year by about 4 days, earnings by 3.8 percentage points and wages by 2.1 percentage points. In
this way, our estimated referral effects are indeed associated with improved labor market
outcomes especially as it concerns participation in the labor market and the intensity of that
participation.43 Similar results obtain when housing controls are included in the analysis.4445
in the analysis. The number of immigrants is lowest, for example, in the fourth specification that selects
the block groups that are most homogeneous with respect to this characteristic.
41 The estimation results for the full sets of individual and block-level covariates are quite standard and are
available from the authors upon request.
42 The first two dependent variables refer to labor market outcomes for the week preceding the census
survey. The last four variables represent labor market outcomes for the preceding year. Earnings and wage
regressions are run for the sample of individuals that were fully-employed in the previous year, defined as
having worked at least 40 weeks and at least 30 hours per week.
43 Recall from our discussion above that this analysis will tend to understate the benefits of improved match
quality at the block level as the quality of local matches will typically be overstated for individuals who
generally provide referrals.
44 Standard errors are corrected for clustering at the block level in all labor market outcome regressions
reported in the paper.
45 It is also worth noting that the estimated coefficients on match quality are qualitatively similar when no
additional controls are included for average neighbor characteristics at the block level. This provides some
confidence that the estimated impact of match quality is robust to the possibility of correlation between the
measurement error in these variables and the measurement error in match quality.
27