About time: the narrowing gender wage gap in Austria

We examine the gender wage gap in Austria from 2005 to 2017 using data from EU-SILC. The raw gap of hourly wages declined from 18.6 log points in 2005 to 14.9 log points in 2017. We use standard decomposition techniques that correct for differences in the distributions of human capital and other variables between men and women. Decompositions of the wage gap indicate that both the explained and the unexplained part of the gender wage gap decreased substantially over the last ten years. Using the approach developed by Neumark (J Hum Resour 22:279–295, 1988), the unexplained wage gap shrank from 8.7 log points in 2005 to 5.1 log points in 2017. The main reason for the decline in wage differences was the relative improvement of women’s observed and unobserved characteristics.


Introduction
demonstrated that the US gender wage gap is much smaller than it had once been and she concludes that this decline is the result of increases in the human capital of women relative to men. In contrast, Blau and Kahn (2017), using PSID data for 1980-2010, stress that because US women exceed men in educational attainment by now, traditional human capital factors, although they were essential for the narrowing of the gender wage gap, explain little of the still existing wage gap. In addition, the unexplained component of the US gender wage gap did not fall much between the 1990s and 2010. They conclude that differences in the selection into occupations and industries are the most important aspect of the persistent US gender wage gap. Böheim et al. (2013) summarize several studies of the gender wage gap in Austria and conclude that the gender wage gap hardly changed during the 1990s and that it decreased between 2002 and 2007 (Böheim et al. 2007(Böheim et al. , 2013. However, there is no study for Austria that uses a consistent source of data to analyze such development over a longer time. Earlier work used different data or different empirical methods. See, for example, Zweimüller and Winter-Ebmer (1994), Pointner and Stiglbauer (2010), Bundesministerium (2010), Grandner and Gstach (2015) or Christl and Köppl-Turyna (2020). This makes an assessment of the gender wage gap's evolution over time difficult.
We provide an analysis of the development of the gender wage gap in Austria for the period 2005-2017, using data from the Austrian EU-SILC (Austria 2018). The EU-SILC is the only long-term yearly survey which is currently available for Austria. It provides a range of personal characteristics and job-specific information which allows us to contrast changes over time, using a unified and consistent approach. In our sample of employees from the private and public sector, the wage gap without controlling for any differences between men and women declined from 18.6 (20.7) log points in 2005 (2006) to 15.0 log points in 2017. For employees in the private sector, this raw gap declined from 21.6 (23.6) in 2005 (2006) to 13.5 in 2017.
We expect to find a narrowing of the gender wage gap over this period for several reasons. First, women became more attached to the labor market over the last decades. Women's labor force participation rate in Austria in 2005 was 51.3% and it was 55.9% in 2017 (OECD 2019). In contrast, men's participation rates were 66.4% in 2005 and 66.8% in 2017. Second, women's educational attainment increased over time (Statistik 2019a). In 2000, 84.9% of women who were between 20 and 24 years of age had at least upper secondary level education; among men, it was 85.3%. By 2017, 90.2% of women and 84.7% of men aged 20-24 years of age had at least upper secondary education. Thirdly, the gender wage gap regularly features in political debates and several attempts have been made to address unfair wage discrimination by gender. For example, since 2014, banks have been required to formulate a quota for the board of directors and the executive directors to improve the representation of underrepresented employees (Wieser and Fischeneder 2019).
However, there are also reasons to expect little change in the gender wage gap over time. Although women's labor market participation increased, much of this increase is due to an increase in part-time work. In 2005, women's part-time rate was 40.4% and it was 48.3% in 2017. In contrast, men's part-time rate increased from 5.7% in 2005 to 11.0% in 2017. In terms of education, women and men tend to choose different fields and there has been little change over the years. For example, in the winter term of 2005, 10.3% of all male students and 7.9% of all female students enrolled in Science, Technology, Engineering, and Mathematics (STEM) studies. 1 In 2017, these numbers were 14.1% and 11.5%. In 2005, the most popular profession for female apprentices was sales (24.9%) and it was automotive engineering for male apprentices (8.6%). 2 In 2017, 23.5% of female apprentices trained in sales, which was still the most popular profession among women (Wirtschaftskammer 2018). Among men, automotive engineering was only the third most popular choice (9.5%) and metal engineering was the most popular (13.7%). While there have been political initiatives and legal reforms to reduce the gender wage gap and gender equality, for example, gender quotas for the board of directors in large companies, most of these seem to mainly raise awareness as they do not involve penalties or fines.
We decompose the gender wage gap into an explained and into an unexplained part, using several standard decomposition methods (Blinder 1973;Oaxaca 1973;Neumark 1988;Reimers 1983;Cotton 1988;Juhn et al. 1991;Oaxaca and Ransom 1995) to analyze which characteristics are associated with the evolution of the gender wage gap over time. In this way, our results contribute to the public debate which usually focuses on the question of how much of this difference is based on differences in characteristics, and how much is not based on such differences, perhaps due to unfair treatment of women. (Unequal treatment either due to prevailing gender stereotypes or limited access, e.g., to certain educational tracks or occupations, can, of course, result in observed differences.) A part of the gender wage gap might be due to other factors not included in our analyses, e.g., proficiency in skills, as these data do not provide such information.
Our results allow an assessment if and how women's improved human capital, such as educational attainment and labor market experience, contribute to a closing of the gender wage gap. The main determinant of the decline in both the private and the public sector is the relative improvement of the women's observed and unobserved characteristics. Using the decomposition method by Neumark (1988), we find that, after controlling for human capital, occupation, and other explanatory variables, the gap in a sample of both public and private sector employees shrank from 8.7 (8.8) log points in 2005 (2006) to 5.07 log points in 2017. Analyzing 1 3 only private sector employees, the unexplained gap declined from 9.9 (9.5) in 2005 (2006) to 5.06 log points in 2017. All our decompositions indicate that the unexplained part of the gender wage gap decreased substantially over the last ten years. The decrease of the unexplained gender wage gap between the largest gap in this period (2006) and the most recent gap (2017) ranges from 3.7 log points to 8.5 log points depending on the decomposition approach.

Data
We use data from the Austrian European Union Statistics on Income and Living Conditions (EU-SILC) covering the years 2005-2017 (Austria 2018). EU-SILC is an annual survey of about 6000 households with about 14,000 persons. The survey is a combined cross-sectional and longitudinal survey where each year about a quarter of observations are dropped from the survey, while a similar number of observations is added. Each quarter of the sample remains in the survey for four years. The survey collects data on income, poverty, social exclusion, housing, labor, education, and health at the household and individual levels.
Our empirical analysis uses wage regressions that also control for sample selection using a Heckman procedure (Heckman 1979). We include persons who are between 20 to 60 years of age and analyze the wages in their main job. Our main sample consists of both private and public sector employees. We repeat some of our analyses also for the private sector separately and show the results in the Appendices 1 (tables) and 2 (figures). EU-SILC does not provide an hourly wage. We calculate the hourly gross wage by dividing the usual monthly earnings (including overtime and bonuses) by the number of usual hours in paid work. We deflate all wage data to prices of 2014 using the Consumer Price Index (CPI) (Statistik Austria 2019b). Men and women might self-select into the labor market and in order to correct our estimates for sample selection, we add non-working persons to our sample.  Figure 4 in Appendix 2 shows the implied distribution of log hourly earnings by gender in 2007, 2009, 2011, 2013, 2015, and 2017. In all years, we observe that the distribution of women's wages was to the left of the distribution of men's wages. This indicates that women's wages were on average lower than men's wages. We also observe that the gap between female and male wages was rather constant over the wage distribution. 4  We use a range of variables which describe personal characteristics, household characteristics, and job-related characteristics. The summary statistics of all these variables are tabulated in Tables 11 (personal and household characteristics) and 12 (job characteristics) in Appendix 1 for the years 2007, 2009, 2011, 2013, 2015, and 2017. The variables are typical indicators that are thought to proxy the productivity of employees and firms, and thus should be relevant in the determination of wages. These are age, education, experience, health status, household size, nationality, occupation, contract type, sector, position in the firm, region, and firm size. 5

Empirical methods
In our empirical analyzes, we examine how wage differences between men and women in Austria have evolved over time. We first analyze the wage gap for each year from 2005 until 2017 using Blinder-Oacaxa-type decompositions (Blinder 1973;Oaxaca 1973;Neumark 1988;Reimers 1983;Cotton 1988;Oaxaca and Ransom 1995). The descriptive analysis shows an increase in the wage differential at the beginning of our observation period and a decrease towards the end of the period. As we are particularly interested in which variables contribute to the observed changes, we analyze in a second step the change in the wage gap between 2006, when the gender wage gap was the greatest in this period, and 2017, the latest available data, using Juhn-Murphy-Pierce decomposition techniques (Juhn et al. 1991).
We decompose the differences in the mean wages of women and men based on the technique first developed by Blinder (1973) and Oaxaca (1973). We estimate wage equations separately for women (W) and men (M) with ordinary least squares, where y ig is the hourly wage of employee i, i = 1, … , N , of gender g, g = M, W ; g are the coefficients to be estimated; X ig is a vector of characteristics; and ig is an i.i.d. error. Besides indicators for human capital, family structure, occupation, and firm characteristics, we also include a Heckman selection term in X ig to account for different probabilities of working (Heckman 1979). In the participation equation, we use as identifying variables the number of children between 0 and 2, between 3 and 5, between 6 and 9, as well as those between 10 and 18. We also use the health status of the partner as an exclusion restriction. The underlying assumption is that children and chronically sick partners constrain the possibility for paid work, but do not impact on the wage itself. (This assumption might be violated if, for example, persons with children have lower bargaining power over wages.) The difference in the mean wages, Y g , can be decomposed (Oaxaca and Ransom 1995): where ̂ * is a weighted average of the coefficient vectors, i.e., ̂ * = Ω̂M + (I − Ω)̂W , with a weighting matrix Ω and the identity matrix I. The first term on the right-hand side of Eq. (2) is the difference between men and women in their mean characteristics, evaluated at ̂ * . It is that part of the wage gap that is due to observable differences, for example, proxies for productivity such as educational attainment. The sum of the second and the third term is the part of the wage gap which cannot be ascribed to observed differences. We calculate the decomposition separately for each year in our sample.
The decompositions proposed by Blinder (1973) and Oaxaca (1973) represent special cases of Eq. (2), where Ω is either equal to I or the null-matrix. In the first case, Eq. (2) corresponds to the "male-based" decomposition which assumes that men are paid their marginal product and women are negatively discriminated against. In contrast, when Ω is the null-matrix, a "female-based" view assumes that women are paid their marginal product and men are positively discriminated against. We first follow Neumark (1988) and estimate a pooled model to derive the counterfactual coefficient vector ̂ * . We then also apply Reimers (1983) who assumes ̂ * = (1∕2)̂M + (1∕2)̂W . Finally, because the number of men, n M , and the number of women, n W , differ in our sample, we use Cotton (1988) who weights the coefficients by the group sizes, n M and n W , i.e., For the second part of our analysis, we decompose the differences in the gender wage gap over time into a portion due to gender-specific factors and a portion due to differences in the overall level of wage inequality (Juhn et al. 1991). 6 Suppose that wages for employee i of gender g in period t is given by the following equation: where y igt are gross hourly wages, X igt is a vector of explanatory variables including a Heckman selection term, gt is a vector of explanatory coefficients, igt is a standardized residual (i.e., with mean zero and a variance of one for each point in time), and gt is the period's residual standard deviation of wages (i.e., the unexplained level of wage inequality among men).
The difference in the average wages of men and women at time t is given by: where Y gt refers to average wages of men and women at time t, and Δ is the difference operator. A change in the difference between two periods t and s can be decomposed as:

3
The first term on the right-hand side measures the change in the differences in observed characteristics X between men and women over time. It describes how differences between men and women in, for example, educational attainment or experience have changed over time. The second term measures how the differences in the observed returns to education or experience have changed over time. A negative change can be interpreted as a reduction in differences in returns to education. The third term adjusts for simultaneous changes in quantities and prices. The fourth term measures the effect of differences in the relative residual wage position of men and women over time, i.e., the relative ranking of women within the male residual wage distribution. Such differences in rankings may reflect gender differences in unmeasured characteristics or the impact of labor market discrimination against women. Unmeasured characteristics could be negotiation outcomes (Artz et al. 2018) or personality traits as the propensity to compete. Research by, for example, Niederle and Vesterlund (2007) has shown that women shy away from competition and men compete too much. With this term we may measure how differences in these characteristics change over time. Again, a negative term may indicate that women have caught up in these characteristics over time. The fifth term measures the part of the wage gap that is due to changes in residual inequality, i.e., how changes in unobserved prices for the unobserved quantities affect the change in the wage gap. This term assesses the changing prices, for example, for the propensity to compete or that the amount of discrimination has changed over time. The last term again adjusts for simultaneous changes in unobserved quantities and unobserved prices.

Decomposition results
We start with a presentation of the results of several Blinder-Oaxaca-

Results of Blinder-Oaxaca decomposition for all years
In Table 2, we present the results from decompositions of the gender wage gap for the years 2005 until 2017. In each year, we use the year's pooled sample as the reference distribution (Neumark 1988 The decomposition results indicate that over the whole period the unexplained gender wage gap decreased as well. In 2005, the differences in observed characteristics explained 9.88 log points of the total gap, or 53%, the unexplained part was 8.74 log points. The unexplained part was 8.81 log points (42%) in 2006 and 5.07 log points (34%) in 2017. From 2006 until 2017, the unexplained part of the wage gap declined by 3.75 log points.
These results indicate that the reduction in the raw gender wage gap was driven by a change in the explained and the unexplained gap. Both a reduction in the difference between men's and women's observable characteristics as well as in unobservable characteristics contributed to a lower gap over time. In Appendix 1 in Table 15, we also present results of a Blinder-Oaxaca decomposition where we use only observations of private sector employees. We observe the same pattern as in the overall sample. The raw gender wage gap was, however, wider than in the overall sample and declined more pronouncedly.
To assess the robustness of these results, we also calculated the explained and unexplained gap from several other decomposition approaches. In Table 3, we present results from a male-based decomposition (Blinder 1973;Oaxaca 1973), from a coefficient weighting scheme as proposed by Reimers (1983) and Cotton (1988), and from a decomposition that uses a pooled regression with a group dummy variable. The decrease of the unexplained gender wage gap between 2006 and 2017 ranges from 3.7 log points to 8.5 log points depending on the decomposition approach. While the true counterfactual wage distribution is unknown, all calculated decompositions indicate that the unexplained part of the gender wage gap decreased substantially over the last ten years. Figure 1 is a graphical representation of Table 3 for selected decompositions. We observe a declining gender wage gap both with and without controlling for differences in characteristics, such as education, experience, occupation, industry, degree of urbanization, firm size, and hierarchy level. 8

Results of step-wise Blinder-Oaxaca decomposition for 2006 and 2017
To illustrate the effect of human capital variables such as education and experience in comparison to occupation, industry, hierarchy, and other explanatory variables, . Raw gap and residuals from different decomposition methods. Explanatory variables used in all decompositions are age, education, experience, health status, household size, nationality, occupation, contract type, sector, position in the firm, region, firm size, and a correction (inverse Mill's ratio) to adjust for non-random selection into employment we estimate Blinder-Oacaxa-type decompositions and add groups of explanatory variables sequentially. We start with a specification that only includes indicators for formal educational attainment. In a next step, we add variables related to experience, followed by occupation, status, and degree of urbanization. Finally, we add firm size and position in the corporate hierarchy. We present the results for 2006 and 2017 in Table 4. In 2006, the raw wage gap was 20.71 log points. Using Neumark's (1988) approach, about 14% of this difference is attributed to the different educational attainments that men and women have in our sample. The in this way corrected gap is about 17.84 log points, i.e., 86.1% of the wage gap cannot be explained by differences in educational background. Observed characteristics which are related to labor market experience (experience, marital status, and part-time work) reduce the gap further to about 12.08 log points, or 58.3% of the raw gap. This shows that the difference in men's and women's labor market experience explains a substantial part of the gender wage gap. Additional characteristics such as status, occupation, industry, degree of urbanization, and country of origin reduce the gap to 10.39 log points, or to about half of the raw gap. Accounting for differences in the size of firms and the hierarchy levels of men and women reduce the gap to 8.81 log points, which leaves 42.5% of the gender wag gap unexplained.
Comparing these numbers with those from 2017 shows that in 2017 differences between men and women in education contributed less to explaining the gender wage gap. This indicates that men and women differed less in their education in  1 3 2017 than in 2006. The gender gap in 2017 was 14.89 and differences in education explain about 3.4% of the overall wage gap. In other words, accounting for differences in formal education between men and women leaves 96.6% of the wage gap unexplained. However, differences in labor market experience are related to differences in wages. The wage gap after accounting for differences in education and experience is 9.34 log points, which leaves 62.7% of the gap unexplained. 9 Accounting for differences in status, occupation, industry, degree of urbanization, and country of origin reduces the gap to 6.26 log points (42%). 10 Accounting for different firm sizes and hierarchy levels lowers the unexplained gap to 5.07 log points (34%). 11 In Table 4, we also show the results using the approach by Reimers (1983). We observe a similar pattern. However, the share of the unexplained part is larger than for the decomposition by Neumark (1988).

Detailed Blinder-Oaxaca decomposition for 2006 and 2017
In Table 5, we show detailed decomposition results for the years 2006 and 2017 and the contributions to the explained and the unexplained part. On average, men and women became more similar in many observed characteristics which is illustrated by the estimated coefficients tabulated in the first panel ("explained") of Table 5. Women acquired notably more labor market experience over time and, on average, became more similar to men in this respect. This can be seen in the estimated explained contribution to the gender wage, which in 2006 was about 3.1 log points. This estimate implies that one part of the gender wage gap in 2006 was due to women having less labor market experience than men. By 2017, this difference had become smaller and differences in labor market experience were responsible for about 1.5 log points of the average gender wage gap. Industrial segregation, however, deepened over time and observed differences between men and women led to greater differences in wages than before.
In the second panel of Table 5, we detail the contributions to the unexplained part of the gender wage gap. A positive (negative) unexplained contribution factor implies that for similar women and men the wage gap increased (decreased) for reasons not related to differences in the characteristic. We estimate that, if men and women had exactly the same characteristics, women would have earned on average about 8.8 log points less than men in 2006. For 2017, we estimate a gap of 5.07 log points. This implies that men and women receive different wages for the same observed characteristics, but the difference became smaller over time. A closer look at the explanatory variables reveals that, for example, men who cohabited with a partner received a wage premium in 2006 of 6.39 log points in comparison to women who cohabited with a partner. This premium decreased over time and we estimate it was 0.16 log points in 2017.
Although women's labour force participation rate increased over time, in 2017, it was 7 percentage points (or 16%) less than men's. It is possible that women who are highly productive, and would therefore command a high wage, do not receive a corresponding wage offer. If this were the case, these women most likely decide against participating in the labor market-and women who actually work are those who are, on average, less productive. Or, in contrast, employed women are those who are on average more productive. This would be the case, if the less productive do not receive a sufficiently high wage offer. An observed gender wage gap could be due to such differences in participation due to underlying differences in productivity. We estimate that women who work are positively selected as the estimated coefficient for the inverse Mill's ratio ( ) is negative, i.e., if women who do not work would work, they would receive a lower wage. However, in most years we fail to obtain statistically significant evidence. We also estimate participation equations for men and their selection into the labour market is similar to women's. Again, the estimated coefficients for the inverse Mill's ratio in the wage regressions are negative, however, in most years they are not statistically significant. This can also be seen by the relatively low estimated contribution of the explained and unexplained component due to selection, their sum was about 0.08 log points in 2006 and about − 0.36 log points in 2017. This evidence suggests that the gender wage gap is not due to who among men and women select to participate in the labour market. Rather, as we document in Table 5, differences in labor market experience, occupation and industrial segregation, and labor market attachment are the predominant reasons for the gender wage gap in Austria.

Juhn-Murphy-Pierce decomposition for 2006 and 2017
In Table 6, we present the decompositions of the change of the gender wage gap over time into three components (Juhn et al. 1991). The three components are calculated separately for the changes in the explained and in the unexplained part of the gender wage gap. The first component attributes changes to changes in the groups' characteristics over time ("quantity effect"), the second component to changes in residual inequality, i.e., changes in prices ("price effect"), and the third component to simultaneous changes in characteristics and prices ("interaction effect"). We estimate that the gender wage gap decreased between 2006 and 2017. Using the decomposition based on pooled regressions as the reference wage distribution, we find that both the explained and the unexplained components decreased. However, the unexplained part decreased more than the explained part. In line with the Oaxaca-Blinder-type decompositions, we find that the gender wage gap in hourly gross wages declined between 2006 and 2017 (− 5.82 log points). The smaller gap was mainly due to a smaller explained component of the wage gap, which is decreasing from about 11.90 in 2006 to 9.82 log points in 2017, a difference of 2.07. The unexplained part of the wage gap decreased from 8.81 in 2006 to 5.07 log points in 2017, a difference of 3.75. The change in the explained part of the gender wage gap and the change of the unexplained part were both due to a substantial shift in observed and unobserved characteristics. Men and women became more similar in those characteristics which are valued in the labor market. 12 The main factor that narrowed the gender wage gap was the convergence in observed and unobserved characteristics. This can be seen by the two quantity effects, tabulated in column (1) of Table 6, which are − 3.5 log points due to changed observed characteristics and − 3.25 log points due to changes in unobserved characteristics. Reasons for the reduction in unobserved characteristics could have been policies that helped women to catch up with men. For example, the change in the equal treatment law that urged firms to be more open about the wages they pay their employees may have provided better guidance for wage negotiations. Riley-Bowles et al. (2005), for example, show that women tend to negotiate more effectively when there is less ambiguity about wages. Additionally, the change in parental leave subsidies introduced in 2010 or the increase in childcare facilities that started in 2007, may have increased women's labor market attachment.
The view that increased similarity of men's and women's characteristics is the main reason for the decline in the gender wage gap is also supported when we use Reimers's (1983) weighting scheme. We estimate a change of − 3.0 (− 5.3) log points due to changed observed (unobserved) characteristics. Unlike the results from Neumark's (1988) approach, however, this approach shows that the change in the gender wage gap was hardly affected by the changes in the explained part as the increasing similarity between men and women was offset by increasing gender-specific differences in how these characteristics are valued (3.33 log points).
In Table 7, we show how various variables contributed to the reduction in the explained part of the gender wage gap based on Neumark's (1988) method for decomposing the wage gap. We observe that women caught up in their educational attainments and labor market experience. Smaller gender-specific differences in occupational segregation and in having a leading position reduced the gender wage gap. Most reductions are driven by the quantity effect, where a negative contribution indicates that men and women became more similar. A positive price effect, for example, wage differences associated with wages in economic sectors (industry), widened the wage gap and off-set some gains due to more similarity in observable characteristics. We interpret this as evidence that women receive lower wages when they (increasingly) work in male-dominated industries.
The quantitative most important part of the change in the gender wage gap between 2006 and 2017 is, however, the reduction in the unexplained gap (Table 6). This reduction is mainly determined by fewer differences in unobservable characteristics, which could include negotiation skills. To a smaller extent, the reduction in the unexplained gap is also caused by a price effect, i.e., a smaller difference in how unobserved characteristics are priced.

Gap over the business cycle
In a further step of our empirical analysis, we relate the gender wage gap to the business cycle. We use the unemployment rate as a measure of the business cycle and show the correlation with the raw and with the unexplained gender wage gap in  Fig. 2. We observe that the unemployment rate is negatively correlated with both the raw and the unexplained gender wage gap. We find that the higher the unemployment rate, the lower is the gender wage gap.
To assess this effect in more detail, we also estimate OLS regressions where we use different measures of the gender wage gap (raw, Heckman adjusted, explained, and unexplained) as dependent variables and the unemployment rate as the explanatory variable. For three of the four dependent variables, we estimate that the gap is negatively correlated with the unemployment rate. The coefficients are statistically significant from zero at conventional levels and support the view that the gender wage gap is responsive to the business cycle. When we use the unexplained wage gap that also controls for selection, we estimate a strong correlation between the two variables. This highlights the necessity to control for selection as job opportunities for men and women change differentially over  the business cycle. Of the four dependent variables, the explained gender wage gap is positively correlated, but its correlation is not significantly different from zero. 13 Increased unemployment may lead to more competition for jobs and this could lead to a lower gender wage gap. An alternative view is that unemployment changes the composition of male and female employees, resulting in a more similar distribution of characteristics which are demanded by firms. Because the correlation between the unemployment rate and the raw wage gap is stronger than with the unexplained wage gap, the first explanation seems more plausible (Table 8). Wage gap in log points 1983 1986 1989 1992 1995 1998 2001 2004 2007 2010 2013 2016 Year Raw   The earliest study on Austria used data from 1983 (Zweimüller and Winter-Ebmer 1994) and showed a rather large gap of more than 30 log points. In addition, Zweimüller and Winter-Ebmer (1994) used net wages which usually result in a smaller wage gap due to progressive taxation.

Comparison to the literature
Hourly wage data were not regularly collected in Austria and thus there was no further study until the mid-1990s. From then onwards, several studies produced estimates of the gender wage gap. These, however, used different data sets (Mikrozensus, EU-SILC or tax data) or different econometric methods (inclusion of an indicator for women in OLS regressions or different variants of Blinder-Oaxaca decompositions). Despite these differences, it becomes quite apparent that the raw and the unexplained gender wage gap declined over time.
Additionally, we estimate a meta-regression using the results from the studies cited above. We use as the dependent variable the raw gender wage gap and the unexplained gender gap observed in various studies on Austria and as explanatory variables characteristics of the data set and the decomposition method the studies used. Table 9 shows the results from this meta-regression for two samples. The first sample includes all studies and the second sample is limited to studies which used more recent data from 2003 onwards.
Our results in columns (1) and (3) of Table 9 show the results for the raw gender wage gap. We observe that in both samples the raw gender wag gap decreased by about 0.3-0.4 log points per year. This gap is by about 2.0-2.2 log points greater if only data from the private sector was used compared to using data from a combination of public and private sector employees. The use of cross-country data implies a greater raw gap by about 4.7-5.2 log points compared to the use of Austrian data only. We do not estimate statistically significant differences for other sample or study specific characteristics.
In columns (2) and (4) of Table 9, we tabulate the results for the unexplained wage gap. This gap decreased by about 0.6 log points per year. This is slightly more pronounced than the decrease of the raw gap over time. The comparison of the different results show that the unexplained gap is greater for private sector employees than for public sector employees. It can be seen that studies that use only data on private sector employees estimate a greater wage gap than studies that use a combination of both private and public sector employees. This suggests that unobserved characteristics or unequal treatment is more important in the private sector than in the public sector.
The estimated unexplained wage gaps are smaller when the samples consist only of full-time employees. Compared to results that are derived from samples of both part-time and full-time employees, the unexplained gap is about 2.4-4.5 log points smaller, although this result is only significantly different from zero (at conventional levels) in the sample of studies that include older ones.
We also find that the unexplained wage gaps are greater when the results are based only on data from larger firms (more that 10 employees). Compared to all firms, the unexplained gap is greater by about 2.5-2.8 log points. This result is significantly different from zero at conventional levels only for the sample that includes older studies. As expected, the chosen decomposition technique is associated with the size of the unexplained wage gap. Studies that use a Neumark decomposition or which use a binary indicator for the gender result in systematically smaller estimates of the unexplained wage gap than studies that use a male-based Blinder-Oaxaca decomposition. Estimates which are based on a Neumark decomposition are 6.0-8.3 log points smaller and those that use the binary indicator are 3.2-3.3 log points smaller than estimates from a male-based decomposition.

Summary and conclusions
We examined the gender wage gap in Austria using EU-SILC data from 2005 to 2017. Using standard decomposition techniques, we decompose the gender wage gap over time. We find that the raw wage gap declined from 18.6 (20.7) log points in 2005 (2006) to 15.0 log points in 2017. Controlling for observed differences between women and men in human capital, occupation, and other explanatory variables, we find that the unexplained part of the gender wage gap decreased substantially over the last ten years. The decrease of the unexplained gender wage gap between the largest gap in this period (2006) and the most recent gap (2017) ranges from 3.7 log points to 8.5 log points depending on the decomposition approach. Using the method by Neumark (1988), we find that the unexplained gender wage gap shrank from 8.7 (8.8) log points in 2005 (2006) to 5.1 log points in 2017. We find that differences in the observable characteristics such as educational attainment, experience, occupation or industry have become a more important part for the gender wage gap between 2006 and 2017. The remaining part of the wage gap between women and men might be caused by differences in unobserved characteristics, e.g., attitude or commitment, or unfair discrimination against women. Using the approach suggested by Juhn et al. (1991), we find that the main determinant of the shrinking wage gap over time is the relative improvement of women's observed and unobserved characteristics. Men and women became more similar in observed and unobserved characteristics over time, and this contributed substantially to the reduction of the gender wage gap. One reason for the reduction in observed and unobserved characteristics could have been the implementation of policies that helped women to catch up with men. For example, the change in the equal treatment law that urged firms to be more open about the wages they pay their employees may have provided better guidance for wage negotiations. Riley-Bowles et al. (2005), for example, show that women tend to negotiate more effectively when there is less ambiguity about wages. Additionally, the change in child-care allowance introduced in 2010 or the increase in childcare facilities that started in 2007, may have increased women's labor market attachment.
Our results are consistent with earlier research that showed a narrowing of the gender wage gap over time. For example, Böheim et al. (2013) found that wage differences declined moderately between 2002 and 2007. However, the difference in the raw gender wage gap is still large due to differences in observed and unobserved characteristics between men and women. Labor market experience, occupation and industrial segregation, and labor market attachment are still important aspects where men and women differ, which results in average wage differences.
Funding Open access funding provided by University of Vienna. This project received funding from the Anniversary Fund of the Oesterreichische Nationalbank (Number 16273). We thank Silvia Rocha-Akis for her comments and the participants at the Workshop Arbeitsmarktökonomie 2017 for their discussions and suggestions. All errors and opinions are the authors' sole responsibility.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.