Skip to main content

Table 6 Estimated effects using other matching and weighting approaches

From: How sensitive are matching estimates of active labor market policy effects to typically unobserved confounders?

 

Regular employment after 36 months

Real monthly earnings after 36 months

 

Std.

Ext.

Diff.

Std.

Ext.

Diff.

Kernel matching (baseline)

0.063\(^{***}\)

0.056

0.006

118\(^{***}\)

109\(^{***}\)

9

(0.017)

(0.018)

(0.006)

(29)

(29)

(9)

Caliper radius matching

0.054\(^{***}\)

0.058\(^{***}\)

− 0.005

107\(^{***}\)

113\(^{***}\)

− 3

(0.019)

(0.019)

(0.014)

(32)

(32)

(18)

Local linear matching

0.066\(^{***}\)

0.060\(^{***}\)

0.006

124\(^{***}\)

116\(^{***}\)

8

(0.018)

(0.018)

(0.006)

(29)

(30)

(9)

Inverse probability weighting

0.063\(^{***}\)

0.057\(^{***}\)

0.006

117\(^{***}\)

111\(^{***}\)

7

(0.017)

(0.017)

(0.004)

(28)

(29)

(7)

Doubly robust

0.058\(^{***}\)

0.054\(^{***}\)

0.005

110\(^{***}\)

104\(^{***}\)

6

(0.017)

(0.017)

(0.004)

(28)

(28)

(7)

Control variables

 Typically observed covariates

\(\checkmark\)

\(\checkmark\)

 

\(\checkmark\)

\(\checkmark\)

 

 Typically unobserved covariates

 

\(\checkmark\)

  

\(\checkmark\)

 
  1. This table shows estimated ATTs. Standard errors are obtained via bootstrapping with 999 replications. Statistical significance based on the normal approximation on the 10/5/1% level is indicated by \(^{*}/^{**}/^{***}\). Local linear matching use a standard bandwidth of 0.06. The caliper for the radius matching is set equal to three times the 90% quantile of the distribution of differences in the propensity score from a nearest neighbor matching with replacement (Huber et al. 2015). The doubly robust estimator combines inverse probability weighting with parametric outcome adjustment (Robins et al. 1995)