• No results found

Is there a gender wage gap for people with tertiary education in the Netherlands?

N/A
N/A
Protected

Academic year: 2021

Share "Is there a gender wage gap for people with tertiary education in the Netherlands?"

Copied!
34
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

Amsterdam School of Economics

Bachelor Economics and Business

Specialization Economics

Is there a gender wage gap for people with

tertiary education in the Netherlands?

Academic year 2017 – 2018

Faculty of Economics and Business

Desislava Zheleva, 10846743

(2)

Statement of Originality

This document is written by Student Desislava Zheleva, who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Table of Contents Abstract: ... 4

 

1.

 

Introduction ... 5

 

2.

 

Literature review ... 6

 

3.

 

Theoretical framework: ... 10

 

4.

 

Methodology ... 12

 

5. Data ... 14

 

6. Econometric results ... 16

 

6.1  Assumptions  ...  16

 

6.2  Hypothesis:  ...  19

 

6.3  Outcomes  and  interpretation:  ...  20

 

7.Limitations and Future Research ... 22

 

8.Conclusion ... 23

 

Bibliography ... 25

 

10.Appendices ... 30

 

10.1  Appendix  A:  Summary  statistics  ...  30

 

10.2  Appendix  B:  Testing  multicollinearity  –  outcome  from  correlation  table  ..  31

 

10.3  Appendix  C:  Regression  analyses  ...  32

 

Table  C1:  Regression  analysis  on  the  main  variables  ...  32

 

Table  C2:  Regression  analysis  on  the  main  variables  and  FemaleAge  ...  33

 

Table  C3:  Regression  analysis  on  the  main  variables  and  FemaleDegree  ...  33

 

Table  C4:  Regression  analysis  on  the  main  variables  and  FemaleDutch  ...  34

 

(4)

Abstract:

Gender inequality has been a long researched issue. While there are many factors affecting the income differential between men and women, after an extensive literature review, it has been hypothesized that one of the most important factors that have an impact is the level of schooling. Hence, this paper examines the question whether there is a difference in earnings between men and women with tertiary education in the Netherlands. Carrying out a multiple regression analysis based on a sample of 47,689,000 observations has yielded the result that highly educated women in the Netherlands earn less than men with comparable characteristics. However, more research needs to be conducted, including more control variables, in order to get a more comprehensive view of what affects the gender wage gap.

(5)

1. Introduction

The topic of gender pay inequality has been studied for a number of decades. Gender pay inequality (also known as gender pay/wage/salary gap) is referred to as the difference in earnings between men and women (Council of the European Union, 2010). However, until today, there is still no clear answer to the question of what constitutes the gender wage gap, how it is affected and how it has changed over time (Blau & Kahn, 2017). The latest evidence by the National Bureau of Economic Research (NBER) suggests that the gender gap has been steadily declining for economically advanced countries, such as the Netherlands (Blau & Kahn, 2017). Nevertheless, according to the World Economics Forum (2016), the Dutch gender wage gap for the financial and science sector has increased from 23.9% in 2010 to 29.1% in 2016. While Blau and Kahn (2017) suggest that human capital variables, such as education, explain little, or nothing, of the gender salary gap, the findings in the 2016 Global Gender Gap Report, by the World Economic Forum, imply that there is indeed gender inequality for people with tertiary education.

Tertiary education provides advanced level of skills and training, which are crucial for teachers, doctors, engineers, (social) scientists, etc. This level of education refers to universities, colleges and institutes (The World Bank, 2002). In the Netherlands, tertiary education includes all Universities (WO) and Colleges/Universities of Applied Science (HBO) (De Graaf & Wolbers, 2003).

Up until now, there has been no focus on the trend in the Dutch gender wage gap, for highly educated people. Previously, it has been researched if women-oriented jobs, such as nursing, pay less than male-dominated jobs (De Ruijter, Van Doorne-Huiskes, & Schippers, 2003). Another paper, published by Bakker, Tijdent and Winkels (1999), discusses how schooling, differences in jobs and differences in match between qualifications and jobs affect the gender wage gap in the Netherlands. Therefore, the purpose of this paper is to examine the question whether there is a gender wage gap in the Netherlands for people with tertiary education. In other words, it will be tested whether having a degree benefits women in the Dutch labour market.

According to Blau and Kahn (2017), women labour participation has been steadily increasing. On one hand, this could be explained by the increase of the number of part-time jobs in the Netherlands. On the other hand, it could be explained by the fact that women have been working on obtaining higher degrees and have increased their commitment to work (Blau & Kahn, 2017). Taking this into account, it

(6)

can be expected that a decline in the gender wage gap between Dutch men and women would be found.

In current academic literature, researchers often isolate groups of factors, which affect the gender wage gap (e.g. background), in order to apply their research methods (Weichselbaumer & Winter-Ebmer, 2005). Nonetheless, according to Weichselbaumer and Winter-Ebmer (2005), such data restrictions usually have the most significant impact on the gender salary gap. Furthermore, omission of data can lead to biased and inaccurate results, which means that data selection is more prominent than the choice of method of research (Weichselbaumer & Winter-Ebmer, 2005). Therefore, the aim of this paper is to analyze if there is a gender wage gap in the Netherlands for people with tertiary education, controlling not only for human capital variables, but also for age and background.

The structure of this paper is as follows: first, a literature review section is going to discuss current opinions on what constitutes the Dutch gender pay gap. Subsequently, a section is going to be dedicated to the theoretical framework, where the Dutch educational system will be explained in more detail, as well as why tertiary education is an important factor for measuring gender wage gap. Next, the research method and collection of data will be discussed, followed by a review of the findings. Finally, limitations of the research will be considered, as well as future research suggestions, and a conclusion will be drawn.

2. Literature review

 

This section discusses the factors that affect the gender wage gap, according to current academic literature.

To begin with, Blau and Kahn (2000) find that gender specific factors account for 33% of the gender pay gap, which is why they discuss these factors in more depth. According to them women have lower levels of labour market experience. This is so, because women expect to spend a greater time at home, doing housework, or taking care of their children (Weichselbaumer & Winter-Ebmer, 2005). This anticipation of a reduced working life discourages women from investing in tertiary education, which leads to smaller contribution to human capital, and thus lower earnings (Becker, 1985). At the same time, employers often may hesitate hiring less educated women for jobs which require advanced skills since providing on-the-job training for such candidates is costly and firms are concerned that they might not get a full return on investment (Blau & Kahn, 2000). For the Netherlands, Van der Meer

(7)

(2008) finds that women labour market participation has risen by 25% since 1985 (from 30% to 55%). He explains that this increase is due to the fact that the number of part-time jobs has multiplied over that period. Moreover, just like Blau and Kahn (2000), Van der Meer (2008) finds that labour supply of women is more elastic, due to the fact that women adjust their labour market behavior to changes in wages.

Another widely discussed factor is labour market discrimination, which is analyzed in the Blinder-Oaxaca decomposition. The Blinder-Oaxaca decomposition splits the wage differential between men and women into two components. The first one includes wage inequality due to differences in human capital variables, such as education. Meanwhile, the second part, which is referred to as the unexplained part, measures discrimination (Jann, 2008). Most of the researchers use this analytical tool when studying the gender wage gap (Neumark, Bank & Van Nort, 1996; Blau & Kahn, 2000; Livanos & Núñez, 2012). In their paper, Neumark et al. (1996), perform an experiment, where men and women with comparable qualifications apply for the same job in a chain of high-priced restaurants. The researchers found that women had 40% lower chances of getting an interview for the position, and for those who got interviewed, the probability of getting the job was 50% lower, compared to men. At the same time, another experiment by Goldin and Rouse (2000) investigated the impact of blind auditions by symphony orchestras. They found that when their identity is obscure, women have a higher probability of taking the job vacancy. This switch from standard to blind auditions led to a nearly 46% increase in the number of women that play in the most prestigious symphony orchestras today (Goldin & Rouse, 2000).

Nevertheless, Blau and Kahn (2000) argue that any analytical approach that relies on an unexplained part, residual, may omit important independent variables that are crucial for obtaining quality results. Furthermore, according to Van der Meer (2008), institutional changes in the Netherlands have increased the level of labour market competitiveness, which has made it harder for discriminating employers to continue existing in the market. Thus, the unexplained part of the Blinder-Oaxaca decomposition is not significant for the Netherlands and wage inequality could be explained solely by differences in productivity (Van der Meer, 2008)

A third much-discussed variable is occupation and industry. In their paper, Blau and Kahn (2000) suggest that when occupation and industry are taken into consideration, the explained part of the gender wage gap is 62% (compared to 33% if only gender specific factors are considered). The authors also point out that the advancement of technology has increased the demand for white-collar workers. This means that due to technological changes, women have been more favored for job

(8)

positions that require computer skills, rather than blue-collar jobs that require physical strength (Berman, Bound, & Griliches, 1994). However, when they examine the scholarly world, Berman et al. (1994) find that there is a glass ceiling hindering women’s occupational achievements. The researchers find that only 16.2 % of the professors are female and that women take only 3-5 % of the senior manager positions in the Fortune 1000 companies. The latest evidence by Eurostat (2015) supports the argument that there is a glass ceiling for white-collar workers. Their statistics (see Table 1 below) show that the wage gap is more narrow for jobs that require physical strength: 17% in electricity supply, 2.2% in water supply, 13 % in construction; compared to 21.8 % in business economy, 18.4 % in information and communication, 29.1% in the financial sector and 23.9% in the scientific sector.

Table 1: Gender pay gap statistics for Europe

Note. Adapted from “Gender pay gap statistics” by Eurostat, 2015, Retrieved from:

http://ec.europa.eu/eurostat/statistics-explained/index.php/Gender_pay_gap_statistics

Another important variable is immigration, especially since immigrants constitute 20.7% of the population in the Netherlands (CBS, 2017). In their paper, Euwals, Dagevos, Gijsberts and Roodenburg (2007), discuss the labour market position of foreigners in the Netherlands and Germany. They find that immigration features, such as language proficiency and country of schooling, have an effect on earnings. Nevertheless, in their analysis, Euwals et al. (2007) discover that only 7% of the men and 3% of the women, who migrate to the Netherlands, have obtained a tertiary level of education. Therefore, since level of schooling and language proficiency are important for one’s earnings in the Netherlands, it is essential that

(9)

immigration is considered as a variable in the analysis (Euwals, Dagevos, Gijsberts, & Roodenburg, 2007).

Last but not least, it has been confirmed that most of the gender discrepancy in Europe is due to differences in the years and level of schooling (Tekgüç, Eryar, & Cindoğlu, 2017). In their paper, De Graaf and Wolbers (2003) study the gender differentiation in the choice of education in the Netherlands. They find that children whose family is highly educated tend to follow the steps of their parents and choose a tertiary level of education. Apart from family background, De Graaf and Wolbers (2003) find other factors that affect children’s preferences for education. For instance, girls are likely to follow a study that can later help them find part-time jobs, since such jobs can be easily disrupted when they want to start a family. Otherwise, according to Galor and Weil (1996), women experience difficulty combining family and career. Consequently, in households with high level of education, there are fewer children. Also, female students rarely enroll in a science, economics or engineering university majors. Instead, women find themselves in the fields of education, behavioral studies, health, art and law (De Graaf & Wolbers, 2003). The following table (Table 2), taken from their research, shows the fraction of male and female students, attending each field of study. Every one of the numbers represents how many people from the sample are studying that particular major. For instance, 32.6% of the boys attending college (HBO) are studying engineering, compared to only 7.1% of the female college students.

Table 2: Fraction of male and female students in every field of study

Field of study Agri-culture Educa- tion Beha- viour & Society Eco- nomics Health Art, Lang.& Culture Law& Public Order Nat. Sciences Engi-neering HBO Male 5.4 12.2 4.5 34.8 6.2 4.1 0.2 - 32.6 Female 2.6 24.6 22.3 20.9 18.6 3.8 0.1 - 7.1 WO Male 0.7 - 12.4 23.7 9.3 4.9 11.7 11.7 25.7 Female 2.5 - 25.8 8.5 16.8 13.3 15.8 8.5 8.8

Note. Adapted from “The effects of social background, sex, and ability on the transition to tertiary education in the Netherlands.” by M.De Graaf and H.Wolbers, 2003, The Netherlands Journal of Social Science, 39(3), pp.183-184.

(10)

In spite of these findings, latest statistics by the World Economic Forum (2016) show that women have started to invest more in tertiary education, with 82% of the females attending HBO or WO in 2016, compared to 69% in 2013 (World Economic Forum, 2013). The number of men obtaining an academic degree has also increased, but it is still lower than the number of women – 75% in 2016, compared to 62% in 2013. According to Blau and Kahn (2000) women’s dedication to obtaining advanced skills and their commitment to the labour force leads to a decline in the wage discrimination against females.

As it can be seen, there are a lot of factors that affect the gender wage gap and most of them could be linked to the level of women’s education – women are required to learn more advanced skills, in order to get a white-collar job; females with higher education have more labour market experience; immigrants with a degree have better paid jobs. Nevertheless, many findings are contradictory when looking at various literature and statistics. While most researchers suggest that the gender wage gap is closing (see Blau & Kahn,1997; O’Neil & Polachek, 1993; Wellinton, 1993), data from Eurostat and World Economics Forum suggest the opposite.

Based on the literature review conducted above, it can be concluded that there are many factors that affect the gender wage gap, such as gender specific factors, immigration or education. Nevertheless, it seems that most of these determinants can related to the level of schooling a person has acquired. Thus, it would be interesting to research whether there is a gender pay gap in the Netherlands for people with tertiary education.

3. Theoretical framework:

 

This section discusses the Dutch educational system in more detail, the meaning of tertiary education, and explains why the latter is a sufficient variable to study the gender wage gap.

First of all, a good understanding of the Dutch educational system is essential for this paper. As stated before, tertiary education refers to higher level of schooling, including private/public universities, colleges, technical institutes and vocational schools (The World Bank, 2017). In the Netherlands, tertiary education has a binary system, meaning that there are two types of higher schooling. On one hand, there are research-oriented studies (wetenschappelijk onderwijs, WO). On the other hand, there are higher professional schools (hoger beroepsonderwijs, HBO). While the former one takes place in research universities (universiteiten), the latter takes place

(11)

in universities of applied sciences/colleges (hogescholen). Research university education usually constitutes of three parts – a Bachelor program of 3 years, a Masters program of 1-3 years and a doctor’s degree of approximately 4 years. Similarly, universities of applied sciences consist of 2 cycles – a Bachelor program of 4 years and a Master program of 1-2 years (Nuffic, 2015). The table below (Table 3) shows the number of universities and colleges in the Netherlands, as well as how many students are attending each of the two types of schooling.

Table 3: Number of colleges and universities in the Netherlands

Type of schooling

College (HBO) University (WO) Number of schools in the country1 37 17 Number of students 445 0002 266 1373

Since HBO and WO are considered the highest levels of schooling in the Netherlands, this paper is going to count them both as tertiary education.

Secondly, as previously mentioned, gender wage gap is known as the difference in earnings between men and women (Council of the European Union, 2010). Some researchers, like Tekgüç, Eryar and Cindoğlu (2017), insist that tertiary education is the most essential element of wage employment, and thus, the most essential element of the gender wage gap. In their paper, Tekgüç et al. (2017) discuss the importance of tertiary education as an explanatory variable for gender pay inequality. According to them, sampling observations based on educational level creates more homogeneous groups with respect to unobserved characteristics, such as skills. This group specification mitigates the level of biasedness in the gender wage gap estimates (Kunze, 2008). Besides, if there is no disaggregation by educational level, Tekgüç et al. (2017) find that the unexplained part of the gender wage gap is larger. In their discussion about the relevance of higher education for analyzing the gender wage gap, Tekgüç et al. (2017) point out several facts that illustrate why tertiary education is important for women. First, the probability for university-educated women to engage in wage employment is higher, than for uneducated females (see also Tansel, 1994). Second, women who have obtained an

                                                                                                               

1    Retrieved  from  

http://info.studielink.nl/nl/studenten/overzichtonderwijsinstellingen/Pages/universiteiten.aspx   2    Retrieved  from  https://www.onderwijsincijfers.nl/kengetallen/hoger-­‐beroepsonderwijs/deelnemers-­‐ hbo/ingeschreven-­‐hoger-­‐beroepsonderwijs  

3    Retrieved  from  https://www.onderwijsincijfers.nl/kengetallen/wetenschappelijk-­‐ onderwijs/deelnemerswo/ingeschrevenen-­‐in-­‐het-­‐wetenschappelijk-­‐onderwijs

 

(12)

academic degree have more control over their personal lives. Female graduates are more independent, especially when it comes to housework and fertility within marriage (Cindoğlu & Toktaş, 2002). Last, but not least, better-educated women have the freedom to change jobs due to the amount of skills they possess, while less educated women are limited to certain, low paid positions (Royalty, 1998).

Other scholars also review the significance of tertiary education when exploring the gender wage gap. In his book, Becker (1975) discusses the human capital theory, according to which, people are rational individuals and as such, they take decisions that maximize their return on investment. According to Becker (1975), education is the most profitable investment one could make, since it is directly related to people’s earnings. The more educated and trained workers are, the higher their marginal productivity is and, therefore, the more they earn (Becker, 1975). Livanos and Núñez (2012) support this claim in their paper, where they insist that an academic degree is a signal of superior abilities that not every person possesses. In their paper, Livanos and Núñez (2012) study what is the wage differential between men and women depending on whether they have obtained a college/university degree. The researchers claim that women who obtained an academic degree are unlikely to quit their jobs when they start a family, or they would prefer to have fewer children, so that they can focus on their career.

Overall, based on previous academic research, it can be concluded that tertiary education is a reasonable factor to consider when analyzing the gender wage gap.

4. Methodology

This section is aimed at explaining what methods have been adopted for the analysis, why have they been adopted, and why not alternative methods.

First, it is essential to be able to distinguish the types of research that can be implemented. There are two main categories of research that scholars apply. The first kind is qualitative research. With qualitative research, academicians usually work with non-numerical data and focus on the meaning and interpretation of previous literature (Crossman, 2017). Most often, they firstly gather information, and then they form a hypothesis. In opposition, there is quantitative research, which relies on numerical data and helps to examine connections between variables. Within the field of quantitative research, there are four categories of analysis: descriptive,

(13)

correlational, quasi-experimental and experimental. Table 4 below shows the difference between them.

Table 4: Types of quantitative research

Quantitative Research

Descriptive Correlational Quasi-experimental Experimental Describes current status of a variable and provides information about a phenomenon. Determines the relationship between two or more variables, using statistical methods and data. This method

examines trends, but does not prove causes. Determines the causal effect between an independent and dependent variable(s). The researcher cannot manipulate data, and he/she does not randomly assign groups. Determines causal effect between variables. The difference between experimental and quasi-experimental research is that in experimental research, the experimenter can manipulate the independent variable and can randomly assign subject to different groups. Note: Reprinted from Key Elements of a Research Proposal. Quantitative design, Retrieved from

https://www.bcps.org/offices/lis/researchcourse/develop_quantitative.html

For this paper, a multiple regression analysis is going to be performed, which falls in the category of quasi-experimental research. The multiple regression model is an extension of the simple regression model which examines the relationship between one dependent variable (Y) and one independent variable (X). Thus, the multiple regression method examines the same relationship, but while also controlling for other independent variables, in order to avoid problems of omitted variable bias (Stock & Watson, 2011).

There are several reasons that make this method suitable for the analysis in this paper. First, as pointed by Stock and Watson (2011), the multiple regression model fights the problem of omitted variable bias. Also, Shalev (2007) said that multiple regression model fits analyses, in which marginal effect is to be estimated.

(14)

Since this paper examines the gender effect on salary, for people with tertiary education, it can be inferred that the model is suitable.

Alternatively, as previously mentioned, researchers often use the Blinder-Oaxaca decomposition. As it was stated before, this is a method, in which the wage differential equation is divided into two parts – one that describes the effect of human capital variables; and another, unexplained part, which is related to discrimination (Jann, 2008). Nevertheless, there are several reasons that justify not using the above-mentioned method in this paper. In her article, Gosse (2002) criticizes the Blinder-Oaxaca decomposition. She states that one of the major problems with this method is the index number problem. This means that the end results are affected by what reference group is chosen - male or female (Gosse, 2002). Moreover, Jones (1983) argues that the Blinder-Oaxaca decomposition results are not interpretable, especially when using dummy variables, and as it would be seen later on, the regression in this paper uses binary variables. To prove his point, Jones (1983) performed an experiment, where he used data on the income gap in Australia in 1976. He calculated the same gender gap in two ways – once using a dummy variable to represent “age left school”, and once using a continuous variable. He found that the results, and especially the unexplained part, differed under the two methods. Therefore, it can be concluded that the Blinder-Oaxaca decomposition is not a suitable model for this particular paper.

5. Data

This section describes the data used in the regression, as well as the motivation behind using these variables.

In this paper, the time frame chosen for the analysis is 2011-2015. The reason that this period is used is that this is the longest time frame, for which the Dutch Bureau of Statistics (CBS- Centraal Bureau voor de Statistiek) provides data. Nevertheless, the sample is sufficiently large, consisting of more than 40 million observations in total, for all the years.

Before proceeding with the data description, it is important to define what the model looks like. The baseline regression equation in this paper is as follows:

(15)

𝑖𝑛𝑐𝑜𝑚𝑒! = 𝛽!+ 𝛽!∗ 𝑎𝑔𝑒!"+ 𝛽!∗ 𝑓𝑒𝑚𝑎𝑙𝑒 + 𝛽!∗ 𝑑𝑒𝑔𝑟𝑒𝑒 + 𝛽!∗ 𝑑𝑢𝑡𝑐ℎ   +  𝛽!∗ 𝑛𝑜𝑛𝑤𝑒𝑠𝑡𝑒𝑟𝑛 +   𝛽!∗ (!2012) + 𝛽!∗ (!2013) +   𝛽!∗ (_2014)   +  𝛽!∗ (_2015) + 𝛽!"∗ 𝐹𝑒𝑚𝑎𝑙𝑒𝐷𝑒𝑔𝑟𝑒𝑒 + 𝜀!"

 

where t = 2011,2012, 2013, 2014 or 2015; and i refers to the ith person from the sample.

All data has been retrieved from CBS, and in the next few paragraphs, each variable will be explained.

To begin with, the dependent variable of this regression is income. In their statistics, CBS define income as the yearly personal income from labour and self-employment. In the numbers provided by the Dutch Bureau of Statistics, premiums for income insurance are deducted from the personal income. In this paper, the personal income of people within the working force, who have obtained at least a Bachelor’s degree, has been recorded for the period between 2011 and 2015. For simplicity, all numbers were divided by 1000. For instance, if there are sixteen people, who earned 40.5 euros in year t, this means that, in reality, there were 16,000 people, whose yearly income was 40,500 euros.

Next, the variable age constitutes of three categories. Category one includes all working people in the age group 15-24. Category two includes workers who fall in the age group 25-44, and finally, Category three includes people aged 45-64. Ages below 15 and above 65 have been omitted, since people from these age groups are not part of the labour force (Dutch Labour Force Survey (LFS), n.d.). Age is followed by the variable female, which is a dummy variable, indicating whether a person is male or female. The variable female takes value of one if the worker is a woman, and zero otherwise.

Following is the dummy variable degree. It takes value of one if a worker has obtained at least a Bachelor’s degree (or a Master’s, or a Doctorate level) in a Dutch research university (WO), or a university of applied sciences (HBO), and zero if a lower level of education was acquired.

Two background variables were included in the regression – dutch and nonwestern. The variable western, which represents immigrants from Europe (excluding Turkey), North America, Oceania, Indonesia and Japan, was omitted in order to avoid a dummy variable trap. The variable dutch is a binary regressor that equals one for people with Dutch background, whose parents were also born and raised in the Netherlands, or zero otherwise. Meanwhile, the variable nonwestern takes value of one for people with non-western migration background, and value of zero in every other case.

(16)

Also, four dummy variables for years were added: (_2012), (_2013), (_2014) and (_2015). That way, it could be examined the yearly change in earnings for an individual. The year 2011 was used as a base variable, against which the outcomes from other years were compared. Thus, 2011 was omitted from the regression, in order to avoid problems with the dummy variable trap. Summary statistics of the main explanatory variables can be found in Appendix A.

After that, four interaction terms were added, since according to Stock and Watson (2011), interaction between regressors may affect the dependent variable. The first interaction variable is FemaleAge, which examines whether the age of a woman has an effect on her income. This term can take values of zero (if male), one, two or three, depending on the age group, to which a woman belongs. The second one, FemaleDegree, is an interaction term, equal to the product of the variables female and degree. Its purpose is to examine whether there is a relation between women who have obtained an academic degree and their income. This term is equal to one if the worker is a female who has obtained at least a Bachelor level of education, or it equals zero if this is not the case. Finally, the variables FemaleDutch and FemaleNW are used to observe whether the background of a woman has an effect on her income. The variable FemaleDutch equals one if a woman comes from the Netherlands, and zero otherwise, while the variable FemaleNW equals one if a woman has a non-western background, and zero otherwise. Several variables, that could have been significant for the analysis, were omitted due to the unavailability of information.

6. Econometric results

6.1  Assumptions  

As mentioned before, the econometric model used in this paper is multiple regression. Nevertheless, before proceeding with the analysis, there is a number of assumptions, known as the Gauss-Markov assumptions, that need to be satisfied, in order for the model to be efficient (Stock & Watson, 2011).

The first one of them is that the conditional mean of the residual, given the independent variable, has a mean of zero (Stock & Watson, 2011). However, Stock and Watson (2011) explain that normality of residuals is often rejected when the model is applied in real life. This statement is confirmed by the results of a Shapiro-Wilk test, shown in Table 4 below. Thus, it can be expected that some of the

(17)

estimates of the model will not be significant. Such a violation of this assumption can occur when there is an omission of explanatory variables (Stock & Watson, 2011).

The second assumption that needs to hold is that the predicted and the explanatory variables are independently and identically distributed (i.i.d.). This assumption holds automatically if the supposition of no autocorrelation in the residual can be proven (Stock & Watson, 2011). For this reason, a Durbin-Watson test was performed in Stata. The resulting d-statistic is 0.0011864, which leads to a conclusion that no serial correlation is present in this model (0.0011864<1.675).

Next, there should be no large outliers. This means that when a scatterplot is requested, there should be a straight line that goes through the observations. This assumption does not hold for the raw data, however, when some of the observations are removed, there is linearity present.

Furthermore, homoscedasticity of the error terms should be present, meaning that the variance of the error term, given the independent variables, should be constant (Stock & Watson, 2011). To test this, a Breusch-Pagan test was performed in Stata. The resulting F-statistic is 9604.33, with p-value=0.000. This means that the error terms are heteroskedastic, and thus their variance is not constant. To resolve this issue, heteroskedasticity-robust standard errors are going to be used in the actual analysis. Table 4 below summarizes these outcomes:

(18)

Table 4: Diagnostics table – testing the assumptions

Diagnostics table

Assumption Test Hypothesis Result Conclusion

Normality of residuals

Shapiro-Wilk W

test Normality Prob>z = 0.000

Residuals are not normal

Doornik-Hansen test for multivariate normality Normality Prob>chi2 = 0.0000 Residuals are not normal Dependent and independent variables are i.i.d. + no autocorrelation of the residuals Durbin-Watson No auto-correlation d-statistic( 11, 47689) = 0.0011864 No autocorrelation, so observations are i.i.d. Homo-skedasticity Breusch-Pagan test Homo-skedasticity Prob>F = 0.000 There is hetero-skedasticity present Finally, it was checked whether there is multicollinearity present in the model. Perfect multicollinearity arises when one of the independent variables is a perfect linear combination of another one (Stock & Watson, 2011). If this occurs, there could be a problem with the computation of the OLS (Ordinary Least Squares) estimator (Stock & Watson, 2011). The table in Appendix B shows the resulting correlation table. From the outcomes, it can be seen that only the variables dutch and degree are highly correlated ( 𝜌!"#$!,!"#$"" = 0.8567) . Nevertheless, the correlation coefficient is not exactly equal -1 or 1, thus implying that there is no perfect multicollinearity, so there is no problem running the regression. In order to prove this point, the same regression was performed once including the variable dutch, and once omitting it. Output 1 below shows that the results stay the same, whether or not the variable was excluded from the model. Moreover, since the sample used for this analysis is large (N=47,689,000), there is no reason to infer that multicollinearity is a problem (Stock & Watson, 2011).

(19)

Output 1: Comparison between the regression model including the variable dutch (1) and the regression model excluding the variable dutch (2)

 

(1)  

(2)  

VARIABLES  

income  

income  

 

 

 

age  

11.70***  

11.70***  

 

(0.0464)  

(0.0464)  

female  

-­‐15.34***  

-­‐15.34***  

 

(0.0712)  

(0.0712)  

degree  

3.152***  

3.153***  

 

(0.135)  

(0.0709)  

dutch  

0.00143  

 

 

(0.131)  

 

nonwestern  

0.904***  

0.902***  

 

(0.193)  

(0.156)  

_2012  

0.233***  

0.233***  

 

(0.0892)  

(0.0892)  

_2013  

0.664***  

0.664***  

 

(0.0891)  

(0.0891)  

_2014  

1.176***  

1.176***  

 

(0.0894)  

(0.0894)  

_2015  

1.439***  

1.439***  

 

(0.0892)  

(0.0892)  

FemaleDegree  

-­‐1.440***  

-­‐1.440***  

 

(0.111)  

(0.111)  

Constant  

11.33***  

11.33***  

 

(0.153)  

(0.153)  

 

 

 

Observations  

47,689  

47,689  

R-­‐squared  

0.791  

0.791  

Robust  standard  errors  in  parentheses  

***  p<0.01,  **  p<0.05,  *  p<0.1  

6.2  Hypothesis:  

 

The following hypotheses will be tested:

𝐻!:  𝛽!"= 0  ;  𝐻!:  𝛽!"< 0

The null hypothesis will test the effect that a degree has on women’s income. If not rejected, meaning that the coefficient of the variable FemaleDegree is not significant, then there is no gender wage gap. The alternative hypothesis will test if the coefficient of the variable female is negative. If it is, then this means that there is a gender wage gap between men and women with tertiary education in the Netherlands.

(20)

6.3  Outcomes  and  interpretation:    

Before proceeding to final results and interpretation, it would be interesting to see the effect that the interaction terms have on the regression. Tables C1-C5, from Appendix C, show the analyses that were conducted. All of the analyses were made based on the assumption of ceteris paribus – holding everything else constant.

First, a multiple regression was performed only on the main variables – age, female, degree, dutch, nonwestern, _2012, _2013, _2014 and _2015 (see Table C1, Appendix C). That way, the effect of every single variable can be determined and later compared to the results when interaction terms are added. All of the tests were performed, using significance level of 1% (𝛼=1%).

To begin with, it can be seen from the regression output that the variable age is statistically significant at 1% (p<0.01). Also, its coefficient is positive and equal to 11.718. Thus, it can be inferred with 99% confidence that people from a higher age group earn 11,718 euros more than comparable, younger individuals. Next, it was confirmed that there is, indeed, a gender wage gap in the Netherlands. The outcomes show that women earn 15,784 euros less per year, compared to men (p<0.01). Following, it can be concluded with 99% confidence that having a degree can indeed increase someone’s earning (p<0.01). On average, graduates tend to earn 2,378 euros more than what their less educated colleagues receive. Meanwhile, both interaction terms showed to have a positive effect on earnings. Finally, the coefficients of the time variables suggest that a person’s income increases yearly. For instance, a Dutch, educated, woman, who belongs to age group 1, earned 10,134.89 in 2012. These earnings are with 234.22 euros more, than what a woman with comparable characteristics earned in 2011.

Nevertheless, this regression has an important limitation: the effects of age, degree and background are the same, no matter the gender. There is no reason to conclude that this is the case. Therefore, the next paragraphs introduce the regression analyses, in which interaction terms were introduced.

First, the variable FemaleAge was included, and income was regressed on the variables: age, female, degree, dutch, nonwestern, _2012, _2013, _2014, _2015 and FemaleAge (see Table C2, Appendix C). An interesting case occurred - the coefficient of the variable female changed from negative to positive. This change suggests that there might be, what they call in academic literature, suppression effects (Tu, Gunnell, & Gilthorpe, 2008). MacKinnon, Krull and Lockwood (2000) suggest that a way to test if such effects are actually present in the model is to

(21)

the previous section, there is no presence of multicollinearity, so it can be inferred that there are no suppression effects. As for the interpretation, it seems from the results that older women earn less. However, it can be calculated that this negative effect cancels out even if a woman belongs to the first age category. For instance, in 2013, a woman that is considered to fall in age group 1, who is Dutch and has at least a Bachelor level of education would earn 16,826 euros (=1000 ∗ (0.5096 + 16.619 ∗ 1 + 6.9835 + 1.8427 + 0.3516 + 0.6729 − 10.1533 ∗ 1)). At the same time, a man with the same characteristics would be paid 19,995 euros (=1000 ∗ (0.5096 + 16.619 ∗ 1 +1.8427 + 0.3516 + 0.6729))

Next, the variable FemaleDegree was added, in order to examine whether women and men who have obtained tertiary level of education earn the same amounts (see Table C3, Appendix C). In this regression, the coefficient of the interaction variable turned out negative, suggesting that the effect of having a degree is different between men and women. Moreover, the results are statistically significant at 1% (p<0.01), which indicates that it could be concluded with 99% confidence that highly educated women earn around 16,775 euros less per year, than men with the same characteristics do. Thus, it can be concluded that there is a gender wage gap for people with tertiary education. This result also aligns with the theory and raw data, used for this research.

Next, the regression analysis, including the variable FemaleDutch, showed that this interaction term was statistically significant for the study (see Table C4, Appendix C). It turned out that Dutch women earn less than men and foreign women. This finding is also supported by the raw data, where it can be seen that in this period, women with western background acquired a higher income. Statistics show that in the previous few years more children were born in the Netherlands, which could have induced Dutch women to quit their jobs, or start working part-time. At the same time, there was a higher inflow of immigrants. These facts are represented in the graph below (Figure 1):

(22)

Figure 1: Net births, Net migration and population growth

Note. Adapted from “Trends in the Netherlands, 2016” by CBS. Retrieved from

https://www.cbs.nl/en-gb/publication/2016/26/trends-in-the-netherlands-2016

The results regarding the regression model that includes the interaction term FemaleNW can be seen in Appendix C. Although this variable was not statistically significant (p>0.01), it can be deduced that coming from a non-western background benefited women. What might have caused this term to be insignificant is omitted variable bias 4.

Overall, the outcomes in each of the regressions followed the same pattern. Given the 1% significance level, it was concluded that women who have a degree earn less than men with comparable characteristics. Thus, it can be inferred that there is a gender wage gap for people with tertiary education. However, there are other variables that need to be taken into account, in order to make this research complete. The next section is going to discuss this matter.

7.Limitations and Future Research

Although the research reached its aims, some limitations need to be discussed. As pointed multiple times in the previous section, some of the outcomes were not as expected. The most logical explanation for this is omitted variable bias. This phenomenon occurs when one or more (significant) regressors are left out from the model, when they should have been included. The omission of important variables may result in imprecise estimation of other components (Barreto & Howland, 2005). Therefore, the outcomes might be have been different if variables such as: occupation, (number of) children, family status, industry and part-time job

                                                                                                               

(23)

were added. Other factors that might affect the gender wage gap are in which school a person has obtained their academic degree, or which study a person followed. Also, the data provided on age was in only three groups. Results could have been different if age would be a continuous variable, taking value from 15 to 64. Nevertheless, there is no data source that provides such specific information per individual.

Another limitation is the time frame. The most amount of information that could be gathered was over a period of only five years. Such a short interval is not enough to observe a certain trend in the wage gap. Therefore, adding more regressors to the equation, and testing for a longer period of time could give more accurate results.

While performing a multiple regression analysis could give insights on the gender wage gap in the Netherlands, more research should be done. There are more comprehensive models that can be conducted, which might derive more meaningful results. Some are the Blinder-Oaxaca decomposition and an Instrument Variable (IV) regression (Weichselbaumer & Winter-Ebmer, 2005). According to Weichselbaumer and Winte-Ebmer (2015), every method yields different results. Therefore, the most important thing is to obtain as specific and as much data as possible. Comparing outcomes from different methods, using a more complete set of data, can give a more extensive and accurate picture of the Dutch gender wage gap and its trends.

8.Conclusion

The aim of this paper was to examine the relation between tertiary education and the difference in earnings between men and women. The alternative hypothesis was that there is a difference in earnings between male and female graduates in the Netherlands. Based on the regression analysis that was performed, there is enough statistical evidence to infer that there is a gender pay gap between highly educated people in the Netherlands. These findings are in line with theory and expectations.

For this paper, a multiple regression model was applied. The main advantage of this method is the easiness to interpret results. Nevertheless, this statistical technique may be too simple to cover all aspects of the topic. In addition, the major shortcoming of this paper is the unavailability of data. Important factors, such as children, occupation, part-time job or kind of study had to be omitted.

In spite of the limitations mentioned before, this paper has contributed to the topic by analyzing the Dutch gender salary gap while controlling both for tertiary education and background, and resulting in a conclusion that an academic degree is

(24)

negatively associated with higher earnings for women. Furthermore, it has emphasized the fact that in future research, either more regressors should be included, or a longer time frame should be studied in order to reach to a more precise conclusion.

(25)

Bibliography

Bakker,  B.,  Tijdens,  K.,  &  Winkels,  J.  (1999).  Explaining  gender  wage  differences.  

 

Netherlands  Official  Statistics  ,  14,  36-­‐41.  

 

Barreto,  H.,  &  Howland,  F.  (2005).  Introductory  econometrics:  using  Monte  Carlo

 

 simulation  with  Microsoft  excel.  Cambridge  University  Press.  

 

Becker,  G.  S.  (1975).  Human  Capital:  A  Theoretical  and  Empirical  Analysis,  with  

 

Special  Reference  to  Education,  Second  Edition.  The  University  of  Chicago  

 

press.  2(3),  13-­‐44.  

 

Becker,  G.  S.  (1985).  Human  Capital,  Effort,  and  the  Sexual  Division  of  Labor.  

 

Journal  of  labour  economics  ,  3  (1),  33-­‐58.  

 

Berman,  E.,  Bound,  J.,  &  Griliches,  Z.  (1994).  Changes  in  the  Demand  of  Skilled  

 

Labor  Within  U.S.  Manufacturing  Industries:  Evidence  From  the  Annual  

 

Survey  of  Manufacturing.  The  Quarterly  Journal  of  Economics  ,  109(2),  

 

367-­‐397.  

 

Blau,  F.  D.,  &  Kahn,  F.  M.  (1997).  Swimming  upstream:  Trends  in  the  gender  wage  

 

differential  in  the  1980s.  Journal  of  labor  Economics  ,  15  (1),  1-­‐42.  

 

Blau,  F.  D.,  &  Kahn,  L.  M.  (2000).  Gender  differences  in  pay  .  Journal  of  Economic

 

 Perspectives,  14(4)  ,  75-­‐99.  

 

Blau,  F.  D.,  &  Kahn,  L.  M.  (2017).  The  Gender  Wage  Gap:  Extent,  Trends,  and  

 

Explanations.  Journal  of  Economic  Literature  ,  55  (3),  789-­‐865.  

 

CBS.  (2016).    Trends  in  the  Netherlands,  2016  [Graph].  CBS.  Available  from  

 

https://www.cbs.nl/en-­‐gb/publication/2016/26/trends-­‐in-­‐the-­‐

 

netherlands-­‐2016  

 

CBS.  (2017).  Population,  households  and  population  dynamics;  from  1988.    

 

Retrieved  from  

 

http://statline.cbs.nl/StatWeb/publication/?DM=SLEN&PA=37556ENG&

 

D1=0-­‐44,53-­‐60&D2=1,11,21,31,41,51,61,71,81,91,101&LA=EN&VW=T  

 

(26)

Cindoğlu,  D.,  &  Toktaş,  Ş.  (2002).  Empowerment  and  Resistance  Strategies  of  

 

Working  Women  in  Turkey:  The  Case  of  1960—70  Graduates  of  the  Girls'  

 

Institutes.  European  Journal  of  Women's  Studies  ,  9  (1),  31-­‐48.  

 

Council  of  the  European  Union.  (2010).  The  gender  pay  gap  in  the  Member  States

 

 of  the  European  Union:  quantitative  and  qualitative  indicators.  Brussels:  

 

Belgian  Presidency  Report.  

 

Crossman,  A.  (2017,  May  17).  An  overview  of  Qualitative  Research  Methods.  

 

Retrieved  from  https://www.thoughtco.com/qualitative-­‐research-­‐

 

methods-­‐3026555  

 

De  Graaf,  P.  M.,  &  Wolbers,  M.  H.  (2003).  The  effects  of  social  background,  sex,  

 

and  ability  on  the  transition  to  tertiary  education  in  the  Netherlands.  The

 

 Netherlands'  Journal  of  Social  Science  ,  39  (3),  172-­‐201.  

 

De  Ruijter,  J.,  Van  Doorne-­‐Huiskes,  A.,  &  Schippers,  J.  (2003).  Size  and  Causes  of  

 

the  Occupational  Gender  Wage-­‐gap  in  the  Netherlands.  European  

 

Sociological  Review  ,  19  (4),  345-­‐360.  

 

Dutch  Labour  Force  Survey  (LFS).  (n.d.).  Retrieved  from  Centraal  Bureau  voor  de  

 

Statistiek  (CBS):  

 

https://www.bcps.org/offices/lis/researchcourse/develop_quantitative.

 

html  

 

Eurostat.  (2015).  Gender  pay  gap  statistics  [Table].  Eurostat.  Retrieved  from    

http://ec.europa.eu/eurostat/statistics-­‐

explained/index.php/Gender_pay_gap_statistics    

 

Euwals,  R.,  Dagevos,  J.,  Gijsberts,  M.,  &  Roodenburg,  H.  (2007).  The  Labour  

 

Market  Position  of  Turkish  Immigrants  in  Germany  and  the  

 

Netherlands:Reason  for  Migration,  Naturalisation  and  Language  

 

Proficiency.  Centraal  Planbureau  ,  1-­‐42.  

 

(27)

 

Goldin,  C.,  &  Rouse,  C.  (2000).  Orchestrating  Impartiality:  the  Impact  of  ‘Blind’  

 

Auditions  on  Female  Musicians.  American  Economic  Review  ,  90  (4),  715-­‐

 

741.  

 

 

Goldschmidt,  J.  (2012).  Protecting  Equality  as  a  Human  Right  in  the  Netherlands:  

 

From  Specialised  Equality  Body  to  Human  Rights  Institute.  The  Equal  

 

Rights  Review  ,  8,  32-­‐49.  

 

Gosse,  A.  M.  (2002).  The  Gender  Pay  Gap  in  the  New  Zealand  Public  Service.  State

 

 Services  Commission  ,  1-­‐49.  

 

Jann,  B.  (2008).  A  Stata  implementation  of  the  Blinder-­‐Oaxaca  decomposition.  

 

Stata  Journal  ,  8  (4),  453-­‐479.  

 

Jones,  F.  L.  (1983).  On  Decomposing  the  Wage  Gap:  A  Critical  Comment  on  

 

Blinder's  Method.  The  Journal  of  Human  Resources  ,  18  (1),  126-­‐130.  

 

Key  Elements  of  a  Research  Proposal.  (2017).    Quantitative  Design  [Table].    BCPS.  

 

Retrieved  from  

 

https://www.bcps.org/offices/lis/researchcourse/develop_quantitative.

 

html  

 

Kunze,  A.  (2008).  Gender  wage  gap  studies:  consistency  and  decomposition  .  

 

Empirical  Economics  ,  35  (1),  63-­‐76.  

 

Livanos,  I.,  &  Núñez,  I.  (2012).  The  effect  of  higher  education  on  the  gender  wage

 

 gap.  International  Journal  of  Education  Economics  and  Development  ,  3  

 

(1),  33-­‐47.  

 

MacKinnon,  D.,  Krull,  J.,  &  Lockwood,  C.  (2000).  Equivalence  of  the  Mediation,  

 

Confounding  and  Suppression  Effect.  Prevention  science  ,  1  (4),  173-­‐181.  

 

Neumark,  D.,  Bank,  R.  J.,  &  Van  Nort,  K.  D.  (1996).  Sex  Discrimination  in  

 

Restaurant  Hiring:  An  Audit  Study.  The  Quarterly  Journal  of  Economics  ,

 

 111  (3),  915-­‐941.  

(28)

Nuffic.  (2015).  The  Dutch  education  system  described.  Retrieved  from  Nuffic:  

 

https://www.nuffic.nl/en/study-­‐and-­‐work-­‐in-­‐holland/dutch-­‐education-­‐

 

system  

Onderwijs  in  Cijfers.  (2017).  Ingeschrevenen  hoger  beroepsonderwijs  [Table  3].  

 

Retrieved  from  https://www.onderwijsincijfers.nl/kengetallen/hoger-­‐

 

beroepsonderwijs/deelnemers-­‐hbo/ingeschreven-­‐hoger-­‐

 

beroepsonderwijs  

 

Onderwijs  in  Cijfers.  (2017).  Ingeschrevenen  in  het  wetenschappelijk  onderwijs  

 

[Table  3].  Retrieved  from  

 

https://www.onderwijsincijfers.nl/kengetallen/wetenschappelijk-­‐

 

onderwijs/deelnemerswo/ingeschrevenen-­‐in-­‐het-­‐wetenschappelijk-­‐

 

onderwijs  

 

O'Neil,  J.,  &  Polachek,  S.  (1993).  Why  the  gender  gap  in  wages  narrowed  in  the  

 

1980s.  Journal  of  Labor  Economics  ,  11  (1),  205-­‐228.  

 

Royalty,  A.  B.  (1998).  Job-­‐to-­‐job  and  job-­‐to-­‐non-­‐employment  turnover  by  gender

 

 and  education  level.  Journal  of  Labor  Economics  ,  16  (2),  392-­‐433.  

 

Shalev,  M.  (2007).  Limits  and  alternatives  to  multiple  regression  in  comparative  

 

research.  Capitalisms  Compared  ,  261-­‐308.  

 

Stock,  H.  J.,  &  Watson,  M.  W.  (2011).  Introduction  to  Econometrics.  Boston:  

 

 Addison-­‐Wesley.  

 

Studielink.  (2017).  Overzicht  hogescholen  &  universiteiten  [Table  3].  Retrieved  

 

from  

 

http://info.studielink.nl/nl/studenten/overzichtonderwijsinstellingen/P

 

ages/universiteiten.aspx  

 

Tansel,  A.  (1994).  Wage  employment,  earnings  and  returns  to  schooling  for  men  

 

and  women  in  Turkey.  Economics  of  Education  Review  ,  13  (4),  305-­‐320.  

 

 

Tekgüc,  H.,  Eryar,  D.,  &  Cindoğlu,  D.  (2017).  Women’s  Tertiary  Education  Masks

 

   the  Gender  Wage  Gap  in  Turkey.  Journal  of  Labor  Research  ,  38  (3),  360-­‐

 

   386.  

 

(29)

 

The  World  Bank.  (2017,  October  5).  Higher  education.  Retrieved  from  

 

http://www.worldbank.org/en/topic/tertiaryeducation  

 

Tu,  Y.-­‐K.,  Gunnell,  D.,  &  Gilthorpe,  M.  S.  (2008).  Simpson's  Paradox,  Lord's  

 

Paradox,  and  Suppression  Effects  are  the  same  phenomenon  –  the  

 

reversal  paradox.  Emerging  Themes  in  Epidemiology  ,  5  (1),  2.  

 

Van  der  Meer,  P.  H.  (2008).  Is  the  gender  wage  gap  declining  in  the  Netherlands?

 

 Applied  Economics  ,  40  (2),  149-­‐160.  

 

Weichselbaumer,  D.,  &  Winter-­‐Ebmer,  R.  (2005).  A  meta-­‐analysis  of  the  

 

international  gender  wage  gap.  Journal  of  economic  surveys  ,  19  (3),  479-­‐

 

511.  

 

Wellington,  A.  J.  (1993).  Changes  in  the  male/female  wage  gap,  1976-­‐85.  Journal  

 

of  Human  Resources  ,  383-­‐411.  

 

World  Economic  Forum.  (2013).  The  Global  Gender  Gap  Report.  World  Economic

 

 Forum,  Geneva,  Switzerland.  

 

World  Economic  Forum.  (2016).  The  Global  Gender  Gap  Report.  World  Economic

 

 Forum,  Geneva,  Switzerland.  

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

(30)

10.Appendices

10.1  Appendix  A:  Summary  statistics  

 

 

_2015 47,689 .1992074 .3994085 0 1 _2014 47,689 .1992703 .3994559 0 1 _2013 47,689 .1996687 .3997555 0 1 _2012 47,689 .2005704 .4004313 0 1 nonwestern 47,689 .0247646 .1554086 0 1 dutch 47,689 .2494915 .4327233 0 1 degree 47,689 .3113087 .4630336 0 1 female 47,689 .4789574 .4995623 0 1 age 47,689 2.242844 .7362595 1 3 income 47,689 31.71334 13.45634 9.3 50.7 Variable Obs Mean Std. Dev. Min Max

(31)

10.2  Appendix  B:  Testing  multicollinearity  –  outcome  from  correlation  table  

 

 

_2015 -0.2498 -0.2491 -0.2488 1.0000 _2014 -0.2499 -0.2492 1.0000 _2013 -0.2502 1.0000 _2012 1.0000 _2012 _2013 _2014 _2015 _2015 0.0298 0.0036 0.0019 0.0150 0.0153 0.0050 _2014 0.0185 0.0016 0.0010 0.0064 0.0108 0.0009 _2013 -0.0009 0.0006 0.0000 0.0001 0.0051 0.0007 _2012 -0.0187 -0.0018 -0.0008 -0.0078 -0.0176 0.0021 nonwestern 0.0295 -0.0001 0.0020 0.2370 -0.0916 1.0000 dutch 0.1330 0.1007 0.0072 0.8567 1.0000 degree 0.1404 0.1015 0.0208 1.0000 female -0.5993 -0.0236 1.0000 age 0.6638 1.0000 income 1.0000 income age female degree dutch nonwes~n

(32)

10.3  Appendix  C:  Regression  analyses    

Table  C1:  Regression  analysis  on  the  main  variables  

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

_cons 11.50772 .1513112 76.05 0.000 11.21115 11.8043 _2015 1.43393 .0893025 16.06 0.000 1.258895 1.608964 _2014 1.170869 .0895444 13.08 0.000 .9953608 1.346377 _2013 .6628502 .0892237 7.43 0.000 .4879706 .8377298 _2012 .2342234 .0893115 2.62 0.009 .0591717 .4092751 nonwestern .9843454 .1931492 5.10 0.000 .6057703 1.36292 dutch .0804017 .1314934 0.61 0.541 -.1773271 .3381306 degree 2.377822 .1262281 18.84 0.000 2.130413 2.625231 female -15.78356 .0572517 -275.69 0.000 -15.89577 -15.67134 age 11.71829 .0468289 250.24 0.000 11.62651 11.81008 income Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = 6.1571 R-squared = 0.7907 Prob > F = 0.0000 F(9, 47679) = 57670.71

(33)

Table  C2:  Regression  analysis  on  the  main  variables  and  FemaleAge  

 

 

 

Table  C3:  Regression  analysis  on  the  main  variables  and  FemaleDegree  

 

_cons .509643 .1374134 3.71 0.000 .2403108 .7789751 FemaleAge -10.15328 .0628895 -161.45 0.000 -10.27655 -10.03002 _2015 1.46014 .0710841 20.54 0.000 1.320814 1.599466 _2014 1.188404 .0714696 16.63 0.000 1.048323 1.328486 _2013 .6729599 .0718131 9.37 0.000 .5322053 .8137145 _2012 .2423181 .0725772 3.34 0.001 .1000658 .3845703 nonwestern 1.250336 .1728092 7.24 0.000 .9116279 1.589045 dutch .3515688 .1174123 2.99 0.003 .1214391 .5816984 degree 1.842685 .1128556 16.33 0.000 1.621486 2.063883 female 6.983515 .1686505 41.41 0.000 6.652957 7.314072 age 16.61931 .0482541 344.41 0.000 16.52474 16.71389 income Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = 4.8976 R-squared = 0.8676 Prob > F = 0.0000 F(10, 47678) = 44445.00

Linear regression Number of obs = 47,689

_cons 11.33185 .1527632 74.18 0.000 11.03243 11.63127 FemaleDegree -1.439836 .1107812 -13.00 0.000 -1.656969 -1.222703 _2015 1.438668 .0891606 16.14 0.000 1.263912 1.613424 _2014 1.175751 .089394 13.15 0.000 1.000538 1.350964 _2013 .6639788 .0890845 7.45 0.000 .489372 .8385856 _2012 .2333334 .0891704 2.62 0.009 .0585582 .4081085 nonwestern .903685 .1931696 4.68 0.000 .5250701 1.2823 dutch .001428 .130675 0.01 0.991 -.2546967 .2575527 degree 3.151946 .1347642 23.39 0.000 2.887806 3.416085 female -15.33585 .0712186 -215.33 0.000 -15.47544 -15.19626 age 11.70126 .0464004 252.18 0.000 11.61031 11.7922 income Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = 6.1481 R-squared = 0.7913 Prob > F = 0.0000 F(10, 47678) = 51422.74

Referenties

GERELATEERDE DOCUMENTEN

In a further, independent aspect there is provided a method of producing electronic circuitry comprising providing a cir cuitboard, producing at least one aperture in

The use of microscopic modeling has highlighted the importance of a rarely considered microscopic parameter, namely, the detachment rate from filament [ +] ends, in determining

The model results reveal the existence of stable equilibrium states with more than one inlet open, and the number of inlets depends on the tidal range and basin width (section 3)..

The international development Master Meter (MM) project in Jakarta aims to improve the water security of low-income households by providing access to drinking

In conclusion, this thesis presented an interdisciplinary insight on the representation of women in politics through media. As already stated in the Introduction, this work

Based on the results from this thesis the answer is: “Yes, to some extent.” The reason for this answer is that for the age group 25-64 years gender norms are a significant

Supportive strategies could also help the shipping sector to comply with regulation that is set by other (external) parties to the chain such as national governments, the EU or

Natural gas through a pipe at typical transport conditions behaves Newtonian and can be regarded as incompressible, because the Mach number of gas transport is