have been raised concerning pairwise bootstrap in small samples (Horowitz, 2000). While our
sample is quite large, we have a very small number of ones in our dependent variable, which may
create similar problems. We verified the accuracy of the pairwise bootstraps by also estimating
standard errors using a pairwise bootstrap with the HC3 correction and also with a wild bootstrap
(Mammen (1993); Flachaire (1999), (forthcoming)).29
5 EMPIRICAL DESIGN - LABOR MARKET OUTCOMES
Having analyzed the impact of local interactions on job referrals, the second portion of
our analysis examines whether such referrals have an impact on labor market outcomes more
generally. In particular, given the characterization of how the strength of social interactions
related to job referrals (i.e., the propensity to work together) varies with the attributes of a pair of
individuals identified in the first portion of our analysis, we explore whether an individual’s labor
market outcomes are related to the idiosyncratic quality of the strength of the potential networks
available on her block. Specifically, we estimate a series of labor market outcome regressions that
include a measure of match quality defined at the individual level along with controls for
individual and average neighbor characteristics (measured at the block level) as well as block
group fixed effects.
The goals of this portion of our analysis are two-fold. First, since we detect informal
hiring effects indirectly, it serves as a check on the plausibility of the first portion of our analysis.
Second, by focusing on outcomes we hope to be able to provide a better sense of the magnitude
of our estimated network effects. It is certainly possible that referrals may be more likely among
neighbors but may have little effect on labor market outcomes - i.e., that without the referral the
individual would find a comparable job through another search method. In addition, our labor
market models are less likely to understate the effect of referrals when compared to the referral
effects model described in the previous section. In particular, with limited sorting within block
groups, expected match quality for individual with others in the same block group is the same as
their actual block match quality. Consequently, the block level index for match quality is likely
to capture the effect of referrals both within the block and from neighboring blocks.
For this analysis, the unit of observation is an individual rather than a pair. For the
employment and labor force participation outcomes, the econometric model is a linear probability
29 Pairwise bootstraps are estimated using a sample based on the pair of the predicted value and the
predicted residual for each observation. The HC3 correction scales the predicted residual for each
observation by the estimated variance of the predicted residual for that observation while the wild bootstrap
multiplies the predicted residual for each observation by a random number.
17
More intriguing information
1. FUTURE TRADE RESEARCH AREAS THAT MATTER TO DEVELOPING COUNTRY POLICYMAKERS2. Naïve Bayes vs. Decision Trees vs. Neural Networks in the Classification of Training Web Pages
3. Bird’s Eye View to Indonesian Mass Conflict Revisiting the Fact of Self-Organized Criticality
4. Integrating the Structural Auction Approach and Traditional Measures of Market Power
5. The name is absent
6. Delivering job search services in rural labour markets: the role of ICT
7. The name is absent
8. Keystone sector methodology:network analysis comparative study
9. Research Design, as Independent of Methods
10. CAPACITAÇÃO GERENCIAL DE AGRICULTORES FAMILIARES: UMA PROPOSTA METODOLÓGICA DE EXTENSÃO RURAL