Measuring the gender wage gap—a methodological note

ABSTRACT

We propose to estimate the Blinder-Oaxaca decomposition by a single-equation model augmented with interactions between the group membership and other predictors. The relative importance of predictors on the discriminatory wage gap is examined by the interaction coefficients, which may lead to very different conclusions than the usual percentage calculations using the detailed decomposition method. Comparisons are made between the traditional interpretations and those suggested here using wage data from Finland. The decomposition analysis suggests that the discriminatory male-female wage gap is largely related to work experience, while our preferred model points to the importance of family gap and working industry.

KEYWORDS:

• Linear regression model
• wage differential
• decomposition
• interaction

JEL CLASSIFICATION:

• J31
• J71
• C21

Acknowledgments

The authors thank Jari Vainiomäki and Ari Hyytinen for their useful comments.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ Neumark (1988) argued that the fair wage structure, ${\mathit{\beta }}^{\ast }$, can be deduced from the employers’ behaviour. Unfortunately, ${\mathit{\beta }}^{\ast }$ is hardly empirically discernible as Neumark (1988, 285) admits. Nevertheless, after some reasoning, he ends up with the practical solution that ${\mathit{\beta }}^{\ast }$ (or its proxy) is estimated from the pooled sample. Reimers (1983) suggested a simple average ${\mathit{\beta }}^{\ast }=\left({\hat{\mathit{\beta }}}_{\mathrm{M}}+{\hat{\mathit{\beta }}}_{\mathrm{F}}\right)/2$. Cotton (1988) took the weighted average ${\mathit{\beta }}^{\ast }=f_M{\hat{\mathit{\beta }}}_{\mathrm{M}}+f_F{\hat{\mathit{\beta }}}_{\mathrm{F}}$, with the weights $f_M$ and $f_F$ consisting of fractions of the corresponding group members.

² The data description can be found at: http://www.tilastokeskus.fi/meta/til/pra_en.html.

³ Classification: 9 years for primary education, 12 years for lower secondary education, 14 years for the lowest level of tertiary education, 16 years for a lower-degree level of tertiary education, 18 years for a higher-degree level of tertiary education and 21 years for a doctorate or equivalent level of tertiary education.

⁴ This difference is found by solving 0.030 = .003x – .002x²/100, where the interaction coefficient ‘Married’ is set equal to the value of the quadratic experience function at x.

⁵ In order to improve the readability of Tables A1 and A2, we do not report the standard errors or the stars that flag the statistical significances of the estimates. We, however, note that all the estimates are statistically significant at least at the 5% significance level, except those marked with ^a (Foreign firm dummy in Column (1) of Tables A1–A2; Education in Column (4) of Tables A1–A2; and Health & services education field dummy in Column (4) of Table A1).