Women’s Tertiary Education Masks the Gender Wage Gap in Turkey

Abstract

This paper investigates the gender wage gap for full-time formal sector employees, disaggregated by education level. The gap between the labor force participation rate of women with tertiary education and those with lower levels of education is substantial. There is no such gap for men. Hence, existing gender wage gap studies for Turkey, where we observe lopsided labor force participation rates by education levels, compare two very different populations. We disaggregate the whole sample by education level to create more homogenous sub-groups. For Turkey, without disaggregation, the gender wage gap was 13% in 2011, and women are significantly over-qualified relative to men on observed characteristics. Once we disaggregate the sample by education level, we show that the gender wage gap is 24% for less educated women and 9% for women with tertiary education in full-time formal employment. Observed characteristics only explain 1 % of this gap in absolute terms. We further disaggregate the data by public and private employment. The gender gap is higher in the private sector. However, women with tertiary education in the public sector are significantly better qualified compared to men, and consequently the adjusted gender wage gap is higher for women with tertiary education in the public sector. Our estimates also indicate a rise in the gender wage gap between 2004 and 2011.

Appendices

Appendix 1: Female Population by Education and Employment

Table 7 Female population by education level (15+)

Appendix 2: Methodology

Derivation of Wage Estimation Model (Eq. 1)

We calculate the average wage gap as follows:

$$ Difference\; in\; hourly\ Log\ wages= R= E\left({Y}_M\right)- E\left({Y}_W\right)= E\left({X}_M\right)^{\prime }{\beta}_M- E\left({X}_W\right)^{\prime }{\beta}_W $$

(2)

where E(Y_M) and E(Y_w) are expected mean log hourly wages, E(X_M) and E(X_w) are mean characteristics of male and female samples, respectively, and β_M and β_W are coefficients that determine the expected hourly wages. Threefold decomposition (Jann 2008) can determine the amount of mean outcome difference due to difference in observable characteristics, and the amount due to difference in coefficient estimates:

$$ \begin{array}{l} R={\left[ E\left({X}_M\right)- E\left({X}_W\right)\right]}^{\hbox{'}}{\beta}_W+ E{\left({X}_W\right)}^{\hbox{'}}\left({\beta}_M-{\beta}_W\right)+{\left[ E\left({X}_M\right)- E\left({X}_W\right)\right]}^{\hbox{'}}\left({\beta}_M-{\beta}_W\right)\ \hfill \\ {}\mathrm{R}=\mathrm{Endowment}+\mathrm{Coefficients}+\mathrm{Interaction}\hfill \end{array} $$

(3)

Endowment component refers to the difference between average observable characteristics related to the job market for men and women, such as age, tenure etc. Coefficients (for example, the amount that an extra year of tenure increase one’s wages) component measure the coefficients’ contribution to hourly wage differences. This component is the measure unexplained part. Finally, the interaction component accounts for the coexistence of simultaneous differences for endowments and coefficients between the two groups. The estimated decomposition takes on the following format:

$$ \widehat{R}={\overset{-}{Y}}_M-{\overset{-}{Y}}_W=\left({\overset{-}{X}}_M-{\overset{-}{X}}_W\right)^{\prime }{\widehat{\beta}}_W+{\overset{-}{X}}_W^{\prime}\left({\widehat{\beta}}_M-{\widehat{\beta}}_W\right)+\left({\overset{-}{X}}_M-{\overset{-}{X}}_W\right)^{\prime}\left({\widehat{\beta}}_M-{\widehat{\beta}}_W\right) $$

(4)

If we re-arrange endowment and interaction terms, Eq. 3 can be reduced to Eq. 1 in the main text, which is the common presentation in the literature, and corresponds to Eq. 9 in Oaxaca and Ransom (1994), when the male wage structure is assumed to be the base wage structure. We present our empirical results in parallel to Eq. 1 because we note that in our empirical analysis, interaction effects are very close to zero in most of the cases and its combination with Endowment effect leads to no loss of significant information.

Appendix 3: Data

Labor Force Survey is the source of data for official statistics about employment. The survey aims to cover all households in Turkey (institutional population is excluded). Data is collected throughout the year (first week of every month) and the sample size is large enough to produce quarterly employment statistics both at national level as well as urban and rural areas of NUTS2 regions (there are 26 NUTS2 regions in Turkey). As a result, there are 52 strata (26 NUTS2 regions and rural and urban areas within each region). First, each Stratum is divided into clusters and the total number of clusters is 350. Than each cluster is further divided into blocks (containing approximately 100 houses in urban areas or whole villages in rural areas). Finally, each quarter 15 of households in each block is surveyed (TurkStat 2012a). The population of the NUTS2 regions differ widely hence weighting of estimates is necessary for representative results. The sampling weights are provided by TurkStat in the dataset. The response rate to the survey is 87% (TurkStat 2012b: XXII). We use the whole sample for the year in our study.

Appendix Table 8 (for 2011) and Table 9 (for 2004) present descriptive statistics for dependent and independent variables used in the regression analysis disaggregated by gender and education level. As can be seen from both tables single women are over-represented among working women compare to men. Moreover, women with zero children is also under-represented in the sample compare to similar men. These findings are probably due to the fact that women tend to leave labor force once they have children. And women with children who stays in the labor force are more likely to have more secure employment. As expected, women with tertiary education are over-represented in overall sample. Women in the formal, full-time employment are more likely to be younger, reside in urban areas, metropolitan cities (Istanbul, Ankara and Izmir) and western regions (first six regions). Finally, women are more likely to work in medium sized firms, in public sector and they have shorter tenure in in their current employers.

Table 8 Averages of dependent and ındependent variables in Tables 3, 4 and 5 (for 2011)

Table 9 Averages of dependent and ındependent variables in Table 6 (for 2004)