Panel A: Effect of early nonemployment in the structural equation:
 
 
OLS

2SLS

Standard errors
^{a}

Robust

Cluster g∗p

Robust

Cluster g∗p

Outcomes:
 
(1)

(2)

(3)

(4)

Salaried empl.

coeff

–0.00169***

–0.00169***

–0.00256

–0.00256


se

(0.00034)

(0.00041)

(0.00375)

(0.00290)


Pval
 
0.00019
 
0.38202


Bootstrap Pval ^{b}
 
0
 
0.45646


Exogeneity test Pval ^{c}
   
0.767

Selfempl.

coeff

0.00054*

0.00054

0.00248

0.00248


se

(0.00030)

(0.00041)

(0.00338)

(0.00258)


Pval
 
0.19177
 
0.34175


Bootstrap Pval
 
0.18619
 
0.37437


Exogeneity test Pval
   
0.438

Overall empl.

coeff

–0.00115***

–0.00115***

–0.00008

–0.00008


se

(0.00021)

(0.00025)

(0.00207)

(0.00151)


Pval
 
0.00005
 
0.95655


Bootstrap Pval
 
0
 
0.96697


Exogeneity test Pval
   
0.467

Log earnings

coeff

–0.0269***

–0.0269***

–0.1002**

–0.1002***


se

(0.0033)

(0.0040)

(0.0419)

(0.0291)


Pval
 
2.51E08
 
0.0013


Bootstrap Pval
 
0
 
0.0060


Exogeneity test Pval
   
0.00970

Log hours worked

coeff

–0.0203***

–0.0203***

–0.0723**

–0.0723***


se

(0.0024)

(0.0029)

(0.0307)

(0.0207)


Pval
 
9.35E09
 
0.0011


Bootstrap Pval
 
0
 
0.0060


Exogeneity test Pval
   
0.0112

Panel B: Effect of the instrument in the first stage : OLS
 
Outcome:

Standard errors:

Robust

Cluster g∗p
  
Early nonempl.

coeff

5.4615***

5.4615***
  

se

(1.7273)

(1.6848)
  

Pval
 
0.00230
  

Bootstrap Pval
 
0.00400
  

F stat
 
10.51
  

Bootstrap F stat ^{d}
 
9.25
  
 Standard errors between parentheses. Panel A reports results from estimating β in Eq. (2). β is the effect of one pp increase in \(y^{0}_{it_{1}}\), i.e. the % of hours spent in nonemployment in the first two and a half years after graduation relative to potential total hours if one would work fulltime during the whole period. For clustered standard errors, we report the Pvalue and the wild bootstrap Pvalue. Column 12 (34) show OLS (2SLS). In 2SLS the provincial unemployment rate at graduation is used as instrument for \(y^{0}_{it_{1}}\). Panel B shows the effect of the instrument on \(y^{0}_{it_{1}}\) in the first stage and the corresponding F statistic

^{***}
p <0.01, ^{**}
p <0.05, ^{*}
p <0.1

^{a}Robust accounts for heteroskedasticity. Clusters are defined by graduation year g and province of residence at graduation p (G=44 clusters)

^{b}Computed according to the wild bootstrap proposed by Davidson and MacKinnon (2010) for 999 repetitions

^{c}With clustered standard errors, this test is defined as the difference between two SarganHansen statistics: one for the equation where \(y^{0}_{it_{1}}\) is treated as endogenous and one for the equation where \(y^{0}_{it_{1}}\) is treated as exogenous. Under the null that \(y^{0}_{it_{1}}\) is exogenous, the statistic is distributed as χ
^{2}(1)

^{d}Bootstrap F statistic is the F statistic corresponding to the bootstrap Pvalue of the t statistic of the instrument: we rely on the equivalence between F and t distribution: for G=44, t
^{2}(G−1)=F(1,G−1)