20 emerge for Greece and Denmark. In Greece, the share of this income group for in- dividuals below 25 is the largest over all countries. On the other hand, this income component has negative share in total income for Danes older than 65. In the next section, first, I analyze aggregate income gap in all countries over the given age groups. Next, the gap in each above-mentioned income component is analyzed in the similar manner. It allows to determine which income component could have the largest contribution to the total gap. Employment income gap is analyzed only for those between age 25-44 and between age 45-65. This is due to the fact that there are very few employed 9 individuals in the other two age groups. On average only 2.5% and 1% of the total sample receive employment income in age groups of below 25 and over 65, respectively. The approach to leave out individuals below age 25 and over age 65 from the employment income gap analysis is in line with Christofides et al. (2013), who analyzed the gender wage gap, using EU-SILC data for year 2007. 5 Results In this section, I present results derived from EU-SILC data by applying uncondi- tional quantile regression and Oaxaca-Blinder decomposition, discussed in Section 3. Before analyzing the gap over the different age groups, first, decomposition re- sults for pooled age groups are presented10. Tables A.3, A.4, and A.5 show detailed decomposition results of total income gap for the 20th, median, and 90th percentiles, respectively. For the low-income individuals, unexplained income gap is mostly due to mar- riage status: being married or cohabiting with partner widens unexplained income gap. The gender income gap at 20th percentile can be explained by managerial occu- pations, as well as training in craft and trade and plant and machinery occupations, and economic status of being inactive. Likewise, as for the gaps at median income, being married has a major contribution to the unexplained portion, while inactive economic status can account for the largest portion of the explained gap. In addi- tion, part-time working can explain significant part of the explained gap for some countries. Similar to the previous two income percentiles, for the high-income indi- viduals, marriage accounts for the largest part of the unexplained gap, while being inactive and having training in managerial, clerical and administrative, and service and sales occupations have positive contribution to the explained income gap. The detailed decomposition of gender employment income gap (the largest part of the total income for individuals below age 65) are reported in Tables A.6, A.7, and A.8 for 20 th, 50 th, and 90 th percentiles, respectively. For the low-income indi- viduals in most of the countries, part-time employment and secondary education contribute positively to the explained gap. Furthermore, training in crafts and re- lated trade could account for significant portion of the explained gap. Similar to the total income gap, the unexplained gap of the employment income is mainly due to marriage status (however, it is not statistically significant for some countries). Likewise, in case of median employment income gap, part-time employment has 9Please note that the term "employed" includes full- and part-time (self-) employed individuals. 10Please note that throughout the section terms "unconditional" and "raw"’ (or "total") are used as synonyms. Also, terms "unexplained" and "conditional" gaps are used interchangeably. 5. Results 21 significant contribution to the explained gap, while marriage accounts for large part of the unexplained gap. And finally, among high-income individuals, economic sta- tus of being a part-time worker or inactive have major contribution to the explained gap for most of the countries. Moreover, training in clerical and service occupations has significant positive effect on the explained gap. As in all other cases, marriage constitutes large part of the employment income gap at the upper end of the dis- tribution. Across the distribution tertiary education tends to lower the explained gap: more women enroll in universities and choose degrees that are primarily male- dominated, also female graduates outnumber their male counterparts. This can be considered as a consequence of the women’s strategy to attain more education to reduce the disparity. However, despite these changes, the concentration at lowest- earning occupations (e.g. teachers) still remains the problem for women. The reason why secondary education does not impact the gap for median and high-income in- dividuals could be explained in a following manner: people at median and upper end of the income distribution are expected to have higher education than those at the lower end, therefore, for them the differences in lower levels of education are not likely to have significant impact on the gap. The findings of marriage having large impact on unexplained gap of either total income or employment income are line with earlier literature, known as “marriage premium” for men (e.g. see Dougherty (2006) for selection effects). Given the fact that women are underpaid compared to their male counterparts, for maximizing the household income the married couple would rather focus on male partner’s labor market activities and allocate female partner’s time to household chores. 5.1 Results for the gap in total income The results are reported for different age groups that allow observing the behaviour of the income gap over different age categories. Tables A.9, A.10, A.11, and A.12 show results of the unconditional quantile regression decomposition obtained for 9 quantiles (10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, and 90%) for the individu- als below age 25, between 25-44 and 45-65, and over age 65, respectively. The upper parts of these tables show how much of the total income gap can be explained, while lower parts report the portion that cannot be explained by differences in observed characteristics. There is a variation in the statistical significance of the gaps, either raw, explained, or unexplained for all age groups. Figure 2 reports the median in- come gap for all age groups across the countries. Detailed decomposition of the total income gap is provided in appendix, Figure B.1. The results shown in Figure 2 are interpreted by age groups. First, I analyze the youngest age group, which is followed by individuals between 25-44 and 45-65, and, lastly, the oldest age group of the sample is analyzed. According to the Figure 2, for individuals below age 25, the unconditional me- dian income gap systematically favours men in all countries, except in Finland, Greece, Portugal, and Sweden, where the gap is in favour of women. On the con- trary, the conditional median income gap favours women only in Estonia, Greece, the Netherlands, Norway, and Sweden. As observed, in Greece and Sweden both unconditional and conditional gaps favour women (i.e., the gaps are negative), how- ever, unconditional gap is lower than the conditional one. This fact implies that the explained part of the total income gap is negative as well, i.e. had men and
22 women had the same returns, the differences in observed characteristics would ben- efit women more than men. Likewise, in Finland, Portugal, Romania, Serbia, the conditional income gap exceeds median raw gap, again implying that the explained part of the total gap is in favour of women. The largest median raw income gap in favour of men is reported to be in Bulgaria (0.695 log points), while the largest median gap favouring women is observed in Greece (-0.807 log points). In Bulgaria the unexplained part of the total gap is also the largest among all studied countries and accounts for 0.726 log points of the total median gap. According to the report on world population prospects by United Nations 11, Bulgaria is the fastest shrink- ing country in Europe, which is mostly due to brain drain among young population. Also, the share of female emigrants with tertiary education in total emigration is ap- proximately 4 percentage points higher than men’s12. This could be the reason why Bulgaria has the largest median total income gap among the youngest. However, sta- tistical significance of either raw or unexplained gaps varies across countries, which can be attributed to the small sample size in some countries (Table A.1). FIGURE 2: Median total income gap for all countries. 0 .2 .4 .6 .8 Gap size <25 25−44 45−65 >65 Age group Austria −.4−.2 0 .2 .4 .6 Gap size <25 25−44 45−65 >65 Age group Belgium .2 .4 .6 .8 1 1.2 Gap size <25 25−44 45−65 >65 Age group Bulgaria 0 .2 .4 .6 .8 Gap size <25 25−44 45−65 >65 Age group Croatia .2 .4 .6 .8 Gap size <25 25−44 45−65 >65 Age group Czechia −.2−.1 0 .1 .2 .3 Gap size <25 25−44 45−65 >65 Age group Denmark −.4 −.2 0 .2 .4 .6 Gap size <25 25−44 45−65 >65 Age group Estonia −.2 −.1 0 .1 .2 .3 Gap size <25 25−44 45−65 >65 Age group Finland 0 .1 .2 .3 .4 Gap size <25 25−44 45−65 >65 Age group France −.2 0 .2 .4 .6 Gap size <25 25−44 45−65 >65 Age group Germany −1.5−1 −.5 0 .5 1 Gap size <25 25−44 45−65 >65 Age group Greece −.2 0 .2 .4 .6 Gap size <25 25−44 45−65 >65 Age group Hungary −.2 0 .2 .4 .6 Gap size <25 25−44 45−65 >65 Age group Latvia −.4 −.2 0 .2 .4 .6 Gap size <25 25−44 45−65 >65 Age group Lithuania −.2 0 .2 .4 .6 .8 Gap size <25 25−44 45−65 >65 Age group Netherlands −.2 0 .2 .4 Gap size <25 25−44 45−65 >65 Age group Norway 0 .2 .4 .6 .8 1 Gap size <25 25−44 45−65 >65 Age group Poland −.4 −.2 0 .2 .4 .6 Gap size <25 25−44 45−65 >65 Age group Portugal 0 .5 1 1.5 Gap size <25 25−44 45−65 >65 Age group Romania −.2 0 .2 .4 .6 .8 Gap size <25 25−44 45−65 >65 Age group Serbia 0 .2 .4 .6 .8 Gap size <25 25−44 45−65 >65 Age group Slovakia −.2 0 .2 .4 .6 .8 Gap size <25 25−44 45−65 >65 Age group Slovenia −.5 0 .5 1 Gap size <25 25−44 45−65 >65 Age group Spain −.4 −.2 0 .2 .4 Gap size <25 25−44 45−65 >65 Age group Sweden −.2 0 .2 .4 .6 Gap size <25 25−44 45−65 >65 Age group UK Raw gap Unexplained gap Note: confidence intervals are shown for 90% confidence level. Source: author’s calculation from EU-SILC 2016. Next, results for individuals between 25-44 are summarized. Unlike the previ- ously analyzed age group, the total median gap favours men in all countries. In 112017 Revision of World Population Prospects is available at United Nations’website . 12Database on Immigrants in OECD and non-OECD Countries: DIOC . OECD, 2011. 5. Results 23 Bulgaria, Latvia, and Lithuania unexplained part of the total gap is larger than the total gap itself. This implies that in these countries differences in observed character- istics benefit women rather than men. However, in other countries both explained and unexplained parts of the total gap favour men. The largest total median gap is reported for Greece and only 0.176 out of 0.733 log points can be explained by the control variables. In the age group of 25-44, Greece is also observed to have the largest unexplained median gap, which accounts for 0.557 log points. On the other hand, Denmark is shown to have the lowest unconditional and conditional median gaps (0.124 and 0.115 log points, respectively). The observed characteristics can ac- count for 201 out of 0.594 and 0.339 log points in the UK and Austria, respectively, which is the largest portion among all other countries (Table A.10). In contrast to the previous age group, the results reported for the individuals between 25-44 are statistically significant in all countries. In the following age group of 45-65, similar to age group of 25-44, the uncon- ditional median income gap is always in favour of men. In Estonia, Latvia, and Lithuania the portion of the total income gap that is due to the differences in ob- served characteristics is in favour of women. In other countries both explained and unexplained gaps benefit men more than women. The largest total median gap is again observed in Greece, where only 0.109 out of 0.849 log points can be explained by the control variables, making Greece the country with the largest unexplained part of the raw gap (Table A.11). Christofides et al. (2013) found that in Greece public sector employment is associated with a reduced gender wage gap. Accord- ing to the labor force survey, conducted by International Labour Organization13, the share of public sector employment decreased drastically following the 2008 crisis. Assuming that most of these people moved to private sector, that could have laid a solid foundation to the enlargement of the employment income and, therefore, total income gaps (e.g. see de Castro et al. (2013)). Greece is followed by the Nether- lands, Spain, and Germany, where the unconditional median total income gaps are estimated to be 0.666, 0.646, and 0.644, respectively. The lowest total median income gap is reported in Lithuania (0.112 log points). However, more interesting indicator is unexplained part of the total income gap, which is the lowest in Slovenia (0.039 log points). As it is shown in Section 5.2, Slovenia is the country with the lowest median employment income gap among people between 45-65, which could be lowering the total median income gap to some extent. In addition, both unconditional and conditional median gaps are statistically significant in all countries for the given age group. And finally, for individuals over age 65, the unconditional median total income gap always favours men over women. The differences in observed characteristics are in favour of women in Denmark and Slovakia (-0.071 and -0.020 log points, re- spectively). In all other countries, both unconditional and conditional gaps are ob- served to be in favour of men. The largest unconditional and conditional median gaps are reported for Austria (0.651 and 0.547 log points, respectively), while the 13Informaton on public employment by sectors and sub-sectors of national accounts is available at International Labour Organization’s website.