Размещено на http://www.allbest.ru/

Размещено на http://www.allbest.ru/

ФЕДЕРАЛЬНОЕ ГОСУДАРСТВЕННОЕ АВТОНОМНОЕ ОБРАЗОВАТЕЛЬНОЕ УЧРЕЖДЕНИЕ ВЫСШЕГО ОБРАЗОВАНИЯ

«НАЦИОНАЛЬНЫЙ ИССЛЕДОВАТЕЛЬСКИЙ УНИВЕРСИТЕТ

«ВЫСШАЯ ШКОЛА ЭКОНОМИКИ»

Международный институт экономики и финансов

Выпускная квалификационная работа

Анализ влияния профессиональной сегрегации по половому признаку на заработную плату мужчин и женщин и ее значение в гендерном разрыве в оплате труда: пример из России

(Analysis of the impact of occupational sex-segregation on the wages of men and women and its significance in the gender pay gap: Russian case)

Коваленко Анжелика Александровна

Москва 2020



Аннотация

Данная работа изучает влияние горизонтальной профессиональной гендерной сегрегации на почасовые зарплаты мужчин и женщин России в период после кризиса 2007-2008 гг. В качестве индикатора профессиональной сегрегации используется процент женщин в индустрии, так называемая профессиональная феминизация. Анализ взаимосвязи включает в себя контролирование способностей и инвестиций в человеческий капитал, а также использование разных выборок.

В результате анализа данных РМЭЗ за 2010-2018 годы была получена преимущественно статически незначимая зависимость между процентом женщин и почасовой зарплатой. Однако, на нескольких выборках эта связь оказалась положительной и статистически значимой.

Также в результате применения метода декомпозиции зарплат Оаскаки-Блайндера-Ньюмарка, на большинстве выборок не выявлен значимый вклад профессиональной феминизации в гендерный разрыв труда.



Abstract

This work studies the influence of horizontal occupational gender segregation on the hourly wages of men and women of Russia in the period after the crisis of 2007-2008. The percentage of women in the industry, the so-called professional feminization, is used as an indicator of occupational segregation. The analysis of the relationship includes controlling on skills and investment in human capital on different samples.

The analysis of the RLMS data for 2010-2018 shows a predominantly statistically insignificant correlation between the percentage of women and hourly wages. However, for several samples the relation turned out to be positive and statistically significant.

Also, as a result of applying the Oaxaca-Blinder-Newmark wage decomposition technique, the contribution of professional feminization to the gender labor gap was not found to be significant in the majority of the samples.



Contents

Introduction

Literature review

Hypotheses

Chapter 1. Gender segregation in Russia

Current context

Segregation

Features of Russian case

Chapter 2. Model and estimation

Model setup

Estimation results

Chapter 3. Contribution to the Gender Wage Gap

Description of an approach

Calculation

Conclusion

References

Appendix



Introduction

Economic and social inequality between men and women has been a topic of discussion for many years. With globalization and worldwide struggle for gender equality, the rights of men and women as well as their part in labor market seems to balance. For example, based on RLMS data, in Russia in 2010 women took 45% of managerial positions, while in 2018 it has become 51%, which means that women are getting better off in term of career improvement. However, despite the supposedly disappearing difference and high female engagement in the labor force, the gender wage gap still exists. According to the Global Wage Report (ILO, 2018/19), factor weighted wage gap, average difference in payments of men and women in Russia, is 24.9% in men`s favor, while the average factor weighed wage gap in the World is only 18.8%. Why this happens?

My conjecture is that occupation in specific industries, an entrenched division into “female” and “male” professions, which differ in prestige and earnings level is one of the causes. That is why I analyze in this paper the “occupational sex-segregation” - phenomena of horizontal gender segregation in some professional areas of work. Jobs that are generally held by women are considered as easier and less skill-intensive than jobs that are mostly held by men (He et al. 2019). Typical female occupations offer lower wages (e.g., Grцnlund and Magnusson, 2013) and fewer career opportunities (e.g., Kriesi et al., 2010). Even men, who occupy typical female jobs, also earn less than men, who occupy typical male professions (Beller, 1982). This leads to the question: is this gap in payments in female and male professions is indeed the consequence of skill and qualification differences or “female” professions are just devalued and discriminated? What is the direct relationship between occupational sex-segregation and wages?

The issue of occupational gender segregation has previously been discussed by other researchers. However, these works were not devoted directly to the relation between occupational sex-segregation and salaries, and the coefficient of segregation was estimated only by including dummies on the industry. This happens because inter-professional and inter-industry gender differences are accepted as given in works devoted to wage differences (Gimpelson and Kapelyushnikov, 2008). My work has three important differences. First, I study not just the factors that affect the wage gap, but directly the relationship between horizontal occupational segregation and employee wages, controlling not only on the default control variables, but also on the variables associated with the specialized human capital and skills. Second, as a proxy for segregation, I use the percentage of women in each industry for each year, which is called occupational feminization and has not been previously applied in Russia. Third, my work will be the first work of this type that is dedicated to the post-crisis 2007-2008 labor force situation in Russia.

In this work, I test on Russian data the existing findings that higher occupational segregation is associated with larger wage gaps (Cohen et al. 2003; Cotter et al. 1997; McCall 2001a; Ogloblin 1999). I analyze the relationship between percent of women in industry and hourly wages, controlling for skills and investments in human capital. I test the findings of Perales (2013) that he obtained on British data that hourly wages negatively correlate with feminization even taking skills into account meaning the devaluation of “female” professions. For robustness, different samples and specifications were examined. Additionally, I use the Oaxaca-Blinder-Neumark wage decomposition to see which share of gender earnings inequality in modern Russia is explained by the occupational gender segregation in a form of occupational feminization.

The results turned to be quite ambiguous. In the full sample and the sample consisting of people who participated in the panel three or more times, feminization turned out to be insignificant. However, in the samples without the salary outliers and in the strongly balanced panel, this relationship was identified as positive and significant. Thus, it can be said that, controlling for skills and specialized human capital, the correlation between the percentage of women and hourly wages is positive and does not support the hypothesis of Perales (2013) about devaluation of female professions and their discrimination. This result is rather surprising, because it contradicts previous findings about negative relation between salaries and occupational segregation, including those made on Russian data. In addition, during the wage decomposition technique I revealed that occupational feminization is not a component of gender earnings gap and even lowers the gap in some cases.

The rest of the paper is organized as follows. In the next section, I provide the literature review of relevant articles, and state the hypotheses. In the main part there will be three chapters. In Chapter 1, I discuss and compare occupational sex-segregation in modern Russia with other countries, highlighting its distinctive features. Chapter 2 is dedicated to modelling and estimation of the relation of hourly wages with occupational feminization and discussing the results. In Chapter 3, I describe the wage gap decomposition approach and calculate the contribution of occupational feminization into the gender earnings gap. Finally, I conclude by discussing the limitations of the research and possible improvements.



Literature review

In this section, I review articles which are dedicated to the issue of occupational gender segregation and its correlation with wages, their results and limitations.

Cohen et al. (2009) analyzed the changes in women status in managerial positions between 1980 and 2000 using U.S. Decennial Census to contribute in gender segregation study and examine the hypotheses posed by the Jacobs (1992) about female managers, one of which is “female managers have a long way to go before they reach parity with their male counterparts” (p. 298).

They examined three critical indicators of gender inequality: gender segregation, relative earnings, and the effect of gender composition on earnings. In regression, they used natural logarithm of annual earnings as the dependent variable and various explanatory variables (as natural logarithm of hours worked per week) together with controls and dummies (female industry controls, gender, education, race, marriage status, presence of young children etc.). Unlike most other works, they propose not one, but two segregation indexes- simple dissimilarity index(D) (Duncan & Duncan, 1955), and the size-standardized segregation index (SSI), which is an adjusted version if D, since D does not consider the shares of occupations in a labor force. SSI measures the percentage of women or men that would have move in other occupations to achieve equality, if the occupations were the same size. Through panels they got the results that occupations with greater female representation have lower wages and that the net gender wage gap is significant, but they found that this gap is decreasing over time, mostly in integrated managerial occupations. The limitation of this work is that they restricted their analyses only on managers, while I want to widen my analysis across all occupations to see more general picture.

Gauchat et al. (2012) considered issue more deeply, taking the U.S. metropolitan market as a representative of labor structure, as they are convinced that metropolitan areas are good approximations of labor markets and display a considerable amount of variation in gender earnings inequality. They refer to Acker (2004) who argued that instead of benefited from labor restructuring with new, higher-paying occupational opportunities, women continue to mainly occupy lower-paying, less prestigious industries; and some empirical research showing that globalization increases gender inequality in wages. Authors addressed the relationship between gender segregation and earnings using data from U.S. metropolitan statistical areas (MSAs) in 2000 year. As an occupational segregation measure, they used Duncan index of dissimilarity (D) (Duncan & Duncan, 1955). Through OLS and 3SLS authors found that men's wages are significantly positive related to women's wages, but not in opposite way, which means that men's earnings are central in labor market and women's earnings are based on them. Globalization is observed to benefit both men and women, but the effect for men is larger. Their work is limited by full-time workers, despite the fact that in work there was flashed the idea that the appearance of “part-time, temporary service, homework, and informal self-employment" is a "dark side" of globalization which harms women by inducing them to occupy positions in these areas of employment, thereby reducing their potential salary. I want to include all types of employment to control for time workers dedicate to work.

He and Wu (2016) in their work address the issue of the role of occupational gender segregation in creating the wage inequality in era of China`s economic development. One of the hypotheses stated in work is that the higher gender segregation is in a prefecture, the larger the gender earnings gap in that prefecture. As data they used one-percent population mini-census about monthly earnings of China in 2005. At first glance it might seem as a very small sample, but in the end, they had 577,363 observations from 283 prefecture-level jurisdictions in 29 provinces of China. As in works above, authors also used the dissimilarity index to measure the occupational gender segregation (Duncan & Duncan, 1955), and OLS regression to see the relations. As a result, the coefficient of interaction term between female and the dissimilarity is negative and statistically significant, indicating that in occupations with a higher level of gender segregation, women tend to earn less. I want to check if this result can be seen on Russian panel data.

Perales (2013) used panel data to investigate the impact of sex-composition of professional occupations on the gender wage in Britain. He used the regression of wages on the so-called “occupational feminization”, which is a proportion of women in individual occupations, controlling for skill specialization and individual-specific unobserved heterogeneity (i.e. person-level unique wage differentials). He computed OLS and FE estimators to check the impact of individual-specific factors. To account for human capital (skills) characteristics, author used five different approaches, from the post level to the job-learning time. The results in all regressions indicate not only negative and significant association of wages and occupational feminization, but also that its effect is larger for women than for men. However, author highlights that all the indications of skills are imperfect as they suffer from measurement error, since they are based not on the individual`s answers on their skill level, but rather on some occupational means. The significance of occupational gender segregation on the gender wage gap was calculated by a simple decomposition technique (Tomaskovic-Devey, 1995), getting the percentage of the gender wage gap that is explained by occupational gender segregation in all regressions used above. Perales got that workers in female-dominated occupations receive lower wages than those in other occupations, which is robust to controls for skills, and that the occupational gender segregation explains a sizeable portion of the gender wage gap, from 14% to 25% depending on the model. I will try to repeat this approach to study the issue of occupational segregation in my work on Russian data by using this “occupational feminization” measure. I will not be able to repeat all the controls on specialized human capital because of the lack of such data in the RLMS, but I will include the controls on professional courses, continuing education courses and skill level based on post level. To find the percentage of the wage gap explained by segregation, I will use the Oaxaca-Blinder wage decomposition method instead of Tomaskovic-Devey method, because the former is more widely applied.

Oglobin (1999), using RLMS household data for 1994-1996 years, analyzed the gender earnings differences in Russia in the conditions of transition from the centralized Soviet economy to Market economy. He did the regression analysis with the selectivity bias correction on factors affecting the gender wage differentials. To find the distribution of gap explained by different characteristics, he used Oaxaca-Blinder-Neumark method of gender wage gap decomposition. Author stated the hypothesis that occupational and earnings patterns remain rigid and unchangeable from the Soviet Union since they are based on cultural and historical foundations which cannot be changed so fast and these patterns remain a key determinant of gender wage gap. He founded that the gender wage gap at that time was 38%. All models, with and without selectivity bias correction, revealed that considerable part of gender wage gap is due to occupational sex-segregation in industries, accounting to 30% in uncorrected and 29,5% in corrected model. In Ogloblin (2005a) he repeated the analysis using the 2000-2005 data set and found that the gender wage gap was 45,4% and occupational sex-segregation in industries held 38,7% of this number. The limitation of these works is that Ogloblin did not look directly at the relationship between segregation and wages and measured effect of horizontal occupation segregation only by including dummies on industries, whereas in my work the special sex-segregation measure will be inserted to reveal the correlation.

Roshchin (2013) studied gender inequality in Russia as a whole, touching upon the topic of occupational segregation and its place in the gender wage gap. Using data from Goskomstat RF and RLMS data, he got that by the 2001 year the gender wage gap was about 40%. However, after the analysis of determinants of this gap he observed that the largest wage differences were in the occupations, where women working force was “in excess”: in those jobs wage gap reached 55%. The author also found out that women invest much more in their human capital, such as education, which helps to reduce the wage gap. He claims that if women invested in human capital on equal terms with men, the 2001 gender gap would be 7.4% more. Along with other determinants of gender differences in Russia in 2001, occupational segregation made a significant contribution, occupying a third of the entire gap. Using the index of dissimilarity (Duncan`s D), Roshchin showed that the segregation between 1994 and 2001 was quite stable on the level of 32%. Graphically, author managed to show that, dropping out the Agriculture and Finances, the wage and the share of women are negatively correlated. The limitation of this work is that the relationship between segregation and salary is discussed only graphically, where the trend can be seen only with the exclusion of several industries, one of which is Finance, which is one of the most important sectors in the economy. Nevertheless, this trend in his work signals the need for a deeper analysis of these relations.

Hypotheses

After the review of corresponding papers, I stated hypothesis that will be tested during the research:

Hypothesis 1

In Russia, wages negatively correlate with occupational feminization.

In other words, my assumption is that working in occupations with higher proportion of women is associated with lower wages. Estimation will be provided for different samples and in various specifications.

Hypothesis 2

Occupational gender segregation in a form of occupational feminization explains considerable part of gender wage gap in Russia.

I want to find a proportion of gender wage gap-average difference of wages of men and women, which can be explained by occupational sex-segregation. The threshold of 25% is based on average from the literature used, from 14% in work of Perales (2013) to the 38,7% in work of Ogloblin (2005).



Chapter 1. Gender segregation in Russia

This chapter is dedicated to the presentation and precise analysis of occupational segregation in Russia. Moreover, there will be discussed how segregation can be measured and what distinctive features does Russian labor market have over other countries.

Current context

To illustrate the current situation with occupational sex-segregation on Russia, I use the most recent RLMS data, which is for 2018 year. I calculated the percent of women in each occupation-industry and sorted them to highlight the most segregated occupations, and which occupations are “male” and “female”.

Graph 1. Feminization of occupations, 2018

Source: RLMS, 2018; own analysis.

As can be seen from the graph, such industries as “IT”, “Construction”, “Oil and Gas Industry”, “Wood, Timber and Forestry”, “Army, MIA, Security”, “Heavy industry”, “Energy and Power Industry”, “Agriculture”, “Transportation, Communication”, “Civil Machine Construction” and “Military” (where percent of women is less than 40 percent), can be considered as “male” occupations. At the same time, “Advertising and Marketing”, “Science, Culture”, “Real Estate”, “Catering”, “Government and Public Administration”, “Services”, ”Health”, “Education”, “Social services”, and “Ecology” (where percent of women is more than 70 percent) can be considered as “female” occupations. This hierarchy is quite reasonable, given the social orientation on the gender division of professions into “male” professions requiring physical strength or working with computers, and “female” professions requiring working with people. At the same time, “Housing and communal services”, “Public Organizations”, “Light and Food Industry”, “Chemical Industry”, “Trade, Consumer Services”, “Jurisprudence”, “Finances”, “ Sports, Tourism and Entertainment”, “Church” and “Mass Media” are observed to be the most integrated professions, where the proportion of male and female workers are more or less equal.

To look at the preliminary relationship between feminization (the percentage of women in the profession) and hourly salary, I calculated and plotted below the average hourly wages, combining with the previous graph:



Graph 3. Correlation of feminization and wage, 2018

Source: RLMS, 2018; own analysis.

There is no obvious connection between the feminization and average wages, although it seems that in the `integrated' industries the salaries are higher than in `female' ones. “Jurisprudence” and “Mass Media” seems to stand out from the integrated professions and can compete in salary level only with “IT” and “Oil and Gas Industry”.

It may be noticeable that in majority of the integrated professions (where the percentage of women is 50-65%) salaries are higher than the average, with a “Church” outlier, which is one of the lowest-paid jobs (89,1 rubles). At the same time, in some “male” professions, like “IT” and “Oil and Gas Industry” there are the highest wages (277,9 rubles and 258,6 rubles respectively), while in “Government and Public Administration”, which is a “female” profession - the lowest (85,4 rubles). Overall, some trend cannot be seen so far, that is why this issue need to be analyzed deeper.



Segregation

Occupational segregation is measured using Duncan's D, or the index of dissimilarity (Duncan & Duncan, 1955), which shows the degree of uniformity of the distribution of two groups in the labor market. D can be used not only for gender analysis, but also for others measures of discrimination (race, ethnicity, etc.). This index represents the proportion of a workers of one group that should move to other occupations to have the uniform distribution of workers of all groups across occupations. The formula for gender segregation index is the following for year t:

where m_it is the male population of the i^th occupation in year t, M_t = the total male population in the country in year t, f_it = the female population of the i^th occupation in year t, F_t = the total female population in the country in year t and N = number of occupations in year t.

Index of dissimilarity is the most vivid and accurate measure to show the overall picture of gender segregation and to compare countries among themselves. To illustrate occupational sex-segregation and its changes in Russia across time, I calculated D for 10-year period from 2008 to 2018 years using the data from RLMS:



Graph 4. Index of Dissimilarity in Russia, 2008-2018

Source: RLMS, 2008-2018, own analysis.

As can be seen, the segregation has been maintained at the level of 40% through all observed years. So, 40% of men or women need to move to other occupations to achieve a uniform distribution of workers of both sex in the labor force. This is relatively higher than the measures founded in the past, when the segregation index in 1994-1996 was 32,4% (Ogloblin, 1999) and in 2000-2002 was 36,5% (Ogloblin, 2005b).

Features of Russian case

To compare Russian figures with the rest of the world, I took several developed and developing countries and calculated the index of dissimilarity in their labor markets, using the most recent International Labour Organization database. I took some European countries (Italy, Germany, France, United Kingdom), United States, China, and India together with Russia to see the difference.



Graph 4. Index of Dissimilarity in the World, 2008-2019

Source: ILOSTAT database, 2020; own analysis.

Using the RLMS and International Labor Organization gives slightly different values of D for Russia, but the difference is not significant, given that official statistics and real answers of respondents may differ due to different samples. In any case, the dissimilarity index ranges from 35-40%. As it can be seen, Russia has the highest index of dissimilarity among selected countries, while China and India have the lowest D. However, all other countries have almost the same D of level 32-33%. This is important as sex segregation in occupations leads to inflexible and thus inefficient labor markets and the economy as a whole (Anker, 1997).

The presence of such high gender segregation compared to other countries can be possibly explained by the institutional basis from the Soviet Union, where the gender division of industries was documented and enshrined in the labor law, which was used by all employers to make decisions on hiring and promoting workers (Ogloblin, 1999). Women were officially regarded as a “specific labor force” (Posadskaya, 2003), and were not given the right to participate in productive industries, as this contradicted their “anatomical-physiological peculiarities” and “female moral-ethical temperament” (Lapidus, 1993). That is, women were assigned to their main role - childbearing, which cannot be connected and combined with complex, specific, and other “non-suitable” for maternity work.

At the same time, the level of economic activity in Russia is one of the highest in the world and can be compared to only Scandinavian countries. For example, it was about 75% for women of working age in 2006 (Gimpelson and Kapelyushnikov, 2008). Such an active participation in economy of Russian women is associated with employment models that have survived from post-war times, when women were forced to work in all industries and at factories due to the lack of labor force.

While the actual reason for segregation cannot be determined with certainty, the fact remains that women are actively involved in the economic life of the country and there is a division that can possibly affect the distribution of salaries and gender wage gap. This relationship will be studied in this paper, controlling for various factors.

occupational segregation wage



Chapter 2. Model and Estimation

In this chapter, the description of the samples and panel regression methods used will be presented, together with the process of estimation and results.

Model setup

As a data, I used the representative sample from the 2010-2018 Russian Longitudinal Monitoring Survey (RLMS).

I considered several sub-samples:

Sample 1 (32 335 observations). Initial sample with restrictions of those respondents who gave answers about their:

a) Monthly wage

b) Working hours in a month

c) Education level

d) Marital status

e) Post level

f) Number of children

g) Region

h) Industry

i) Professional and refresher courses study

j) Disability status

Sample 2 (25 907 observations). Respondents who participated for less than 3 years were removed.

Sample 3 (25 446 observations). Top and bottom 1 % of wages were removed to exclude outliers

Sample 4 (4 221 observations). Balanced sample with respondents who participated in each year of a panel.

Each model has two specifications. They differ by controls on hours worked:

a) Number of respondents' working hours in a month.

b) Indicator of part-time contract.

Main dependent variable, Natural Logarithm of Hourly Wage, is calculated as the logarithm of the ratio of the monthly wage on respondent's main job to the number of working hours in the month. Main explanatory variable - Occupational feminization is calculated as the number of women working in a given occupation in a given year divided by number of all workers working in a given occupation in a given year. In other words, this is the share of women in each occupation each year. Model controls is a set of variables, which include interaction term of feminization and gender time, age, number of children, region, working hours or contract type indicator, industry, sex(gender), disability, education level, marital status. Skills and specialized human capital controls are variables on skill level, professional courses and refresher courses completed by respondents. For details and summary statistics see Appendix 1.

The relation between individuals` wages and occupational feminization is analyzed by linear regression model:

where the i and t indicate individual and time, denotes logged hourly wages, is the constant term, is the measure of occupational feminization, is a vector of observable time-varying variables, is a vector of observable time-invariant variables, represents individual-specific time-constant unobservable characteristics and is a random error.

I estimated the parameters using Ordinary Least Squares (OLS), Fixed Effects (FE), First-Difference (FD) and Random Effects (RE) estimators. So, is treated either as fixed effect or as random effect depending on the parameter estimation method used.



Estimation results

In this section, the brief results for each samples and specification are provided. I used Stata 15 to perform the empirical analysis in this paper. Robust standard errors are used because of heteroscedasticity in all samples with an exception of the strongly balanced sample. Hausman (1978) specification test shows the P-value of zero in all samples and specifications, which means that FE prevails over RE. For the presentation of estimators of all the coefficients, together with the coefficients of feminization without skills controls, and the results of statistical tests, see Appendix 2.

Sample 1.

1a. Specification with working hours

VARIABLES	OLS	FE	RE	FD
feminization	0.0936	0.153	0.139	-0.00981
	(0.103)	(0.0915)	(0.0858)	(0.0882)
Feminization*gender	-0.195***	-0.0819	-0.125***	-0.0426
	(0.0310)	(0.0637)	(0.0428)	(0.0647)
Gender	0.402***	-	0.400***	-
	(0.0176)		(0.0247)
Skill controls	Yes	Yes	Yes	Yes
Other controls	Yes	Yes	Yes	Yes
Constant	4.958***	5.135***	5.139***	-
	(0.113)	(0.356)	(0.124)
Observations	32,335	32,335	32,335	19,504
R-squared	0.521	0.376		0.282

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

1b. Specification with contract type

VARIABLES	OLS	FE	RE	FD
fem	0.0159	0.138	0.0912	-0.00314
	(0.106)	(0.0945)	(0.0888)	(0.0939)
Feminization*gender	-0.0742**	-0.00971	-0.0173	0.0534
	(0.0320)	(0.0647)	(0.0430)	(0.0672)
Gender	0.303***	-	0.291***	-
	(0.0181)		(0.0248)
Skill controls	Yes	Yes	Yes	Yes
Other controls	Yes	Yes	Yes	Yes
Constant	4.958***	5.135***	5.139***	-
	(0.113)	(0.356)	(0.124)
Observations	32,335	32,335	32,335	19,504
R-squared	0.483	0.267		0.155

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

As can be seen, for the full sample all models show the insignificant coefficient on occupational feminization for both choices of the variables reflecting the working hours. However, the interaction term of feminization and gender turned to be negative and significant for OLS in the first specification and for OLS and RE in the second specification, which means that for men the effect of feminization is lower than for women. Skill level 1, which is the highest possible, shows positive and significant correlation with wages (except for FD, where the coefficient is insignificant), while skill level 2, medium level, shows the opposite, negative and significant correlation with wages using all estimators, comparing to the lowest skill level 3. In addition, only OLS shows the positive and significant coefficient for professional courses study (0,0317). Nevertheless, as for refresher courses studying, coefficient is positive and significant for all estimation methods (from 0,015-0,02 in FE and FD to 0,08-0,09 in OLS). Gender has positive and significant coefficient (0,3-0,4), which means that men, on average, earn more, compared to women. Hausman(1978) specification test has Chi-square value of about 580-590 for both specifications, so the hypothesis of equal power of FE and RE is rejected.



Sample 2.

2a. Specification with working hours

VARIABLES	OLS	FE	RE	FD
fem	0.0580	0.131	0.106	-0.0627
	(0.110)	(0.0935)	(0.0903)	(0.0883)
Feminization*gender	-0.205***	-0.0887	-0.137***	-0.0295
	(0.0339)	(0.0662)	(0.0494)	(0.0664)
Gender	0.392***	-	0.380***	-
	(0.0192)		(0.0286)
Skill controls	Yes	Yes	Yes	Yes
Other controls	Yes	Yes	Yes	Yes
Constant	5.058***	5.041***	5.316***	-
	(0.116)	(0.385)	(0.122)
Observations	25,907	25,907	25,907	18,221
R-squared	0.528	0.382		0.281

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

2b. Specification with contract type

VARIABLES	OLS	FE	RE	FD
fem	0.00803	0.123	0.0763	-0.0549
	(0.113)	(0.0965)	(0.0933)	(0.0944)
Feminization*gender	-0.0864**	-0.0236	-0.0374	0.0699
	(0.0350)	(0.0670)	(0.0492)	(0.0689)
Gender	0.301***	-	0.287***	-
	(0.0197)		(0.0285)
Skill controls	Yes	Yes	Yes	Yes
Other controls	Yes	Yes	Yes	Yes
Constant	4.316***	4.078***	4.351***	-
	(0.136)	(0.400)	(0.0944)
Observations	25,907	25,907	25,907	18,221
R-squared	0.493	0.273		0.124

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

For the sample with restrictions on respondents who participated in panel three or more times the coefficient of occupational feminization is also insignificant. Interaction term turns to be significant only in OLS with negative coefficient. Gender is positive and significant like in the previous sample. The highest skill level has positive and significant coefficients in both specifications (about 0,1, on average), while the medium - negative and significant using all estimators except FD (about -0,1, on average). Professional courses show positive and significant coefficient using OLS and negative and significant using FE (0,03 and -0,03 respectively), and refresher courses have positive and significant result using all estimators in both specifications. Hausman test for the specification with working hours has value of 538, while for the specification with part time indicator this value is 607, so FE is better than RE.

Sample 3.

3a. Specification with working hours

VARIABLES	OLS	FE	RE	FD
fem	0.237**	0.210**	0.213***	-0.00859
	(0.0991)	(0.0818)	(0.0782)	(0.0817)
Feminization*gender	-0.235***	-0.125**	-0.167***	-0.0808
	(0.0322)	(0.0627)	(0.0466)	(0.0635)
Gender	0.399***	-	0.391***	-
	(0.0183)		(0.0270)
Skill controls	Yes	Yes	Yes	Yes
Other controls	Yes	Yes	Yes	Yes
Constant	5.037***	4.978***	5.320***	-
	(0.112)	(0.319)	(0.110)
Observations	25,446	25,446	25,446	17,780
R-squared	0.540	0.418		0.348

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

3b. Specification with contract type

VARIABLES	OLS	FE	RE	FD
fem	0.172*	0.195**	0.177**	-0.00590
	(0.104)	(0.0872)	(0.0835)	(0.0890)
Feminization*gender	-0.107***	-0.0448	-0.0574	0.0319
	(0.0335)	(0.0640)	(0.0468)	(0.0664)
Gender	0.300***	-	0.290***	-
	(0.0189)		(0.0271)
Skill controls	Yes	Yes	Yes	Yes
Other controls	Yes	Yes	Yes	Yes
Constant	4.236***	3.706***	4.305***	-
	(0.135)	(0.346)	(0.0834)
Observations	25,446	25,446	25,446	17,780
R-squared	0.496	0.293		0.176

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

On the sample with wage outliers removed, it can be seen that coefficient of occupational feminization is positive (about 0.2) and significant for all estimators apart from the FD estimator. Since Hausman test shows that Fixed Effects is more appropriate model to choose, we can rely on its result about positive relation between feminization and hourly wage (0.195), significant on 5% significance level. Interaction term of feminization and gender has significant coefficient only using OLS, and gender is again positive and significant. Skill level 1 shows the positive and significant result (from 0,01 to 0,2), comparing to the lowest skill level, while the skill level 2 shows the opposite - negative and significant, comparing to the lowest skill level (about -0,1 on average).

Sample 4.

4a. Specification with working hours

VARIABLES	OLS	FE	RE	FD
fem	0.549**	0.672***	0.636***	0.00854
	(0.265)	(0.240)	(0.233)	(0.224)
Feminization*gender	-0.116	-0.171	-0.184	0.162
	(0.0875)	(0.183)	(0.149)	(0.172)
Gender	0.424***	-	0.513***	-
	(0.0494)		(0.0915)
Skill controls	Yes	Yes	Yes	Yes
Other controls	Yes	Yes	Yes	Yes
Constant	5.228***	5.785***	5.267***	-
	(0.205)	(1.057)	(0.219)
Observations	3,729	3,729	3,729	3,303
R-squared	0.561	0.466		0.347

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

4b. Specification with contract type

VARIABLES	OLS	FE	RE	FD
fem	0.440	0.602**	0.546**	-0.0760
	(0.285)	(0.244)	(0.241)	(0.247)
Feminization*gender	0.0418	-0.196	-0.128	0.0850
	(0.0910)	(0.171)	(0.135)	(0.177)
Gender	0.310***	-	0.447***	-
	(0.0506)		(0.0848)
Skill controls	Yes	Yes	Yes	Yes
Other controls	Yes	Yes	Yes	Yes
Constant	4.149***	4.184***	4.113***	-
	(0.202)	(1.101)	(0.227)
Observations	3,729	3,729	3,729	3,303
R-squared	0.523	0.355		0.169

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

For the strongly balanced panel, where all the respondents participated in each year of the panel, the results are almost the same as for previous sample - positive (about 0.6) and significant coefficient for all models except for the First-Difference. As Hausman test again shows that Fixed Effects model prevails over Random Effects, it can be said that the correlation between feminization and hourly wages is positive and significant on 1% significance level for the specification on number of working hours and on 5% for the specification on contract type (0,672 and 0,6, respectively). Gender, as in all samples, has positive and significant result indicating advantage of men over women in hourly wages. Interaction term of feminization and gender is insignificant using both specifications. The highest skill level is positive and significant only using OLS and RE (0,1 on average), and the medium skill level is negative and significant only in OLS. Finished professional courses have positive and significant coefficient using OLS (0,08-0,09), refresher courses positive and significant in OLS and RE for the specification with working hours and positive and significant only in RE for the specification with contract type.

Overall, the positive relation can be established for the last two samples, controlling for skills and investment in human capital. In other samples, the relation is not significant at all. So, the hypothesis about negative correlation of feminization and hourly wages stated above is rejected. These results contradicts the Perales`s (2013) conclusion about devaluation and discrimination of female professions, which means that despite of high gender segregation in Russia, directly the percent of women in industry has no effect or has even positive effect on wages and, consequently, lowers the gender wage gap. To see the exact figures on the shares which occupational feminization can hold in gender wage differentials, the wage decomposition technique will be explained and used in the next section.



Chapter 3. Contribution to the Gender Wage Gap

In this chapter, the procedure of calculating the share which occupational sex-segregation take in the gender pay gap will be presented and results for all samples and specifications will be shown.

Description of an approach

For the calculation I will use Oaxaca-Blinder-Neumark gender wage gap decomposition method described by Ogloblin (1999, 2005) for similar works. This method was proposed by Oaxaca (1973) and Blinder (1973) and has the following steps:

1. Gender wage gap can be presented as:

,

where and are mean wages of mean and women respectively, and are the vectors of mean productivity related characteristics of men and women, and are the coefficients of OLS regression for men and women.

2. Further, it can be decomposed as:

or

,

where on the right-hand sides the first component is the gap that explained by different characteristics, and the second is the unexplained part of gap with unobserved characteristics that is usually interpreted as discrimination (Ogloblin, 1999).

Cotton (1988) and Neumark(1988) indicated the limitation of this approach and improved it so in non-discriminatory settings neither male nor female wage structure prevails in the model, like it is in previous version presented above. As a result, the gender wage gap decomposition technique is given as:

,

where is the estimated non-discriminatory wage structure which was showed by Oaxaca and Ransom (1994) to be the coefficient from the pooled regression. On the right-hand side the first term is productivity differential component, and the second and the third terms are the estimates of the male wage advantage and female male disadvantage, respectively.

So, the contribution of occupational feminization in the gender wage gap will be calculated as:

where and are averages of occupational feminization of men and women respectively, is the coefficient of occupational feminization of pooled regression, and are logarithms of average hourly wages of men and women respectively. Hourly wages are chosen because the independent variable in all my regressions is the logged hourly wage.

Calculation

The contribution was calculated for all samples and specifications despite the significance of insignificance of coefficients of regressions. I did not include the interaction term of feminization and gender while estimating the coefficients in this section, because the model used and presented above provides for the same effect of feminization on men and women. Numbers for significant coefficients indicated. For the detailed tables and figures, see Appendix 3.

Samples	OLS	FE	RE	FD
Sample 1
a	0%	-12%	-8%	2%
b	2%	-14%	-9%	-3%
Sample 2
a	5%	-11%	-5%	8%
b	8%	-14%	-7%	1%
Sample 3
a	-16%	-20%**	-18%*	6%
b	-16%	-23%**	-20%*	-1%
Sample 4
a	-66%*	-80%***	-74%**	-14%
b	-64%*	-68%**	-66%**	4%

Significance: *** p<0.01, ** p<0.05, * p<0.1

Overall, it can be concluded that occupational feminization lowers the gender wage gap, and, actually, is not a component of it, if judged on significant results. This contradicts the findings of Perales (2013) on Britain and of Ogloblin (1999, 2005a) and Roshchin (2003) on Russia. However, in the works for Russia the dummies on industries were used instead of percent of women, so the actual comparison is impossible to make. If controlling for skills and specialized human capital, the occupational sex-segregation seems to has a negative effect on gender wage differentials, which means that the gap in earnings in “male” and “female” industries arises from differences in qualifications and investments in its development, depending on the prestige and difficulty of profession. So, the hypothesis stated previously is rejected and the opposite relationship is proven.



Conclusi...