A Blinder–Oaxaca decomposition for double-hurdle models with an application to migrants’ remittance behaviour

Abstract

This article extends the Blinder–Oaxaca decomposition to double-hurdle (DH) models. The proposed methodology focuses on a generalized non-Gaussian DH specification with correlated errors and allows decomposing the gaps in participation probabilities and outcome levels for a wide range of nested specifications. The decomposition technique is applied to analyse differences in remittance behaviour of temporary and permanent migrants.

Keywords:

• decomposition analysis
• double-hurdle models
• IHS transformation
• remittances

JEL Classification:

• C21
• C24
• F22
• F24

Notes

¹ Equation 2 is a special case of the generalized decomposition proposed by Oaxaca and Ransom (1994).

² I do not explicitly model heteroscedasticity to avoid introducing further confounding elements in the decomposition analysis.

³ Other authors consider instead the Box-Cox transformation (see Aristei and Pieroni, 2008).

⁴ See Bauer and Sinning (2008) and Mussa (2010) for alternative decompositions. The choice of the SE in the counterfactual affects decomposition results only if large differences in σ_g for the two groups exist.

⁵ See Amuedo-Dorantes and Pozo (2006) for a detailed discussion on EMIF data.

⁶ Test for overidentifying restrictions supports the validity of the instruments considered (which include the number of different jobs held, the level of English proficiency and a dummy indicating whether the worker has received any job training). F-test for the significance of instruments in the first-stage regression rejects the null of weak instruments.

⁷ The estimated income coefficient implies an elasticity of remittances equal to 0.9863, which is in line with the findings of Bettin et al. (2012) and is considerably higher than estimated elasticities of studies not addressing the potential endogeneity of income.

⁸ The underlying IHS DH estimates are not presented here, but available upon request.