[01] Democracy, Economic Freedom and Taxation in the European Union... 5 Petru-Ovidiu MURA Table of content [02] Compatibility or Dichotomy Between Fiscal Discipline and Growth: Are There Lessons to be Drawn From the Turkish Experience?... 24 Sinan SÖNMEZ [03] Education and Economic Growth: an Empirical Analysis of Interdependencies and Impacts Based on Panel Data... 39 Liana SON Graţiela Georgiana NOJA Mihai RITIVOIU Roxana TOLTEANU [04]... 55 Karolina GORAUS [05] Bankruptcy Prediction Ahead of Global Recession: Discriminant Analysis Applied on Romanian Companies in Timiş County... 70 Daniel BRÎNDESCU-OLARIU Ionuţ GOLEŢ 3 Timisoara Journal of Economics and Business ISSN: 2286-0991 www.tjeb.ro Year 2013 Volume 6 Issue 19
GENDER GAP IN LABOR MARKET PARTICIPATION IN ROMANIA Karolina GORAUS 1 This article concentrates on the problem of the access of females to the labor market in Romania. The raw gap in labor market participation rates over the period 1997-2008 amounts to app. 22%. Demographic characteristics do not seem to explain this gap - the adjusted gap in labor market participation obtained after controlling for the age, education, place of living, and presence of young kids in the household is as high as the raw gap. After adding to the explanatory characteristics the existence of previous working experience, part of the gap, amounting to around 7%, becomes explained by differences in endowments. This means that although demographic characteristics do not explain differences in participation of females and males in Romanian labor market, jobrelated characteristic may do so. However, access to valuable job-related characteristics for females might be also determined by discriminatory behavior, thus it is not clear if they should be controlled for when decomposing gender gaps. In order to decompose gender gaps in labor market participation in Romania we implemented a relatively new non-parametric decomposition technique. After covering already 11 years of data, we were not able to identify any improvement when it comes to the access of females to the labor market in Romania. According to our findings the gender gap in labor market participation in Romania has even increased over the analyzed period. Keywords: Labor Market Participation, Gender Gap, Nõpo Decomposition JEL Classification: J21, J70. 1 MA, University of Warsaw, Poland.
1. Introduction One of the most important challenges in societies facing demographic change is to raise their employment rates. Earlier entrance to the labor market, higher retirement age, reducing unemployment, and improving labor market participation very often becomes targets of implemented policy measures. Much attention is allotted to the activation of groups that are more often separated from the labor market, especially females. However, before making decisions on policy recommendations, it is necessary to properly assess the situation of females on the labor market. The first question that should be answered is, if females who are not employed have characteristics that make them different from people currently working, or are there some other reasons that separate them from work. In order to address this question, we estimate adjusted gap in labor market participation between males and females. Gender differences on the labor market have been gaining considerable attention in the last decades. In addition to numerous theoretical and empirical papers emerging in this field, significant development of statistical tools created to decompose gender wage differentials have been observed. There are two main streams of these approaches, the traditional parametric and the relatively newer non-parametric decompositions. To the best of our knowledge, there is no empirical research using non-parametric methods to assess gender gap in labor market participation in Romania. Also the research on gender wage gaps or posttransition changes on the labor market is scarce. We aim to fill this gap by implementing decomposition technique developed by Nõpo (2008) in order to estimate gender wage gap in employment and labor market participation in Romania, accounting for the period 1997-2008. The analysis of differences in characteristics between males and females in Romania indicates that females are slightly less educated, and a smaller share of females has working experience. Thus it can be expected that part of the gap in labor market participation could be explained by those differences. However, the results of performed decompositions suggest that adjusted gap in participation in the labor market is almost the same as the raw gap, when controlling for demographic characteristics, such as age, education, place of living, and presence of kids under six years of age in the households. Thus, the gap in labor market participation over the period 1997-2008 that amounts to around 22% remains unexplained by differences in characteristics. In the second type of decompositions, we added one more characteristic, namely we controlled if the individual has working experience. The results of such decompositions indicate, that adjusted wage gap in labor market participation is around seven percent lower than the raw wage gap. This article is divided into three parts. After the introduction, the second section contains the literature review, the description of the data and research methodology, as well as the empirical results. The last section concludes. The findings suggest that the gender gap in labor market participation in Romania was increasing over the period 1997-2008. Both the raw gap, and the adjusted gap, was in 2008 around 90% higher than in year 1997. 56
2. Gender Gap on Romanian Labor Market 2.1. Literature review The issue of gender differences on the labor market has been a significant area of concern for theoretical and empirical research in economics, as well as a topic in social and political discussions, or even were important elements of election campaigns. However, transition economies had a significant delay in having their academic, business, and political elites concentrated on this issue. According to the most recent The Global Gender Gap Report 2012 (Hausmann et al., 2012), the value of Global Gender Gap Index for Romania has not decreased in comparison to previous years. Index that included variables on economic participation and opportunity, educational attainment, health and survival, and political empowerment puts Romania on place 67 among 132 countries covered. The value of the index in 2012 was only slightly higher than in the year 2006, when the index was for the first time calculated, thus the improvement in that area was minor. The index indicates that there is still much space for improvement when it comes to the issue of gender gaps in Romania. The mentioned report includes among its variables the ratio of female labor force participation over the male value, which is basic assessment of gender gap in labor market participation. However, this approach has limited explanatory power as it does not account for differences in characteristics between females and males. When explaining gender differences on the labor market, some people may claim that it is due to discrimination, and others that it simply reflects gender differences in some observable characteristics of the individuals that are determinants of success in finding employment. The question about the most important explanations accounting for gender gaps are typically answered using decomposition methods. This field of economics is not only deserved to explore differences in pay or employment rates between males and females. They can be used to control for observed characteristics in any measure for which it is expected to find some sort of explained and unexplained components. But it is in labor economics that decomposition techniques have been used the most extensively (Fortin et al., 2010). Seminal papers by Oaxaca (1973) and Blinder (1973) are among the most cited in labor economics, and the Blinder-Oaxaca decomposition is now a standard tool in applied economics. This technique requires the linear regression estimation of earnings equations for both females and males. Based on these earning equations, the counterfactual situation, that answers the question about the male (female) wage if the compensation scheme for his (her) individual characteristics aligned with the compensation schemes for females (males), can be generated. After some algebraic manipulations the difference in average wages between males and females is decomposed into two additive components: one attributable to differences in average characteristics of the individuals, and the other to differences in the rewards that these characteristics have. The latter component is considered to contain the effects of both 57
unobservable gender differences in characteristics that the market rewards and discrimination in the labor market. In a large number of methodological papers attempts to refine the Blinder-Oaxaca decomposition have been observed. One direction of developments was connected to the assumption in standard Blinder-Oaxaca decomposition that the male wage structure prevails in the absence of discrimination. Thus other non-discriminatory wage structures have also been observed in the literature (Cotton, 1988; Neumark, 1988; Oaxaca & Ransom, 1994; Reimers, 1983). As standard Blinder-Oaxaca decomposition is only informative about the average unexplained difference in wages, numerous papers aimed at expanding it to the case of distributional parameters besides the mean (Juhn et al., 1991; Machado & Mata, 2005; DiNardo et al., 1996; Firpo et al.2007). Blinder-Oaxaca decomposition is very useful in identifying causes of racial or gender differences not only in wages, but also in educational, labor market, and other outcomes. The technique is relatively easy to apply and only requires coefficients estimates from linear regressions for the chosen outcome variable and sample means of the explanatory variables used in the regressions. However, if the outcome variable is binary, such as employment, college attendance, or teenage pregnancy, the problem arises. Coefficients from a logit or probit model cannot be used directly in the standard Blinder-Oaxaca decomposition equation (Fairlie, 2003). A solution to the problem described above, was constructed by Fairlie (2003), who suggested a method of decomposition, in which estimates from logit or probit models were used. This relatively simple method was described in the analysis of the causes of the black/white gap in self-employment rates. Bauer & Sinning (2008) have generalized the Blinder-Oaxaca decomposition to other non-linear models and demonstrated how it can be applied to models with discrete and limited dependent variables. Another problem associated with the Blinder-Oaxaca decomposition is the misspecification caused by differences in the supports of the distribution of individual characteristics for females and males, as there may be combinations of characteristics for which it is possible to find males but not females in the society, and vice versa (Nõpo, 2008). With such distribution of characteristics one cannot compare wages or labor market status across genders. The problem with comparability is enhanced when job-related variables are included in the explanation of gender gap, as females tend to concentrate in certain occupations that demand particular abilities e.g. soft skills or empathy, while males concentrate more often in risky or managerial occupations. Nõpo (2008) adapted the tool of the program evaluation literature, matching, to construct a non-parametric alternative to Blinder-Oaxaca decomposition method and fix the problem of differences in the supports of distribution of characteristics between females and males. 58
Matching comparison techniques serve to find matched samples with similar observable features except for one particular characteristic, the treatment, which is used to group observations into two sets, the treated and the control group. After controlling for these observed characteristics it is possible to measure the impact of treatment alone. After the introduction of propensity scores in experimental design (Rosenbaum & Rubin, 1983) matching techniques started to be a useful tool in the estimation of causal effects in economics. For example Pratap & Quintin (2002) used propensity score matching to measure wage differences between the formal and informal sectors in Argentina. Nõpo (2008) went a step further and considered the gender variable as a treatment and used matching to select sub-samples of males and females in such a way, that there are no differences in observable characteristics between matched males and matched females. It should be mentioned that the assumption of Rosenbaum and Rubin (1983) about the ignorability of treatment required for propensity score matching is not likely to be satisfied in case the gender is perceived as treatment. Thus matching individuals in Nõpo is based on characteristics, not propensity scores. After grouping both females and males into matched and unmatched sub-samples Nõpo was able to develop decomposition that accounts for differences in the supports. 2.2. Data and research method The empirical part of this paper relies on the data on the level of occupational activity of the population in Romania by demographic and social features. The information comes from the European Union Labor Force Survey data collected from member countries by Eurostat. The available data set contains quarterly data from year 1997 to 2008. It should be also mentioned that for the years 1997 and 1998 the information was available only for the second quarter, thus the remaining quarters are just a replica of the 1997q2, and 1998q2 respectively. Thanks to the relatively big data set it is possible to analyze gender differences in characteristics and in participation in the labor market for each of 42 periods separately and analyze their evolution over time. Additionally the pooled data set was created, as sometimes presenting the results for each of the periods would not be transparent. This data set contains 1 724 634 observations. It should be mentioned that in the analyzed dataset there are only individuals between 15 and 64 years of age. The available datasets contain limited set of variables. Specifically we dispose of information on labor market status, age, education, place of living, presence of kids in the household, and existence of working experience. Table 1 contains descriptive statistics of the mentioned variables obtained from pooled sample containing all quarters 1997-2008. In the analyzed period the average age of individuals who can participate in the labor market is around 39 years. There are slightly more females than males, as their share in the sample is 50.9%. 59
Around 10% of individuals live in the household where kids under 6 years of age are present. When it comes to the level of education, the biggest share of the population, namely 56%, finished the stage of upper secondary education, which means that the highest level of education successfully completed by an individual was the third of fourth level of education, according to the International Standard Classification of Education 1997. Almost 9% finished third level education (5th of 6th stage of ISCED), and the remaining 35% has low education level (ISCED levels 0-2). Among the analyzed individuals of working age, more than 80% has some working experience. However, at the moment of the surveys, around 60% were wageemployed or self-employed, 65% were active on the labor market. Table 1 Variables at disposal Continuous variables Number of Standard Mean observations deviation Age 1,724,634 39.20 14.20 Categorical Variables Number of observations Percent Cumulative Sex 1,724,634 100.00 Females 877,001 50.90 50.90 Males 847,633 49.10 100.00 Presence of kids < 6 years in the household 1,724,634 100.00 No 1,545,608 89.60 89.60 Yes 179,026 10.38 100.00 Education 1,724,633 100.00 Low : lower secondary 608,188 35.30 35.30 Medium: upper secondary 966,405 56,00 91.30 High: third level 150,040 8.70 100.00 Region: NUTS-2 level (8 development regions) 1,724,634 100.00 Has working experience 1,681,406 100.00 No 317,968 18.90 18.90 Yes 1,363,438 81.20 100.00 Working 1,724,634 100.00 No 682,834 39.60 39.60 Yes 1,041,800 60.40 100.00 Active 1,724,634 100.00 No 606,907 35.20 35.20 Yes 1,117,727 64.80 100.00 Source: Own preparation. The technique that will be used to decompose gender gaps in labor market participation in Romania, is the non-parametric approach constructed by Nõpo (2008). When developing his method, Nõpo has been originally concentrated on wage gaps, and he was relating it to the Blinder-Oaxaca decomposition, which has been a traditional and broadly used tool to decompose wage gaps between two groups in society. Thus, it will be convenient to present technical aspects of the method also concentrating on wage gaps, although the chosen non-parametric decomposition method is equally useful to decompose gender gaps in case of binary variables, like labor market participation gaps which will be made in the empirical part of this work. 60
Nõpo s methodology that uses matching comparisons to explain gender wage differentials is a nonparametric alternative to Blinder-Oaxaca decomposition. Thus in order to present Nõpo s approach in the most understandable way it is worth providing the details of Blinder-Oaxaca decomposition in the first place. Almost forty years ago, Blinder (1973) and Oaxaca (1973) constructed a methodology to decompose differences in mean wages across two groups into explained and unexplained components. This decomposition requires the linear regression estimation of earning equation for both groups, in our case, for females and males:, and, where is an average wage of females or males, is the vector of average characteristics in each group, and is a vector of estimated coefficients of characteristics for females or males respectively. With such notations the raw gender wage gap can be expressed as. After adding and subtracting the average counterfactual wage that male workers would have earned under the wage structure of females,, the expression becomes.then, after some algebraic manipulations it takes the form ( ) ( ). Alternatively, the added and subtracted term might be the earning for the female with average individual characteristics, in the case she is rewarded for her characteristics in the same way as the average male is rewarded,. Then the wage gap takes the form ( ) ( ). It is worth mentioning that this alternative form is especially important for the purpose of this work, as Nõpo s decomposition is related precisely to this one. In both forms of decomposition the first components on the right-hand side, ( ) or ( ), are the part of the gap that is due to differences in average characteristics between males and females. In a broader context it is called the composition effect (Fortin et al., 2011). The second component, ( ) or ( ), is attributed to difference in average rewards to individuals characteristics and it is called the wage structure effect. The wage structure effect is also called the unexplained part of the wage differentials, or the part due to discrimination, although more precisely it should be perceived as the component containing the effects of both unobservable gender differences in characteristics and discrimination in the labor market. The Blinder-Oaxaca decomposition is very easy to use in practice, as it is only necessary to plug in the sample means and the OLS estimates in the presented formula. Various good implementations of this procedure are available in the existing software packages. Despite the undeniable advantages of Oaxaca-Blinder decomposition, Nõpo (2008) pointed out its limitations and developed an improved method for decomposing the gender wage gap. Nõpo points out that there are combinations of individual characteristics for which it is possible to find males, but not females, in the labor force, while there are also combinations of characteristics for which it is possible to find females, but not males. With such combinations of characteristics one cannot compare wages across genders. 61
The traditional Blinder-Oaxaca decomposition fails to recognize these gender differences in the supports by estimating earnings equations for all working females and all working males without restricting the comparison only to those individuals with comparable characteristics. In the Blinder-Oaxaca decomposition it is necessary to make an out-of-the-support assumption that the fitted regression surface can be extended for individual characteristics that have not been found empirically in the data set, using the same estimators computed with the observed data. The use of the matching criterion in Nõpo decomposition does not require any parametric assumptions and is solely based on the modeling assumption that individuals with the same observable characteristics should be paid the same regardless of sex. Nõpo also does account for gender differences in the supports. The traditional interpretation of two components as developed by Blinder and Oaxaca applies, but only over the common support. Additionally, in the Nõpo s four-element decomposition there are two elements that are attributable to differences in the supports. The mathematical reasoning of Nõpo is far more complicated than the one from Oaxaca- Blinder decomposition and presenting it in details lies beyond the scope of this work for more information one may refer to Nõpo (2008). However, for the purpose of this thesis the details about the matching procedure, and the estimated components of the decomposition should be introduced. Nõpo decomposes the gap in average earnings between females and males with the use of matching based on their characteristics, such as age, education and marital status. The procedure that is used to estimate the components of Nõpo s decomposition starts with resampling all females without replacement and matching each observation to one synthetic male, with exactly the same observable characteristics and having the wage obtained from averaging wages of all males exhibiting this set of characteristics. In the paper where the methodology is introduced Nõpo considers only characteristics that can be described with discrete variables and perfect matching. As a result of the matching procedure a partition of the data set is generated. The new data set contains observations of matched males, unmatched males, matched females, and unmatched females. Based on this partition the raw gender wage gap can be decomposed into four components:. The first of the four additive components,, is the part of the gap that can be explained by differences between two groups of males those whose characteristics can be matched to female characteristics and those who cannot. This component would disappear in two situations: if for each combination of individual characteristics exhibited in the group of males, it would be possible to find comparable females, or if those unmatched males would earn on average as much as the average matched males. As described by Nõpo (2008) this component is computed as the difference between the expected male wages out of the common support minus the expected male wages in the common support, weighted by the probability measure (under the distribution of characteristics of males) of the set of characteristics that females do not reach. 62
The second component,, is the part of the wage gap that can be explained by differences in the distribution of characteristics of males and females over the common support. This part corresponds to the component attributable to characteristics from Blinder-Oaxaca decomposition, namely ( ), however limited to the common support. The third component is called by Nõpo the adjusted gender wage gap. It is the part of the raw wage gap that remains unexplained by differences in characteristics of the individuals and is typically attributed to a combination of both the existence of unobservable characteristics that the labor market rewards and the existence of discrimination. This component correspond to the second component of the Oaxaca-Blinder decomposition, that is attributable to differences in average rewards to individuals characteristics for females and males, ( ), however it is also limited to the common support. The last component,, is the part of the gap that can be explained by the differences in characteristics between two groups of females, those who have characteristics that can be matched to male characteristics and those who cannot. As stated in Nõpo (2008) it is computed as the difference between the expected female wages in the common support minus the expected female wages out of the common support, weighted by the probability measure (under the distribution of characteristics of females) of the set of characteristics that males do not reach. Three components in Nõpo s decomposition can be attributed to the existence of differences in individuals characteristics that the labor market rewards ( ) and the other ( ) to the existence of a combination of both unobservable characteristics that should be included in the wage equation if these would be observed by an econometrician, and the discrimination. Thus the wage gap might be expressed as ( ), and interpreted as it is traditionally done in the linear Blinder-Oaxaca decomposition, with two components: one attributable to differences in observable features of males and females, and the other perceived as an unexplained component. It should also be mentioned that Nõpo s methodology has its limitations. It is burdened by the course of dimensionality. While the extent to which the raw gender wage gap can be explained depends on the number of explanatory variables, the likelihood of matching decreases with the number of explanatory variables. Variables that suit the methodology developed by Nõpo should thus be discrete, allow for precise estimation of unexplained component of wage gap, and at the same time keep the likelihood of matching females to males possibly high. To sum up, it can be said that the most important advantage of Nõpo s methodology over Blinder- Oaxaca decomposition is that it accounts for differences in the supports of the distribution. According to Nõpo, it is an empirical regularity that the unmatched males have average wages above the average wages of their matched peers and estimating earnings equations for all males without accounting for this regularity tends to overestimate the unexplained component ( ) in the Blinder-Oaxaca decomposition. However, in cases of countries where females exhibit desirable 63
characteristics that the labor market rewards to a greater extent than males, the unexplained component from the Blinder-Oaxaca decomposition could be actually underestimated. When Nõpo methodology is used for the case of binary variable, like labor market participation, the same procedure applies. In the following part of this article, after males and females were matched based on their characteristics, the adjusted gap in labor market participation is evaluated. 2.3. Empirical results Among 877,001 females covered by the pooled dataset 58.8% are participating in the labor market, namely are wage-employed, self-employed, or unemployed. After excluding those unemployed woman, we get the employment rate which amounts to 55%. The labor market participation rate and the employment rate are much bigger in the case of males: 71% of man between 15 and 64 years of age are active on the labor market, and 65.8% are working. The gap in labor market participation amounts to 22% of this rate for females. The gap in employment rate is slightly smaller, and amounts to around 20% of females employed over the analyzed period. That is due to the fact that among males active on the labor market, the bigger share is unemployed, than in case of active females. It may indicate that those women, who are not in employment, relatively more often, as compared to men, are leaving labor force instead of moving to unemployment. When it comes to demographic characteristics of males and females, the differences seem relatively smaller. Females are on average one year older. They also are slightly more often in the households where kids under 6 years of age are present this is probably related to the fact that kids are staying with the mother after parents decide to end their marital relation. The most important differences are related to education, and working experience. Among women covered by the surveys over the period 1997-2008 as much as 40% have completed only preprimary education, primary education of first stage of basic education, or lower secondary education or second stage of basic education, which corresponds to levels 0 2 from ISCED 1997 classification. In the case of men, 30% have completed only the lowest education levels. Thus, a bigger share of men - compared to women - have completed medium or high level of education. Among the females, around 52% have completed third or fourth ISCED 1997 level of education (upper secondary education, and post-secondary non-tertiary education). In case of men the share of those with medium education was around 60%. Also the share of men that have successfully completed first of second stage of tertiary education was bigger, then in the case of females, as there is 9% with fifth or sixth ISCED 1997 education level among males, and 8% among females. There are also gender differences with regard to the working experience. Among males between 15 and 64 years of age, 85% have some working experience, while among females the share is around 77%. After analyzing gender differences in characteristics, the decomposition of the gender gaps in labor market participation rate, and employment rate, can be performed. Firstly, the 64
decompositions are performed on the pooled dataset, so that the average gaps for the entire analyzed period could be divided into explained and unexplained component. Both for the gap in labor market participation, as well as for the gap in employment rate, two types of decomposition were performed. The first one included only demographic characteristics among control variables. Thus males were matched to females according to information on age (categories of five years span), place of living according to NUTS-2 division, presence of kids under 6 years of age in the households, and education. Although in Table 2 only three education levels are presented, in the datasets from year 1998 onwards the ISCED 1997 division into levels from zero to six is available, and the matching is based on this more detailed division then. It should be also mentioned, that we control for time, so that individuals are only matched within the sample of a particular quarter. Table 2 Comparison of characteristics between males and females Males Females Labor force participation rate 71.0% 58.8% Employment rate 65.8% 55.0% Mean Age 38.7% 39.7% Presence of kids<6 years in the household 10.1% 10.6% Education Low : lower secondary 30.4% 40.0% Medium: upper secondary 60.4% 51.8% High: third level 9.2% 8.2% Working experience 85% 77.3% Source: Own preparation In the second type of decomposition, we added one more variable to characteristics, namely the information if the individual has working experience. However, it is not sure if job-related characteristics should be included among controls. Adamchik and Bedi (2003) discussed possible criticism of inclusion of job characteristics while decomposing gender gaps. For instance, a number of job-related characteristics might be endogenous on the labor market. It is not clear if differences in job characteristics for males and females reflect employment discrimination, or different tastes and preferences, or both. In case of working experience, it could clearly be influenced by the presence of gender discrimination in access to the labor market. Still it is valuable exercise to see if the inclusion of this job-related characteristic will have an impact on the results of the decomposition. According to the results of the decomposition of gender gap in labor market participation rate, differences in demographic characteristics do not explain the gap that amounts to 22% of the labor market participation rate for females. In this case the adjusted gap is almost the same as the raw gap, and the share of matched individuals is 99% for females, and also 99% for males. However when we add the working experience, the results of the decomposition change strongly. The adjusted gender gap in labor market participation goes down to around 15%, which is due to the fact, that the component of the gap that can be explained by differences in the distribution of characteristics of males and females over the common support grows to 2%, 65
and the component which is due to differences between females matched to males, and those unmatched, amounts to 5%. The percentage of females matched goes down to around 94%. The results suggest that there are females in the sample, that most probably have no working experience, and taking into account also their demographic characteristics included, there are no men in the sample that can be matched to them. Those females are on average less often participating in the labor market, as compared to women that have sets of characteristics observable also among male individuals. Table 3 Results of decompositions Gender Gap in Labor Market Participation Rate Controls D D0 DX DM DF % of Females % of Males matched matched Demographic characteristics 22% 22.4% -0.6% 0% 0.2% 99% 99% + working experience 22% 15.3% 1.6% 0.1% 5% 93.6% 98.5% Gender Gap in Employment Rate Controls D D0 DX DM DF % of Females % of Males matched matched Demographic characteristics 20.3 21.3-1.2% 0% 0.2% 98.8% 99% + working experience 20.3% 11.8% 3.2% -0.1% 5.4% 93.6% 98.4% Source: Own preparation. When we analyze the gender gap in employment rates, the results are similar, as in case of the gap in labor market participation. When only demographic characteristics are taken into account, the adjusted gap that amounts to around 21% is even slightly bigger than the raw gap of 20%. In this decomposition 99% of females were matched to males, and vice versa. After adding the working experience the change in the results is even stronger than in case of the gap in participation rates. The adjusted gap goes down to around 12% and 8% become explained by differences in characteristics. The component that amounts to 3% is part of the gap due to the differences in the distribution of characteristics over the common support, which suggests that among females that can be matched to males there are less of those with characteristics important to become successfully employed. Especially important is the component of the raw gap in employment rates that amounts to slightly more than 5%, which can be explained by differences between females who are in the common support (93.6% of all females analyzed) and those unmatched females, who are relatively less often in employment. However, it cannot be judged if it is due to the preference of some women not to work and gain experience, or it is relatively harder for them to get the first job, because of discriminatory practices. Figure 1 and Figure 2 present the results of decompositions of gender gap in labor market participation rates, and in employment rates, for each quarter separately. It can be observed that gender gaps in both cases have increased over the period 1997-2008 of about 60%. The raw gap in labor market participation in 1997 amounted to around 17% and in the last quarter of 2008 it was already as high as 28%. Adjusted gap controlled for differences in demographic 66
1997q1 1997q3 1998q1 1998q3 1999q1 1999q3 2000q1 2000q3 2001q1 2001q3 2002q1 2002q3 2003q1 2003q3 2004q1 2004q3 2005q1 2005q3 2006q1 2006q3 2007q1 2007q3 2008q1 2008q3 1997q1 1997q3 1998q1 1998q3 1999q1 1999q3 2000q1 2000q3 2001q1 2001q3 2002q1 2002q3 2003q1 2003q3 2004q1 2004q3 2005q1 2005q3 2006q1 2006q3 2007q1 2007q3 2008q1 2008q3 Goraus, K. (2013). characteristics in each quarter was very similar to the raw gap. Differences of around one percent were observed in the first years of the analyzed period, but in the last years they were almost identical. Adjusted gaps that were controlled additionally for working experience in 1997 amounted to only 11%, while in the last analyzed quarter it was already 21%. Thus both the raw gap and adjusted gap were in last analyzed period around 90% higher than in the first analyzed period, indicating that the situation of females on the labor market in Romania is deteriorating. 0.27 Raw gap 0.22 0.17 0.12 0.07 Adjusted gap based on demographic characteristics Adjusted gap based additionally on experience Figure 1. Decompositions of gender gap in labor market participation rates over time Source: Own preparation. 0.3 Raw gap 0.25 0.2 0.15 Adjusted gap based on demographic characteristics 0.1 Adjusted gap based additionally on experience Figure 2 Decompositions of gender gap in employment rates over time Source: Own preparation. 67
In the case of gender gap in employment rates, both raw and adjusted, the increase over the period 1997-2008 is smaller and amounts to around 40%. The raw gap in employment rate in 1997 was 17%, and after 11 years it was already 25%. When working experience is included among controls, the adjusted gap increased from 11% at the beginning of the analyzed period to 15% in the last quarter of 2008. The analysis of employment rates might provide a clearer view, than the analysis of participation gaps, as females may join labor force, but remain unable to find employment due to discrimination. On the other hand, if females more often had preference to stay away from employment, they may more often register as unemployed, while in fact not being looking for employment. What is more, for people participating in the surveys the difference between unemployment and inactiveness can be vague. However, the analysis of gender gaps in participation over the period 1997-2008 confirms that the gender gap in accessing labor market in Romania is increasing. The results also indicate that there are seasonal changes in gender gaps in participation in labor market in Romania. As the gaps in the first and fourth quarter are much higher than in the second and fourth quarter, one may suppose that females are finding employment in agricultural employment. Explaining the seasonality of gender gaps in labor market participation in Romania lies beyond the scope of this work, but may be interesting area for further research. 3. Conclusions As Europe is facing demographic change, increasing employment rates becomes one of the most important targets of policies implemented. One of the areas for improvement is female labor force participation, which is often relatively low as compared to men. In case of Romania the gap in labor force participation over the period 1997-2008 amounted to 22%. What is especially important the raw gap has been increasing over the analyzed period, and in the last quarter of 2008 was about 90% higher than in 1997. The analysis of the gap in employment rates confirms the finding that the situation of females on the labor market in Romania is deteriorating. However, gender gaps in labor market participation or employment may simply reflect differences in observable characteristics between males and females in the working age. The real challenge lies in providing reliable measures of gender gaps, and huge development in this field has been observed in the last decades. For the purpose of this work, relatively new non-parametric approach developed by Nõpo (2008) has been applied. It decomposes gender gaps into four components, two of which are equivalents to those from traditional parametric Blinder-Oaxaca approach (first attributable to differences in characteristics and second to differences in rewards), and other two account for gender differences in the distribution of characteristics. Performed decompositions suggest that demographic characteristics do not explain gender differences in labor market participation in Romania. Adjusted gaps obtained when controlled for age, education, place of living, and presence of kids in the household as equally high as the raw gaps. Controlling if the individual has working experience has changed the results of the decomposition, namely the adjusted gap in labor market participation over the period 1997-2008 decreased to 15%. This suggests that job-related characteristics could possibly explain gender differences in labor market participation in Romania. However, it cannot be judged if differences in job characteristics for males and females reflect different skills, tastes and preferences, or they are related to employment discrimination. 68
References Adamchik, V.A., & Bedi, A. S. (2003). Gender Pay Differentials during the Transition in Poland. The Economics of Transition, 11(4), 697-726. Bauer, T., & Sinning, M. (2008). An Extension of the Blinder Oaxaca Decomposition to Nonlinear Models. Advances in Statistical Analysis, 92(2), 197-206. Blinder, A. (1973). Wage Discrimination: Reduced Form and Structural Estimates. The Journal of Human Resources, VII(4), 436-455. Cotton, J. (1988). On the Decomposition of Wage Differentials. Review of Economics and Statistics, 70(2), 236-243. DiNardo, J., Fortin, N. M., & Lemieux, T. (1996). Labor Market Institutions and the Distribution of Wages, 1973-1992: A Semiparametric Approach. Econometrica, 64, 1001-1044. Fairlie, R. W. (2003). An Extension of the Blinder-Oaxaca Decomposition Technique to Logit and Probit Models. Working Papers 873, Economic Growth Center, Yale University. Firpo, S., Fortin, N. M., & Lemieux, T. (2007). Decomposing Wage Distributions using Recentered Influence Functions Regressions, mimeo, University of British Columbia. Fortin, N., Lemieux, T., & Firpo, S. (2011). Decomposition Methods in Economics, in Handbook of Labor Economics, ed. Ashenfelter, O., & Card, D., vol. 4A, Amsterdam, North Holland, 1-102. Hausmann R., Tyson, L. D., & Zahidi, S. (2012). The Global Gender Gap Report 2012. World Economic Forum, Geneva. Juhn, C., Murphy, K. M., & Pierce, B. (1991). Accounting for the Slowdown in Black-White Wage Convergence, in Workers and Their Wages: Changing Patterns in the United States, ed. by M. H. Kosters, American Enterprise Institute, Washington. Machado, J. F., & Mata, J. (2005). Counterfactual Decomposition of Changes in Wage Distributions Using Quantile Regression, Journal of Applied Econometrics, 20, 445-465. Neumark, D., (1988). Employers' Discriminatory Behavior and the Estimation of Wage Discrimination. Journal of Human Resources, 23(3), 279-295. Nõpo, H. (2008). Matching as a Tool to Decompose Wage Gaps. Review of Economics and Statistics, 90(2), 290-299. Oaxaca, R. (1973). Male-Female Wage Differentials in Urban Labor Market. International Economic Review, 14(3), 693-709. Oaxaca, R. L., & Ransom, M. R. (1994). On Discrimination and the Decomposition of Wage Differentials. Journal of Econometrics, 61(1), 5-21. Pratap, S., & Quintin, E. (2002). Are Labor Markets Segmented in Argentina? A Semiparametric Approach. Instituto Tecnologico Autonomo de Mexico, Discussion Paper 02-02. Reimers, C. W. (1983). Labor Market Discrimination against Hispanic and Black Men. The Review of Economics and Statistics, 65(4), 570-79. Rosenbaum, P., & Rubin, D. (1983). The Central Role of the Propensity Score in Observational Studies for Causal Effects. Biometrika, 70, 141 55. 69