Dynamic decomposition of regional wage differentials in Korea

Abstract

Using a Juhn–Murphy–Pierce (JMP) decomposition, this study analyses the dynamic changes in regional wage differentials between the Seoul Metropolitan Areas and other regions in South Korea. Data from the Korean Labor and Income Panel Study for three years (2000, 2004, and 2008) is used. JMP decomposition provides information about the components that explain changes in regional wage differentials over time. Between 2000 and 2004, the variations in observed and unobserved components are associated with counteracting effects on regional wage differentials. While changes in observed components contribute more to widen regional wage differentials, those in unobserved components narrow them. However, between 2004 and 2008, both observed and unobserved components move in the same direction to narrow regional wage differentials. Based on our empirical results, we discuss some policy implications.

JEL classification:

• J31

Keywords:

• Regional wage differentials
• Dynamic decomposition
• Seoul Metropolitan Area
• South Korea

Notes

1 The SMA is a region located in northwest South Korea including three different administrative districts: Seoul Special City, Incheon Metropolitan City, and Gyeonggi Province. The SMA is ranked as the second largest metropolitan area in the world (following the Tokyo Metropolitan Area in Japan) with a population of around 23.5 million as of 2010 (Forstall, Greene, & Pick, 2009).

2 Other important studies on wage differentials in the Korean labor market include the following: Cho, Cho, et al. (2010) and Cho, Lim, et al. (2010) investigate the dynamics of gender wage differentials in the dual (core and peripheral) labor markets; Cho and Cho (2011) study the different dynamic changes of gender wage gaps in the formal and informal sectors; and Kim and Cho (2009) examine entry dynamics of self-employment and their determinants.

3 For female workers, we account for a sample selection problem in wage determination, and we thank two anonymous referees who raised this issue. Following Yun (1999), we estimate the wage equation for each region using two-step approach by Heckman (1976). Thus for the decomposition at time t, we have ${\overline{D}}_t\equiv {\overline{Y}}_t^{SMA}-{\overline{Y}}_t^{NSMA}=({\overline{X}}_t^{SMA}-{\overline{X}}_t^{NSMA}){\hat{\beta }}_t^{SMA}+{\overline{X}}_t^{NSMA}({\hat{\beta }}_t^{SMA}-{\hat{\beta }}_t^{NSMA})+({\hat{\lambda }}_{SMA}{\overline{IMR}}_{SMA}-{\hat{\lambda }}_{NSMA}{\overline{IMR}}_{NSMA})+({\overline{e}}_t^{SMA}-{\overline{e}}_t^{NSMA})$.

4 As explained in footnote 3, for female workers we have to add an inverse Mills ratio (IMR) effect term in Eq. (3(3) ${\overline{D}}_t=({\overline{X}}_t^{SMA}-{\overline{X}}_t^{NSMA}){\hat{\beta }}_t^{SMA}+{\overline{X}}_t^{NSMA}({\hat{\beta }}_t^{SMA}-{\hat{\beta }}_t^{NSMA})$(3) ).

5 ${\sigma }_t^{SMA}$ is used in the process of standardization because ${\hat{\beta }}_t^{SMA}$, the estimated coefficient of the SMA wage equation, is used in Eq. (5(5) $Y_{it}^{NSMA}=X_{it}^{NSMA}{\hat{\beta }}_t^{SMA}+e_{it}^{*NSMA}\text{,}$(5) ).

6 We include the inverse mills ratio to control for sample selection bias for the female subgroup in Eqs. ⁽⁴⁾(4) $Y_{it}^{SMA}=X_{it}^{SMA}{\hat{\beta }}_t^{SMA}+e_{it}^{SMA}$(4) ⁽⁵⁾(5) $Y_{it}^{NSMA}=X_{it}^{NSMA}{\hat{\beta }}_t^{SMA}+e_{it}^{*NSMA}\text{,}$(5) . Thus, for female workers, the mean difference at time t will be ${\overline{D}}_t\equiv {\overline{Y}}_t^{SMA}-{\overline{Y}}_t^{NSMA}=\Delta {\overline{X}}_t{\hat{\beta }}_t^{SMA}+\Delta {\overline{IMR}}_t{\hat{\beta }}_t^{SMA}+{\sigma }_t^{SMA}\Delta {\overline{\theta }}_t$.

7 Since Eq. (4(4) $Y_{it}^{SMA}=X_{it}^{SMA}{\hat{\beta }}_t^{SMA}+e_{it}^{SMA}$(4) ) is estimated by OLS, we have ${\overline{e}}_t^{SMA}={\overline{\theta }}_t^{SMA}=0$. Thus in Eq. (6(6) $\begin{array}{c}{\overline{D}}_t\equiv {\overline{Y}}_t^{SMA}-{\overline{Y}}_t^{NSMA}\\ =({\overline{X}}_t^{SMA}-{\overline{X}}_t^{NSMA}){\hat{\beta }}_t^{SMA}+({\overline{e}}_t^{SMA}-{\overline{e}}_t^{*NSMA})\\ =({\overline{X}}_t^{SMA}-{\overline{X}}_t^{NSMA}){\hat{\beta }}_t^{SMA}+{\sigma }_t^{SMA}({\overline{\theta }}_t^{SMA}-{\overline{\theta }}_t^{*NSMA})\\ =\Delta {\overline{X}}_t{\hat{\beta }}_t^{SMA}+{\sigma }_t^{SMA}\Delta {\overline{\theta }}_t\end{array}$(6) ), it will be the case that $\Delta {\overline{\theta }}_t=-{\overline{\theta }}_t^{*NSMA}$. However, for ease of interpretation, we will continue to use the difference prefix (Δ), when explaining the decomposition process.

8 As explained in footnote 6, for female workers we add IMR change effect in Eq. (8(8) ${\overline{D}}_1-{\overline{D}}_0=(\Delta {\overline{X}}_1-\Delta {\overline{X}}_0){\hat{\beta }}_1^{SMA}+\Delta {\overline{X}}_0({\hat{\beta }}_1^{SMA}-{\hat{\beta }}_0^{SMA})+(\Delta {\overline{\theta }}_1-\Delta {\overline{\theta }}_0){\sigma }_1^{SMA}+\Delta {\overline{\theta }}_0({\sigma }_1^{SMA}-{\sigma }_0^{SMA})$(8) ).

9 The numbers of individual respondents in the surveys are 11,205 in 2000, 11,661 in 2004, and 11,734 in 2008.

10 We thank an anonymous referee who raised this issue.

11 The KLIPS also includes earning information for part-time and unpaid family workers. The shares of part-time workers are 8.96%, 7.24%, and 5.76% for 2000, 2004, and 2008. Moreover, the shares of part-time workers who report monthly wage in the KLIPS sample are just 2.27% in 2000, 2.07% in 2004, and 1.43% in 2008. Because the KLIPS is an annual survey for urban workers, it potentially causes another bias when we covert the earnings of part-time workers into monthly wages. Hence, in our final sample we exclude part-time workers and focus on the full-time workers.

12 We observe different labor force participation (LFP) rates between males and females. While the LFP of male workers ranges from 74% (2000) to 76% (2008), female workers range from 42% (2000) to 43% (2008). Kim (2009) suggests that self-selectivity of female labor in Korea is more prominent and, therefore, the female wage equation should be specified to deal with this problem. We correct selectivity in the case of women for the wage equation, as in Garciá et al. (2001). Consequently, all JMP decompositions for female workers in the next section are based on parameter estimates obtained from the Heckman two-step procedure. As shown in ^{Table A1Table A2}, in the first step probit specification for female workers, binary LFP variable is the dependent variable, and the set of independent variables includes female worker's age, squared age, the number of children aged less than 7, the number of children aged 7–15, household income excluding her wage income, marital status, and residential region.

13 We check the sensitivity of the results to the exclusion of occupational controls. We find the major implication of JMP decomposition remains the same in the case of the inclusion. Since other previous studies analyzing regional wage differentials include the occupational controls, we also adopt similar specifications.

14 Someone may be concerned that the potential self-selectivity into SMA and NSMA regions is involved in the wage equation. If the decision to work in a particular region is endogenously determined with respect to a worker's wage, selection-bias correction should be included as an additional regressor in the wage equation. However, most of previous related literature treats the employment region variable exogenously. Nonetheless, we implemented a statistical test on whether the region variable is endogenous or not, which is not reported here but is available upon request. We find that all IMR (inverse Mills ratio) coefficients representing the existence of selectivity bias are insignificant even at a 10% level. We thus confirm that sample selection into regions is not a serious issue in our wage equation and that the region variable can be treated as an exogenous variable.

15 Following Oaxaca and Ransom (1994), we transform the coefficients of the dummy variables such that they reflect deviations from the grand mean rather than deviations from the reference category. All decomposition results for education, industry, and occupation variables are generated with this transformation.

16 Although these decreasing trends of wage differentials are similar, the exact values of changes are different from what we observe in Tables 2 and 3. This is because in the procedure of JMP decomposition some of the observations are dropped if any of their explanatory variables is missing.

17 For example, for male workers in the first period, changes in observed components widen the regional wage differentials by 1.09%-p while changes in unobserved components narrow them by 1.99%-p.

18 As Pereira and Galego (2011) note, this is a limitation of both the Oaxaca–Blinder and JMP decomposition methods.

19 This is in fact the case with our data. The wage structure regressions of SMA in Table A1 in Appendix show the increases of the standard deviations of residuals in the SMA for both male and female workers.