24 lowest unconditional and conditional gaps are observed in Estonia 14. These find- ings can mostly be attributed to the gap in public transfers as it is shown below, in Section 5.4, where it is reported that in Austria and Estonia there are the largest and the lowest median gaps in public transfers, respectively. Since the prime interest of this study is to examine the portion of the gap that cannot be explained by the differences in observed characteristics, next, the condi- tional median gap is analyzed over the age groups. In each country the conditional median gap demonstrates several patterns over these age categories. In Bulgaria and France, the unexplained median gap has U-shaped pattern (unexplained part of the total gap is high at the lower and upper tails of the distribution). In particular, the unexplained median gaps for age groups 25-44 and 45-65 are lower compared to the other two groups. However, the conditional median gap has predominantly inverse U-shaped form (unexplained part of the total gap is lower at the low and upper tails of the distribution) in some countries. Namely, in Belgium, Denmark, Esto- nia, Finland, Greece, Hungary, Latvia, Lithuania, the Netherlands, Portugal, Serbia, and Spain, where the unexplained gaps peak at age groups 25-44 or 45-65. From these countries, in Denmark, Estonia, Finland, Hungary, Latvia, Lithuania and Ser- bia the gap is the largest for individuals between 25-44, while in Belgium, Greece, the Netherlands, Portugal, and Spain the gap is largest in age group 45-65. In other countries, the gap demonstrates either increasing or decreasing, or much more com- plex form. For example, in Austria, Germany, Norway, Sweden and the UK the unexplained gap peaks for the oldest age group. On the other hand, in Croatia, Czechia, Poland, and Slovakia the gap decreases over age groups. In Romania and Slovenia, the unexplained median gap has relatively complex form. In addition to the the median income gap, there is a significant variation within each of these age groups. This within-group variation is easily observed in Figure B.1. Similar to median income gap, the unexplained gap behaviour within groups is also analyzed from the youngest to the oldest age groups across the countries. In Austria, Bulgaria, Croatia, Czechia, Lithuania, and Spain the unexplained gap in favour of men is systematically lower for people with high income (upper end of the distribution) in the youngest age category. On the contrary, in Estonia, Fin- land, Germany, Greece, Hungary, the Netherlands, Norway, and Slovenia, the unex- plained gap in favour of men is the largest for high-income individuals15. In France, Latvia, Poland and Slovakia, the unexplained gap is the largest in the middle of the distribution (i.e., inverse U-shaped pattern). The gap is the lowest around median in Portugal and the UK (i.e., U-shaped pattern). Belgium, Romania, Serbia and Sweden have rather complex pattern of the unexplained gap across the distribution of total income among the individuals below age 25. 14Please note that in Estonia both unconditional and conditional gaps (0.024 and 0.007, respectively) are the lowest among all countries for the given age group, and they are both statistically insignificant. Also, Estonia is the only country in the age group of >65, where results are statistically insignificant at all confidence levels (Table A.12). 15Please note that in these countries, excluding Germany and Hungary, the unexplained part of the total income gap favours women rather than men at the lower end of the distribution. However, this effect gradually decreases or totally fades away at the higher end of the distribution. The decrease in the unexplained gap in favour of women could be interpreted as an increase in the unexplained gap, which favours men. 5. Results 25 In the age group of 25-44, the size of the conditional gap is predominantly in- creasing. Specifically, in Belgium, Denmark, Estonia, Finland, France, Hungary, Lithuania, Norway, and Sweden the unexplained gap is the largest among people with high income. In Austria and Germany, the unexplained gap is lower for the low-income individuals but is relatively flat from median upwards. Unlike from these countries, in Croatia, Greece, Romania, Slovakia, and Spain the gap is larger for low-income individuals (in Spain the gap first decreases and then is relatively flat for upper end of the distribution). Also, in Poland, the gap is largest for the lower end of the distribution, becoming flat starting from median. The gap is observed to be relatively low around median in Bulgaria, Czechia, Portugal and the UK. Some- what similar U-shaped pattern is reported for the Netherlands and Portugal, where the lowest unexplained gap is at the lower end of the distribution. In this age group Slovenia and Serbia are the only countries with inverse U-shaped form of the unex- plained gap. Next, results for age group 45-65 are summarized. In Austria, Croatia, Greece, the Netherlands, Poland, Romania and Spain the largest unexplained gap is present at the lower end of the income distribution. On the other hand, in Denmark, Estonia, Finland, Hungary, Latvia, Lithuania, Slovakia and Sweden the largest unexplained gap is observed at the upper end of the distribution. In Bulgaria, France, Norway, and the UK the lowest gap is reported around median income. In Germany and Serbia, the unexplained gap demonstrates inverse U-shaped pattern, however, the gap is observed to be the lowest not around median but at the lower percentiles of the distribution. In Belgium, Portugal, and Slovenia the gap shows rather complex pattern. In the oldest age group, the unexplained gap has rather complex shape over the whole distribution in Croatia, Czechia, Denmark, Greece, Hungary, Romania and Slovakia. In Romania and Hungary the lowest unexplained gap is reported at 20 th and 30 th percentiles, respectively. Likewise, in Croatia and Denmark, the lowest gap is observed at 40 th percentile. On the contrary to these countries, in Slovakia the lowest gap is at 80th percentile, while in Czechia and Greece at 90th percentile. In Austria, Belgium, and Norway the unexplained gap is systematically lower for high- income individuals. Somewhat similar pattern is observed in France, Germany, and the UK, where the gap decreases throughout the distribution and increases for the individuals with the highest income in the given age group. In Spain, the lowest gap is reported around median, however, for upper income percentiles the gap increases and becomes relatively flat. In Bulgaria, Estonia, Finland, Latvia, Lithuania, Poland, Slovenia, and Sweden the gap is systematically larger at the upper end of the distri- bution. In Serbia, for the individuals with the lowest income the gap is the largest. However, for the 20 th percentile it is in favour of women and starts to benefit men more from 30th percentile onwards. 5.2 Results for the gap in employment income The following part of the paper examines the gap in greater details. First, as de- scribed in Section 4, the gap in the employment income is analyzed. The set of explanatory variables includes the same controls as for the total income. The gap is analyzed for the age groups of 25-44 and 45-65. Detailed results are reported in 26 Appendix A, Table A.13 and Table A.14 The upper panel of Figure 3 reports unconditional and conditional employment income gaps in all 25 countries. In the age group of 25-44, the unconditional and conditional median employment gaps are always in favour of men. The largest un- explained gaps are observed in Latvia and Estonia, where they amount to 0.463 and 0.456 log points, respectively. The fact that Estonia has one of the largest employ- ment income gap in Europe has also been highlighted by earlier studies (see, inter alia, Christofides et al. (2013), Rõõm and Anspal (2010)). In this age group the un- explained employment gap is the lowest in Romania (0.087 log points). The recent study by Boll and Lagemann (2018) also provides evidence that in Romania there is lowest unconditional and conditional employment income gaps. In 15 out of 25 countries, the explained part of the total gap is in favour of women (there countries are: Bulgaria, Croatia, Czechia, Denmark, Estonia, Greece, Hungary, Latvia, Lithua- nia, Poland, Portugal, Romania, Serbia, Slovakia, and Slovenia.). In the rest of the countries, both explained and unexplained gaps are in favour of men. FIGURE 3: Median gap in employment income for all countries. −.2 0 .2 .4 .6 .8 Gap size Austria Belgium Bulgaria Croatia Czechia Denmark Estonia Finland France Germany Greece Hungary Latvia Lithuania Netherlands Norway Poland Portugal Romania Serbia Slovakia Slovenia Spain Sweden UK Age group 25−44 −.2 0 .2 .4 .6 .8 Gap size Austria Belgium Bulgaria Croatia Czechia Denmark Estonia Finland France Germany Greece Hungary Latvia Lithuania Netherlands Norway Poland Portugal Romania Serbia Slovakia Slovenia Spain Sweden UK Age group 45−65 Raw gap Unexplained gap Note: confidence intervals are shown for 90% confidence level. Source: author’s calculation from EU-SILC 2016. In the age group of 45-65 the results are somewhat different from the previous age group (lower panel of Figure 3). The largest unconditional median gap is re- ported for the Netherlands (0.66 log points), where the unexplained part is also the largest among all other countries (0.425 log points). The unconditional median em- ployment income gap in the Netherlands is almost equal to the median total income gap for the same age group, shown in the previous subsection. In the given age group the only country, where the unconditional median gap is in favour of women, is Romania (however, statistically insignificant). The lowest unconditional median
- Results 27 gap, which is in favour of men, is reported for Slovenia (0.061 log points). This is also in line with the findings of Boll and Lagemann (2018). In Latvia and Estonia, the unconditional and conditional median gaps are lower compared to the previ- ous age group, where these indicators were the largest. Moreover, the number of countries, where the explained part of the gap was in favour of women, has also de- creased from 15 to 8. These countries are: Bulgaria, Croatia, Estonia, Latvia, Lithua- nia, Poland, Serbia, and Slovenia. In order to check how an unobserved segregation among industries and firm sizes could have influenced the analysis, the robustness check has been done for 5 countries (Czechia, Estonia, the Netherlands, Norway, and the UK) for the median employment income gap. The detailed results are reported in Table 4. It is observed that compared to the employment income specification, when industry and firm size controls are included, the current study underestimates the explained part of the employment income gap, while the unexplained part is overestimated in most of the cases. This leads to the overestimation of the total employment income gap in all cases apart from Estonia (age group 45-65). Hence, it must be noted that exclu- sion of these controls is a clear limitation of the study. The analysis revealed the significant heterogeneity across the countries not only for median employment income gap but for the whole distribution. The gap reveals interesting behaviour across the distribution, which is shown in Table A.13 and Ta- ble A.14. These results allow to examine “glass ceiling” and “sticky floors” effects. TABLE 4: Robustness check for industry and firm size covariates. Country Explained Unexplained Total Study Robust Diff. Study Robust Diff. Study Robust Diff. Age group 25-44 Czechia -0.028 0.012 -0.040 0.340 0.294 0.046 0.312 0.306 0.006 Estonia -0.030 0.004 -0.034 0.456 0.377 0.079 0.426 0.381 0.045 Netherlands 0.125 0.114 0.011 0.224 0.177 0.047 0.349 0.291 0.058 Norway 0.030 0.069 -0.039 0.263 0.183 0.080 0.293 0.252 0.041 UK 0.111 0.137 -0.026 0.182 0.155 0.027 0.293 0.292 0.001 Age group 45-65 Czechia 0.017 0.020 -0.003 0.260 0.253 0.007 0.277 0.273 0.004 Estonia -0.039 -0.002 -0.037 0.255 0.253 0.002 0.216 0.251 -0.035 Netherlands 0.235 0.236 -0.001 0.425 0.289 0.136 0.660 0.525 0.135 Norway 0.101 0.129 -0.028 0.195 0.097 0.098 0.296 0.226 0.070 UK 0.159 0.187 -0.028 0.261 0.231 0.030 0.420 0.418 0.002 Note: Statistical significance is not shown for convenience purposes. Source: author’s calculations from EU-SILC 2016 data.