The labor force participation rate of female immigrants in the Netherlands

The labor force participation rate of female immigrants in

the Netherlands

BSc Thesis

Name: Lily Waller

Student number: 10329765

Track: Economics

Field: Labor Economics

Number of credits thesis: 12

Supervisor: Sabina Albrecht

Date: Monday 26 June 2017

Words: 5747

2 Statement of Originality

This document is written by Student Lily Waller, 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 Abstract

In the majority of the countries, the average female labor force participation rate is lower than the male labor force participation rate. The last few decades, there have been enormous changes in the female labor force participation rate. Religious effects on economic behavior have been ignored by economists in the past. Therefore, the research question in this thesis is: What are the differences in female labor force participation of different ethnical groups in the Netherlands? The Enquete van beroepsbevolking (1987-2012) will provide all the relevant information about labor force participation. The probit model will be used to assess whether there is a difference between Turkish and Moroccan female and natives. Turkish and

Moroccan females appeared to work less than native females in the Netherlands in every year, except for 1990 and 1995. In conclusion, the most recent data shows that if a female

participant is Turks, the probability of participating in the labor force will decrease with 10,91 percent and if a female participant is Moroccans, it will decrease with 12,72 percent.

Consequently, one can conclude that Turkish and Moroccan female have a lower female labor force participation rate in comparison with Dutch native females.

4

Table of contents

1. INTRODUCTION ... 5

2. THEORETICAL FRAMEWORK ... 7

2.1 CHOICE THEORY BEHIND LABOR FORCE PARTICIPATION ... 8

3. LITERATURE REVIEW: FACTORS THAT INFLUENCE FEMALE LABOR FORCE PARTICIPATION ... 11 3.1 FERTILITY ... 11 3.2 EDUCATION ... 12 3.3 CULTURE AND IMMIGRANTS IN THE NETHERLANDS ... 13 3.4 WAGE ... 15 3.5 CHANGES IN THE NETHERLANDS ... 15 4. HYPOTHESIS ... 15 5. RESEARCH METHOD ... 16 5.1 ASSUMPTIONS ... 18 5.2 SIGNIFICANCE LEVEL ... 18 6. RESULTS ... 18 6.1 FIRST REGRESSION ... 19 6.2 SECOND REGRESSION ... 20 6.3 MARGINAL EFFECTS ... 22

7. CONCLUSION AND DISCUSSION ... 22

8. REFERENCES ... 24

APPENDIX A: RESULTS ... 26

APPENDIX B: TEST FOR MULTICOLLINEARITY ... 30

5

1. Introduction

In the majority of the countries, the average female labor force participation rate (LFPR) is lower than the male labor force participation rate. Although the female labor force

participation rate is rising, it is far from equal. The last few decades, there have been enormous changes in the female labor force participation rate. Between 1940 and 1972 the female labor force participation rate grew from 15 percent to 41.5 percent (Leibowitz, 1975). Before 1940 women usually stopped working when they got married. However, after the second world war it was increasingly normal for women to keep working after they got married.

Graph 1: Labor force in the Netherlands (source: OECD)

As is apparent in the graph, the total labor force participation rate has been increasing since 1975. The increase in female LFPR has been the most important factor of the rise in the aggregate participation rate (Burniaux, Duval & Jaumotte, 2004).

Approximately half of the Netherlands’ human endowment consists of women

(Psacharopoulos & Tzannatos, 1989). Nonetheless, the female labor force participation rate (LFPR) is lower than the male labor force participation rate. The differences between male and female participation rates is a commonly known phenomenon. However, the differences

0 10 20 30 40 50 60 70 80 90 1975 1980 1985 1990 1995 2000 2005 2010 2015

Labor force in the Netherlands

6 become fascinating when we look at the differences of female labor force participation rate between different ethnical origins. The Netherlands is known as a multicultural society.

Ethnic group % of total males % of total females

Native 84,41 83,18 Western immigrants 7,10 7,78 Turkish 1,41 1,45 Maroccan 1,37 1,28 Surinamers 1,73 1,97 Antilles 0,84 0,89 Other 3,03 3,32 Unknown 0,12 0,12

Table 1: Ethnic groups in the Netherlands (source: EBB 2012, CBS)

According to The Enquete van beroepsbevolking (2012) 15,59 percent of all male- and 16,82 percent of all female Dutch residents are foreigners. As one can see in table 1, there are many different ethnic groups with different cultures in the Netherlands. These cultures, with

different religions, have different female LFPR. Religious effects on economic behavior have been ignored by economists in the past (Heineck, 2004). Nowadays, economists acknowledge that religion influences economic behavior. For example, almost 25 percent of Protestants believe that it is the man’s responsibility to be the breadwinner and the women must look after the children (Lehrer, 1995). Exclusivist Protestant groups are the least equal considering the male/female representation and atheist are the most equal. Originally, Roman Catholics were the most conservative. However, they have become less traditional with respect to gender equity in comparison to the Protestant groups. The traditional point of view still has a negative effect on female LFPR.

Increasing the female LFPR is important because of multiple reasons. In the majority of the EOCD countries the preferred female LFPR is higher than the actual female LFPR due to market failures (Jaumotte, 2003). Hence, extending the female participation could lead to a higher level of welfare in the Netherlands. To increase the female LFPR to the optimal level of welfare, it is important to examine the effect of being religious on the female LFPR. Thereby, the population of developed countries are is greying. Because a larger share of the population is in the group of 65+, the ageing of the population will decrease the total labor supply, which causes negative effects for public finances. Therefore, increasing the female

7 LFPR could be a solution for this problem. In addition, after the second world war the high birth and death rates were decreasing because of the industrialization and the total population was growing. This is called the ‘demographic transition’. Because of this there was a higher proportion of the total population in the labor force and the female labor supply increased substantially which had a significant positive impact on economic growth (Bloom, Canning, Fink & Finlay, 2009)

In conclusion, to increase the female labor force participation it is very important to

investigate the reasons why women choose to work or not. If we know what the reasons are, we can stimulate women to work and hence increase the welfare in the Netherlands. The unequal gender distribution is one of the major reasons why there is a great gap between the ‘modern’ Dutch natives and the immigrants with an Islamic culture (Verloo & Roggeband, 2006). Verloo & Roggeband said that the emancipation of Islamic females is the key to reduce the gap. Religion potentially causes the female labor supply to be lower than preferred. Therefore, because religion and ethnicity are closely related, the research question in my thesis is:

What are the differences in female labor force participation of different ethnical groups in the Netherlands?

2. Theoretical framework

Firstly, it is important to define what the definition is for the labor force. The labor force participation rate is the ratio of the numerator and the denominator (Psacharopoulos & Tzannatos, 1989). The numerator refers to the labor force, which consist of all people who work and all people who seek work. The denominator refers to all people who can work. Hence, all children, elderly and disabled people are excluded. The labor force (LF) is given by

LF = E + U (= employed + unemployed) The labor force participation rate (LFPR) is given by

LFPR = LF/P

8

2.1 Choice theory behind labor force participation

The choice for women whether to spend their time on work or leisure can be explained by the neoclassical model of labor-leisure choice. This model works with utility and indifference curves (Borjas, 2000, p. 27). Utility increases from two kinds of consumption, consumption of goods (C) and the number of hours of leisure (L). Hence, the utility function is:

U = f(C,L)

Both C and L have a positive relation with utility. Which means that both kinds of consumption increase the happiness of a person.

Graph 2: indifference curves (source: Borjas, 2000, p.28)

All points on the same indifference curve yield the same utility. There are four aspects that are important for indifference curves. First, the slopes of the indifference curves are negative. A person gets utility from both goods. Hence, if a person wants more leisure and stays on the same curve, it must give up a certain amount of goods. Secondly, point Z has a higher utility than point X, considering that point Z has more L and more C. Consequently, the higher the indifference curve, the greater the utility. Thirdly, the curves cannot cross each other. If the curves intersect, this would disaffirm the fact that a person would like both kinds of

consumption. Finally, the shape of the curves is convex.

Another aspect of this model is the budget constraint. L and C is constrained by the available time of a person and the available income. Income is divided into two different types.

Namely, nonlabor income (V) and labor income (wh). This can be summarized in the budget constraint:

9 where w is hourly wage rate and h is the number of hours worked during a certain period of time. The person must divide the total available time (T) between h and L. Substituting T = h + L into the budget constraint forms the budget line:

C = (wT + V) – wL

Graph 3: budget line (Borjas, 2000, p.33)

Point E illustrates the endowment point. Point E tells what a person can spend if she does not work and spends all of her time to leisure. The slope of C = (wT + V) – wL is w. A typical rational person wants to maximize her utility taking the budget line into account. The optimal point is the point where the slope of the budget line (w) is equal to the slope of the

indifference curve (MUl/MUc).

An assumption in this model is that leisure is a normal good (Mincer, 1962). A good is normal if demand increases when income increases (Cooper, 1983). According to Mincer (1962) the model has a positive substitution effect. This means that if real wage increases, the opportunity costs of one hour leisure also increases. Hence, because leisure is a normal good, the number of hours worked will increase if real wage increases. On the other hand, the model has a negative income effect. When real wage increases, income also increases. Higher

income leads to a higher demand for goods and consumption, including leisure as a good. Consequently, this means that an increase in real wage will decrease the number of hours worked.

10 Graph 4: income and substitution effect (source Borjas, 2000, p.38)

If there is a wage increase and non-labor income does not change, the slope of the blue line (w) becomes steeper and the line moves from FE to GE in graph 4. To distinguish the income and substitution effect, a line parallel to FE has to be drawn (DD). The income effect is illustrated by the movement from P to Q. The increase in wage rate increases the hours of leisure from 70 hours to 85 hours and thus decreases the number of hours worked with 15 hours. The substitution effect is illustrated by the movement from Q to R. The increase in wage rate decreases the number of hours leisure from 85 to 75 and thus increases the number of hours worked with 10. If the substitution effect dominates, the total effect of the increase in wage rate is positive (4a). If the income effect dominates, the total effect is negative (4b).

In addition, Mincer (1962) made an extension to the neoclassical model of labor-leisure choice. Women do not choose only between labor and leisure. They allocate their time between labor, leisure and work at home. Work at home also includes child caring and

household chores, which are some of the most frequent factors why women do not participate in the labor force.

11

3. Literature review: factors that influence female labor force

participation

Research about this subject has been conducted in different countries. Previous research determined the main factors that influences the female LFPR.

3.1 Fertility

Neoclassical models predict that children are a normal good (Ahn & Mira, 2002).

Nevertheless, the data of the last few decades disproves this prediction. During the industrial revolution, the income went up and the fertility rate was decreasing. To explain this

contradiction Becker (1965) developed a model for the time allocation decisions for women.

Graph 5: OECD fertility and female participation rate (source: Ahn & Mira, 2002)

As is apparent in graph 5, fertility and female participation rate clearly have a negative correlation. Several researches showed that mothers have lower wages, work less, and have less respected jobs (Lundborg, Plug & Rasmussen, 2014, p.18). Children change their mothers’ labor supply (Bloom et al., 2009). Using abortion legislation as instrumental

variable (IV), Bloom et al. found that each additional child reduces female LFPR 5/10 percent during their fertile years. The concept of ‘demographic dividend’ elucidates the economic benefits that a country can gain if it experiences a decline in fertility (Bloom et al., 2003). This decline increases the proportion of people between age 15 and 65 (the working-age) compared to the total population, which in turn increases the income per capita. This decline in fertility benefits the population because the physical and human capital per capita also

12 increases.The female labor force participation responds positively on the decline in fertility, which in turn further increases the labor supply and income per capita. Becker (1964) argued that the quantity-quality tradeoff models predict that investment in child health and education per child increases, which is another positive effect caused by the decline in fertility.

It is difficult to prove a causal relation between having children and female labor supply. It is possible that women who choose to have children attach less value to labor than women without children. Therefore, Lundborg, Plug & Rasmussen (2014) used in vitro fertilization as instrumental variable. In this way, they found a causal relationship between having children and female labor supply. In this research, they used wage as a measure of labor supply. Fertility has a negative, large and long-lasting effect on female labor supply. The strongest result was in the extensive margin (first born children). In the intensive margin (second born or later) the causal relationship was ambiguous. In the short run they found some negative (but not strong) effect. In the long run there was no significant effect.

3.2 Education

Higher educated women have a higher probability to participate in the labor force (Leibowitz, 1967). This is explained by the human capital theory; higher educated people are more

productive in the labor market. Because of this, the opportunity costs for children increases when a woman is higher educated. Therefore, education and fertility should have a negative correlation.

The last few decades there has been a sharp increase in women’s education (Hotz, Klerkman & Willis, 1997). The change in education participation is an important factor for the change in labor supply side (Vlasblom & Schippers, 2004). However, they also found that female LFPR increased in every educational level for women. Consequently, it is very important to look at social perspectives. Literature also explains that the large increase in women’s education caused an increase in female earnings (Hotz et al., 1997). Consequently, the opportunity costs of having children are also larger, which is the reason why there is a negative correlation between educational attainment and fertility.

13

3.3 Culture and immigrants in the Netherlands

The labor force of immigrants in the Netherlands has been analyzed (CBS, 2000). At the end of the fifties, the demand for lower schooled labor was higher than the supply. Therefore, many Turkish, Moroccan, Italian and Spanish males immigrated into the Netherlands. The majority of these immigrants were factory workers. However, seven out of ten Spanish and six out of ten Italian workers moved back to their home county within ten years. On the other hand, most of the Turkish and Moroccan males stayed in the Netherlands and at the end of the sixties family reunifications began.

Graph 6: share of Turkish and Moroccan female immigration in the Netherlands (source: CBS, 2000)

As from 1973 it was not allowed anymore to legally migrate to the Netherlands. However, the number of immigrants from Turkey and Morocco continued to grow, which was mainly due to the family reunification. This is one of the causes why the number of Turkish and

Moroccan immigrants is still high compared to other ethnicities.

There are many differences between cultures. In some cultures, it is more common for women to work than in other cultures. Not only the way people look at this subject is different, but also the age of having children, educational level and the language is different. One of the biggest obstacles that second-generation migrants face in the Netherlands is the weak education (Crul & Doomernik, 2003). Their Dutch is worse compared to their native peers

14 which causes lower school performance. This arrear already begins at primary school, causing many children to go to lower education at high school.

Table 2: Success rates of high school students in the Netherlands among different ethnicities (source: CBS, 1999)

As apparent in table 2, Turkish and Moroccan children are not only lower educated, but their success rates are also substantially lower compared to native children.

Besides the fact that immigrant females are lower educated, the social perspective of gender role distribution differs from natives, especially in Islamic cultures. Since 2003, female immigrants are an important subject in Dutch politics (Verloo & Roggeband, 2006).

Increasing the female immigrant labor force participation rate is one of the key determinants to improve the integration between the immigrants and natives (Dagevos, 2001). Dagevos sees religious attitudes towards the unequal gender role distribution as one of the major obstacles for the emancipation. The distance between the first-generation Turks and

Moroccans compared to natives is very large. Especially lower educated immigrants are less integrated. In case the immigrant went to school in the Netherlands, the integration increased substantially. However, the share of second generation Turks and Moroccans who are

religious does not decrease and the attitude towards gender role distribution does not change substantially. Hence, this contributes to a lower female labor force participation rate among Muslims (Heineck, 2004).

15

3.4 Wage

Another important factor is wage. According to the standard work-leisure model and the assumption that leisure is a normal good, an increase in real wage makes the opportunity costs for one hour of leisure more expensive (Mincer, 1962). Hence, an increase in real wage should increase the number of hours worked.

3.5 Changes in the Netherlands

Kea Tijdens (2006) analyzed the change of the female labor force participation rate in the Netherlands and the factors contributing to these changes. The labor force participation of women between 15 and 64 doubled between 1947 and 2004. It appears that the contributing factors before 1970 are different from the contributing factors after 1970. Before 1970 the increase was mainly caused by the increase in hourly wages. After 1970 it was primary caused by the increase of higher educated women. In addition, it is remarkable that before 1970 the female labor force participation rate was highly correlated with marriage. Women tended to stop working when they got married. In the seventies, this attitude changed and most women worked until they had their first child. However, almost all women stopped working at that moment. This changed in the eighties, an increasing number of women kept working after they gave birth to their first child. These results primarily applied for higher educated women. However, it is important to note that only 34 percent of the total female labor force in 2003 had a full-time job (Valk & Boelens, 2004). On the other hand, 86 percent of the total male labor force had a full-time job. Especially women above 29 had part-time jobs. This effect is probably caused by the influence of having a first child.

4. Hypothesis

I expect that there will be a difference between the labor force participation rate of

Turkish/Moroccan females and the female labor force participation rate of native females. Possible explanations for this negative effect could be that immigrants are lower educated (Tijdens, 2006) and immigrants have a more traditional perspective towards the gender role distribution (Lehrer, 1995)

𝐻_":b_$%&,()* = b_$%&,&,* = 0 𝐻_.: b_$%&,()* 𝑎𝑛𝑑/𝑜𝑟 b_$%&,&,* < 0

16

5. Research method

In this thesis, I would like to investigate the differences between the female labor force participation rate in the Netherlands by ethnic origin. I obtain the data from ‘Data Archives and Networked Services’ (DANS), an online data center which reuses datasets. I use the data set from Centraal Bureau Statistiek. The Enquete van beroepsbevolking (1987-2012) will provide all the relevant information about labor force participation. Women over 65 will be removed from the sample because they do not participate in the labor force. Participants are considered native if they have Dutch nationality. People are considered immigrant if they do not have Dutch nationality or have Dutch nationality and are born outside the Netherlands. The last few decades the total female LFPR has increased in the Netherlands. In this thesis, I want to assess whether the female LFPR among different ethnical origins has increased as well. To this purpose, I will use several different regressions for every year from 1990 until 2012 and use the probit model. I choose this model because the dependent variable is binary. The values 0 and 1 cannot be exceeded within the probit model while this is possible within a linear regression model. Firstly, I will look whether there are differences in female labor force between natives and Turkish/Moroccan female immigrants. The purpose of the first equation is to estimate whether there is indeed a difference between the native female LFPR and the Turkish and Moroccan female LFPR in the Netherlands. This can be done in two different ways. The first possibility is to add two interaction terms; 𝛽_$%&()*∗ 𝑓𝑒𝑚 ∗ 𝑡𝑢𝑟𝑘𝑖𝑠ℎ and 𝛽$%&&,*∗ 𝑓𝑒𝑚 ∗ 𝑚𝑜𝑟𝑜𝑐𝑐𝑎𝑛. The second possibility is to exclude all male participants from

the sample. The advantage of the first method is that the power of this regression is higher in comparison with the second method. The advantage of the second method is that all the coefficients are only estimated for women causing the estimations to be more precise. In this research, I choose for the second method because of the large sample size of the EBB. In this regression, the effect between ethnicity and labor force participation is solely tested among women. Consequently, if the 𝛽C(D is significantly different from zero, one can conclude that

Turkish and Moroccan females work less than native females in the Netherlands. The first regression:

Pr(𝑌F=1|𝐴𝑔𝑒F, 𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛F, 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑐ℎ𝑖𝑙𝑑𝑟𝑒𝑛F, 𝑀𝑎𝑟𝑟𝑖𝑎𝑔𝑒F, 𝐸𝑡ℎ𝑛𝑖𝑐F,) = 𝛼 +

𝛽_CP)𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛_F+ 𝛽_Q%*𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑐ℎ𝑖𝑙𝑑𝑟𝑒𝑛_F + 𝛽_&R*𝑀𝑎𝑟𝑟𝑖𝑎𝑔𝑒_F+ 𝛽_C(D𝐸𝑡ℎ𝑛𝑖𝑐_F + 𝛽_ST%𝐴𝑔𝑒_F + 𝛽_ST%UV𝐴𝑔𝑒W

F + 𝑢F

The dependent variable (Y) is employment. Y = 1 if a person is in the labor force and Y = 0 is a person is not in the labor force. All persons who are willing to work, regardless whether

17 they are employed or not, are included in the labor force. According to the definition of CBS, a participant is in the labor force if he or she works (or is willing to work) more than twelve hours per week. For the independent variables, the factors that have the largest influence on LFPR should be included in the regression. These are Age, Education, Number of children, Gender, Marital status and Ethnicity. Education, gender, number of children, marital status and ethnicity are dummy variables. Of course, in the first regression 𝛽$%& is excluded,

because all participants are female. Education takes the value 1 for MAVO/VMBO, 2 for HAVO/VWO/MBO, 3 for HBO and 4 for WO and 0 if the participant only participated in elementary school. Marital status takes the value 1 for divorced/never been married and 0 for married. Ethnicity takes the value 1 for Turkish, 2 for Moroccan and 0 for Native. Female takes the value 1 for female and 0 for male. 𝛽ST%UV𝐴𝑔𝑒W_F is added because age and labor force

participation probably do not have a linear relationship. It probably has an inverted U-shaped relationship since the labor force does not continue increasing at all ages. Because a probit model is not linear, you cannot directly interpret the coefficients.

Subsequently, I will look at the changes of the female labor force participation rate for natives, Turkish and Moroccan in time. The second probit regression is:

Pr(𝑌F=1|𝐴𝑔𝑒F, 𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛F, 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑐ℎ𝑖𝑙𝑑𝑟𝑒𝑛F, 𝑀𝑎𝑟𝑟𝑖𝑎𝑔𝑒F, 𝐸𝑡ℎ𝑛𝑖𝑐F, 𝐺𝑒𝑛𝑑𝑒𝑟) =

𝛼 + 𝛽_CP)𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛_F + 𝛽_Q%*𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑐ℎ𝑖𝑙𝑑𝑟𝑒𝑛_F + 𝛽_YR*𝑀𝑎𝑟𝑟𝑖𝑎𝑔𝑒_F+ 𝛽_C(D𝐸𝑡ℎ𝑛𝑖𝑐_F + 𝛽_$%&𝐹𝑒𝑚𝑎𝑙𝑒_F + 𝛽_ST%𝐴𝑔𝑒_F + 𝛽_ST%UV𝐴𝑔𝑒W

F + 𝑢F

The outcomes of 𝛽$%& will be plotted in a graph to look how the trends are. Firstly, this will

be done for the whole sample. After that, all the participants other than Turkish will be removed from the sample. 𝛽$%&,()* will be plotted in the same graph. The same will be done

for Moroccan. To conclude whether there is indeed a difference between 1990 and 2009 for all three 𝛽_$%&’s, the Wald test of simple and composite linear hypothesis about the parameters will be used.

In a probit model it is not possible to interpret coefficients directly. However, it is interesting to look at how much less likely Moroccan/Turkish women are to work. It is possible to use the ceteris paribus (holding all other regressors constant) effects on changes in the regressors. The average marginal effects of the coefficients 𝛽()* and 𝛽&,* will be calculated. This

18 estimation about how much less likely these women are to work. All male participants will be removed from the sample and the first regression will be used again.

5.1 Assumptions

To use the binary probit model a few assumptions are needed; the dependent variable must be binary, the error terms needs to be independent and normally distributed, the multicollinearity should not be too high, the independent variables must be linear and the dataset must be randomly sampled.

The multicollinearity will be tested with the variance inflator factor (vif). Hereby a rule of thumb is assumed; if the variance inflator factor is larger than ten, the correlation between the independent variables may be too high and further analysis is required (Marquardt, 1980). The interaction terms are naturally highly correlated. Consequently, the high factor for these terms can be ignored. In STATA it is not possible to perform the ‘vif’ command on a probit model. Therefore, this will be done on a normal regression. This will give an accurate result for estimating whether the independent coefficients are highly correlated. To test for the linearity, an interaction term is included. If the interaction term is significant, it must be included in the regression. To ensure that the data set is randomly sampled the CBS uses weight to correct for the unequal draw probabilities. Because of this they make sure that none of the groups are overrepresented.

5.2 Significance level

The most usual significance level for a study is 5 percent, a=0,05. a stands for the probability of a test rejecting the null hypothesis, given that it was true. I will also use this significance level in this thesis.

6. Results

In this chapter the results will be discussed. The datasets from 1990 until 2009 consist of, on average, 189112 participants. A remarkable fact is that the data indeed shows that the population in the Netherlands faces ageing, as mentioned in the introduction. In 1990 the average was 39,21 and in 2009 it was 42,31. A two sample t-test showed that the average age between 1990 and 2009 indeed significantly increased (t=-40,2094, Pr(T<t)=0,000).

19 In general, education appears to have a positive effect on the labor force participation. For all years it applies that the dummy value 4 for ′𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛′ has the highest positive 𝛽%P) and

value 1 the lowest. This means that the higher a participant is educated, the higher the probability of participating in the labor force. All results were significantly different from zero. Children have, as expected, a significantly negative effect on the probability of participating in the labor force. For this coefficient counts that, the higher the value of the dummy value ′𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑐ℎ𝑖𝑙𝑑𝑟𝑒𝑛’, the more negative the coefficient 𝛽Q%* . This means

that, the more children the participants have, the smaller the probability of participating in the labor force. The coefficient 𝛽ST% is positive and significant for every year which illustrates

that the older a person is, the higher the probability that the participant participates in the labor force. However, the coefficient 𝛽ST%UVis negative and significant in all the regressions.

This shows that the relationship between the probability of being in the labor force and age is indeed not linear, but U-shaped. Since 𝛽ST%UV is significant it must be included in the

regression equation. Being an immigrant also has a negative and significant effect on the probability of being in the labor force, regardless of ethnicity. However, in this second regression this effect is for males and females together and in the first only for women.

The assumption of no perfect multicollinearity is not violated. As one can see in table 10 in appendix B, no coefficient has a variance inflator factor greater than ten. Except age and age2 are highly correlated, but this is because they are interaction terms. The linearity assumption is also tested. As apparent in table 11 in appendix C, age2

is negative and significant. Hence, the relationship between age and the labor force is not linear and age2 _{must be included in the} regression.

6.1 First regression

In the first regression all the male participants are removed from the sample. In this way the effects of the independent variables on the probability of females being in the labor force can be estimated. The signs of 𝛽CP), 𝛽Q%*, 𝛽YR*, 𝛽ST% 𝑎𝑛𝑑 𝛽ST%UV do not change compared

to what is mentioned above and they are still significant. As can be seen in table 3 in appendix A, all the 𝛽()* coefficients are significant, except for the 𝛽()* from 1990 and 1995. This

means that, for these years, Turkish female immigrants in the Netherlands have a lower participation in their total workforce in comparison with Dutch native females. In 1992, 1994 and 1996 the ethnicity of the participants was unknown. As apparent in table 4 in appendix A, all the 𝛽&,* coefficients are significant, except for the 𝛽&,* from 1995. This means that, for

20 these years, Moroccan female immigrants in the Netherlands have a lower participation in their total workforce in comparison with Dutch native females. In 1992, 1994 and 1996 the ethnicity of the participants was unknown. Concluding, the probability that Turkish and Moroccan females are in the labor force is smaller than native females.

6.2 Second regression

Secondly, the regression is done for the total population to see whether the coefficient female is indeed increasing. An increase in 𝛽$%& means that it is becoming less negative because the

female labor force participation is increasing. Hence, the probability of participating in the labor force for women is increasing.

As apparent in table 5 in appendix A, the coefficient 𝛽$%& has a significant effect on the

probability of participating in the labor force for every year. The sign before the b is always negative, which means that if the participant is female, the probability of participating in the labor force is smaller compared to male participants.

Graph 7: 𝛽$%& plotted in a graph from 1990 until 2009 -1,2 -1 -0,8 -0,6 -0,4 -0,2 0 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009

Bfemale

21 As can be seen in graph 7, the 𝛽$%& increases over time. This means that the female labor

force participation is increasing over time. But to determine if there is indeed a significant difference between the effect of being female in 1990 and being female in 2009 the following hypothesis needs to be tested:

𝐻0: 𝛽_$%&,.\\" = 𝛽_$%&,W""\ 𝐻1: 𝛽_$%&,.\\" ≠ 𝛽_$%&,W""\

To test whether the 𝛽$%&,.\\" and 𝛽$%&,W""\ are significantly different from each other, the

Wald test of simple and composite linear hypothesis about the parameters is used. As apparent in table 6 in appendix A, one can conclude that the null hypothesis can be

rejected. Hence, the two coefficients are significantly different from each other and the female labor force participation indeed increased in the Netherlands between 1990 and 2009.

Thirdly, the same is done for Turkish and Moroccan females. The same probit regression is used when all the ethnicities are removed from the sample except for Turkish/Moroccan participants.

Graph 8: 𝛽$%&, 𝛽$%&,()* and 𝛽$%&,&,* plotted in a graph from 1990 until 2009 -1,6 -1,4 -1,2 -1 -0,8 -0,6 -0,4 -0,2 0 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009

22 As can be seen in graph 8, the 𝛽$%&,()* and 𝛽$%&,&,* also increase over time. But to

determine if there is indeed a significant difference between the effect of being female in 1990 and being female in 2009 the following hypothesis needs to be tested:

𝐻0: 𝛽_$%&,()*,.\\" = 𝛽_$%&,()*,W""\ 𝑎𝑛𝑑 𝐻0: 𝛽_$%&,&,*,.\\" = 𝛽_$%&,&,*,W""\ 𝐻1: 𝛽_$%&,()*,.\\" ≠ 𝛽_$%&,()*,W""\ 𝑎𝑛𝑑 𝐻1: 𝛽_$%&,&,*,.\\" ≠ 𝛽_$%&,&,*,W""\

To test these hypotheses, the Wald test of simple and composite linear hypothesis about the parameters is used. As apparent in table 7 and 8 in appendix A, one can conclude that both null hypotheses can be rejected. Hence, the two coefficients are significantly different from each other and the female labor force participation for Turkish and Moroccan females indeed increased in the Netherlands between 1990 and 2009.

6.3 Marginal effects

Finally, the average marginal effects have been calculated for 𝛽()* and 𝛽&,* to see how much

less likely Turkish/Moroccan women are to work. This is done for the year 2009. As apparent in table 9 in appendix A, the dy/dx for Turkish females in 2009 is -0,1091487 and for

Moroccan females it is -0,1271674. This can be interpreted as; if a female participant is Turks, the probability of participating in the labor force will decrease with 10,91 percent and if a female participant is Moroccans, it will decrease with 12,72 percent.

7. Conclusion and discussion

In this thesis the main question is: ‘What are the differences in female labor force

participation of different ethnical groups in the Netherlands?’. Using the probit model, the effect of being female and the effect of being a Turkish/Moroccan female is estimated. Finally, by testing the differences in time, this thesis tried to look whether the total female labor force participation has increased between 1990 and 2009 and if this applies to Turkish/Moroccan women as well.

Firstly, every year 𝛽$%&,()*/&,* appeared to have a negative effect on the probability of

participating in the labor force, except for the unknown years, 1990 (Turkish) and 1995 (Turkish and Moroccan). Therefore, one can conclude that Turkish and Moroccan females have a lower female labor force participation rate in comparison with Dutch native females. Secondly, the overall female participation rate increased over time. The female coefficient was -0,95 in 1990 and -0,66 in 2009. In addition, the same is done for the female coefficient of Turkish and Moroccan females. These two coefficients increased as well. The female

23 coefficient increased from -0,99 to -0,80 and from -1,34 to -0,92 for Turkish and Moroccan females respectively.

In conclusion, the most recent data shows that if a female participant is Turks, the probability of participating in the labor force will decrease with 10,91 percent and if a female participant is Moroccans, it will decrease with 12,72 percent. Hence, it is a fact that Turkish and

Moroccan female immigrants have a smaller labor force participation rate in comparison to native females in the Netherlands. However, the total female labor force participation has increased over time. In addition, the Turkish and Moroccan female participation rate has increased as well. Therefore, based on these findings, there is no reason to assume that the total female labor force participation rate is increasing faster than the Turkish and Moroccan female labor force participation rate. However, given the fact that the labor participation rate for immigrant female is still lower compared to natives, it is very important for the integration that the immigrant female labor force participation rises. The government can stimulate lower educated immigrants to increase their educational level.

It is important to mention that all the relations mentioned in this thesis are just correlations and never causal relations due to selection effects. Further investigation is needed to

determine the causal effects that contribute to the female labor force participation. For further investigation, it would be interesting to investigate what the causal reasons are that contribute to the lower female labor force participation rate of Turkish and Moroccan females. Another disadvantage in this research is that there is no distinction between part time workers and full-time workers. Regardless whether a person has a part-full-time job or a full-full-time, he or she had been included into the labor force. However, as mentioned in the literature review, there is a substantial difference between the percentage of women having a full-time job and the percentage of man having a full-time job.

24

8. References

Ahn, N., & Mira, P. (2002). A note on the changing relationship between fertility and female employment rates in developed countries. Journal of population Economics, 15(4), pp. 696.

Becker, G. S. (1965). A Theory of the Allocation of Time. The economic journal, pp. 493-517.

Bloom, D. E., Canning, D., Fink, G., & Finlay, J. E. (2009). Fertility, female labor force participation, and the demographic dividend. Journal of Economic Growth, 14(2), pp. 79-101.

Borjas, G. J., & Van Ours, J. C. (2000). Labor economics (Vol. 2). Boston, MA: McGraw- Hill.

Burniaux, J. M., Duval, R., & Jaumotte, F. (2004). Coping with Ageing.

Centraal Bureau voor De Statistiek, Divisie Kwartaire Sector en Leefsituatie. Allochtonen in Nederland... CBS [Host], 2001.

Cooper, R. (1983). A note on overemployment/underemployment in labor contracts under asymmetric information. Economics Letters, 12(1), pp. 81-87.

Crul, M., & Doomernik, J. (2003). The Turkish and Moroccan second generation in the Netherlands: Divergent trends between and polarization within the two

groups. International Migration Review, 37(4), pp. 1039-1064.

Dagevos, J. (2001). Perspectief op integratie; over de sociaal-culturele en structurele integratie van etnische minderheden in Nederland, pp. 41-46.

Heineck, G. (2004). Does religion influence the labor supply of married women in Germany?. The Journal of Socio-Economics, 33(3), pp. 307-328.

Hotz, V.,, J. Klerman & R. Willis (1997), The economics of fertility in developed countries, in M. Rosenzweig, O. Stark (eds.), Handbook of population and family Economics, Vol. 1A, Elsevier, Amsterdam, pp. 8-10.

Jaumotte, F. (2003). Female labour force participation: past trends and main determinants in OECD countries.

Lehrer, E. L. (1995), The effects of religion on the labor supply of married women, Social Science Research, 24, pp. 281-301.

25

and human behavior NBER, pp. 171-198.

Lundborg, P., Plug, E., & Rasmussen, A. W. (2014). Fertility Effects on Female Labor Supply, p.2

Marquardt, D. W. (1980). You should standardize the predictor variables in your regression models. Journal of the American Statistical Association 75: pp. 74–103.

Mincer, J. (1962). Labor force participation of married women: A study of labor supply. In Aspects of labor economics, Princeton University Press, pp. 63-105.

Psacharopoulos, G., & Tzannatos, Z. (1989). Female labor force participation: An international perspective. The World Bank Research Observer, 4(2), pp. 187-201.

Tijdens, K. (2006). Een wereld van verschil: arbeidsparticipatie van vrouwen 1945-2005, pp. 1-9.

Van der Valk, J., & Boelens, A. (2004). Vrouwen op de arbeidsmarkt. Sociaal-economische trends (3). Den Haag: Centraal Bureau voor de Statistiek, p.21.

Verloo, M. M. T., & Roggeband, C. M. (2006). Nederlandse vrouwen zijn geëmancipeerd, allochtone vrouwen zijn een probleem: De ontwikkeling van beleidskaders over gender en migratie in Nederland (1995-2005), pp. 158-162

Vlasblom, J. D., & Schippers, J. J. (2004). Increases in female labour force participation in Europe: Similarities and differences. European Journal of Population/Revue

26

Appendix A: results

Year btur _Error Std. z 1990 0,1231638 0,0483088 2,55** 1991 -0,7826646 0,1947945 -4,02*** 1992 1993 -0,1957844 0,0779472 -2,51** 1994 1995 -0,0346462 0,0650214 -0,53 1996 1997 -0,2092641 0,0730333 -2,87*** 1998 -0,2859171 0,077599 -3,68*** 1999 -0,3522093 0,0588821 -5,98*** 2000 -0,287382 0,0519456 -5,53*** 2001 -0,3744241 0,0402316 -9,31*** 2002 -0,3519725 0,0389271 -9,04*** 2003 -0,2549913 0,0370582 -6,88*** 2004 -0,2919427 0,0334814 -8,72*** 2005 -0,3222512 0,0329243 -9,79*** 2006 -0,3979361 0,0300954 -13,22*** 2007 -0,4507076 0,032247 -13,98*** 2008 -0,4353129 0,031457 -13,52*** 2009 -0,3696564 0,0295196 -12,52*** Note: *p<0.1, **p<0.05, ***p<0.01

27 Year bmor Std. Error z 1990 -0,3775021 0,0761403 -4,96*** 1991 -1,158196 0,2500707 -4,63*** 1992 1993 -0,5122127 0,1128323 -4,54*** 1994 1995 -0,0718417 0,0974087 -0,74 1996 1997 -0,2171366 0,0975626 -2,23** 1998 -0,2093856 0,09286 -2,26** 1999 -0,4080741 0,0767051 -5,32*** 2000 -0,4871413 0,0737802 -6,60*** 2001 -0,4813051 0,0538523 -8,94*** 2002 -0,3953341 0,0523573 -7,55*** 2003 -0,2447682 0,049102 -4,89*** 2004 -0,2940235 0,0417468 -7,04*** 2005 -0,2355163 0,0389584 -6,05*** 2006 -0,4684694 0,0359216 -13,04*** 2007 -0,4286016 0,0380323 -11,27*** 2008 -0,525682 0,037412 -11,38*** 2009 -0,4281217 0,0338182 -12,66*** Note: *p<0.1, **p<0.05, ***p<0.01

28

Year bfem Std. Error z

1990 -0,9516124 0,0088087 -108,03*** 1991 -0,1583759 0,0323369 -4,90*** 1992 -1,0431950 0,0104753 -99,59*** 1993 -1,0437610 0,0110171 -94,74*** 1994 -0,9699614 0,1015726 -91,74*** 1995 -1,0230030 0,0105332 -97,12*** 1996 -9,7558220 0,0104854 -93,04*** 1997 -0,9842615 0,0111574 -88,22*** 1998 -0,9772200 0,0115822 -84,37*** 1999 -0,9007631 0,0109335 -82,39*** 2000 -0,8922523 0,0107676 -82,86*** 2001 -0,9425982 0,0062055 -151,9*** 2002 -0,9266404 0,0060262 -153,77*** 2003 -0,8767250 0,0059377 -147,65*** 2004 -0,8279834 0,0055023 -150,48*** 2005 -0,8046374 0,0054784 -146,87*** 2006 -0,7651998 0,0054286 -140,96*** 2007 -0,7306253 0,0056522 -129,26*** 2008 -0,7074116 0,0056811 -124,34*** 2009 -0,6627457 0,0057699 -114,86*** Note: *p<0.1, **p<0.05, ***p<0.01

29

Value

𝝌2 _{= 763,44}

p > 𝝌2 = 0,000

Table 6: Wald test 𝛽$%&,.\\" and 𝛽$%&,W""\

Value

𝝌2 _{= 5,35}

p > 𝝌2 = 0,000

Table 7: Wald test 𝛽$%&,()*,.\\" and 𝛽$%&,()*,W""\

Value

𝝌2 _{= 14,72}

p > 𝝌2 = 0,0001

Table 8: Wald test 𝛽$%&,&,*,.\\" and 𝛽$%&,&,*,W""\

Variable dy/dx Std.

Error z

Turkish female -0.10915 0.00905 -12.06***

Moroccan female -0.12717 0.010448 -12.17***

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

30

Appendix B: test for multicollinearity

Variable VIF 1/VIF

Marital status 1.90 0.526687 Education VMBO/MBO 1.84 0.544026 HAVO/VWO/MBO 1.98 0.505456 HBO 1.47 0.682380 WO 1.22 0.817835 Unknown 1.02 0.981060 Gender 1.02 0.984248 Ethnic Turks 1.05 0.951022 Moroccans 1.06 0.944113

Other, Medirerranean Sea 1.00 0.996795

European (excl. Mid. Sea) 1.00 0.996088

Antillean 1.00 0.997347 Surinamers 1.01 0.990386 Indonesian 1.01 0.992120 Other 1.01 0.992568 Children 1 1.23 0.811647 2 1.34 0.748561 3 1.15 0.870285 4 or more 1.09 0.921368 Age 48.14 0.020774 Age2 45.25 0.022100 Mean VIF 5.56

31

Appendix C: test for linearity (age

added)

Variable Beta Std. Err. z

Marital status -.0210435 .0122274 -1.72* Education VMBO/MAVO .2891372 .0125441 23.05*** HAVO/VWO/MBO .5205001 .0124315 41.87*** HBO .8383905 .0174786 47.97*** WO .9325292 .0271163 34.39*** Unknown .6955251 .0696971 9.98*** Gender Female -.9516124 .0088087 -108.03*** ethnic Turks -.0344684 .0329440 -1.05 Moroccans -.2760413 .0440495 -6.27***

Other, Mediterranean Sea .0236122 .0571789 0.41 European (excl. Mid. sea.) -.1282454 .0321843 -3.98***

Antillean/Aruban -.3168317 .0634637 -4.99*** Surinamers -.0377242 .0375650 -1.00 Indonesian -.0620675 .0393703 -1.58 Other -.3244105 .0376732 -8.61*** Children 1 -.3244049 .0118359 -27.41*** 2 -.5392987 .0128972 -41.82*** 3 -.6543131 .0191197 -34.22*** 4 or more -.7762859 .0315240 -24.63*** Age .2433086 .0023038 105.61*** Age^2 -.0033694 .0000291 -115.83*** consant -3.207039 .0453062 -70.79*** LR 𝝌2 _39318,03*** Pseudo R2 0,2565 Note: *p<0.1, **p<0.05, ***p<0.01