Family size, birth order and educational attainment in The Netherlands

1

Family size, birth order and educational

attainment in The Netherlands

Thesis, June 2016

BSc in Economics and Business

Written by

Donna de Jonge

10589449 Supervised by

L. Geijtenbeek

University of Amsterdam

2

_________________________________ Abstract _______________________________ This paper aims to research the relationship between family size, birth order and educational attainment in the Netherlands. Research done on this topic has not yet been conducted with data from The Netherlands. This research is done using wave 1 of the NKPS. In an attempt to answer the research question, an ordered probit regression has been done first with educational attainment as the dependent variable, family size as the independent variable and demographics used as control variables. This regression will be repeated several times, adding a birth order index, parental cohorts and family background attributes. The first regression shows a negative correlation between family size and educational attainment. The second regression shows us that adding birth order does not significantly affect the coefficient of family size and also that birth order has a negative correlation with educational attainment. The third and fourth regressions show the effect of adding family attributes and the coefficients of family size and birth order remain negative. To interpret coefficients more effectively and as a robust check, a linear regression is done using amount of years studied as the dependent variable. This regression shows similar results for the effect of family size and birth order on educational attainment and suggests a negative relationship between family size and educational attainment and a negative relationship between birth order and educational attainment. For a monotonic check of birth order and to see the differences of birth order effects in different family sizes, a separate regression was done per family size. This regression shows a monotonic relationship.

This document is written by Donna de Jonge who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are 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

3

Contents

1. Introduction ... 4

2. Literature review ... 5

Family size ... 5 Birth order ... 6

3. Data ... 7

4. Method ... 11

Setting up a birth order index ... 12

Linear regression using amount of years studied ... 12

5. Results ... 14

Regression estimates ... 14

Ordered probit regression... 14

Birth order effects by family size ... 18

6. Discussion ... 21

Accomplishments ... 21

Shortcomings and further research ... 22

References ... 23

Appendix ... 24

Appendix A: List of variables ... 24

Appendix B: Educational system in The Netherlands ... 25

Appendix C: Full table distribution educational attainment across family size ... 27

Appendix D: Full table distribution educational attainment across birth order ... 28

4

1. Introduction

A negative stereotype of only children exists and they are viewed as selfish, lonely and maladjusted (Thompson, 1974). In general people perceive having siblings to have both advantages and disadvantages, while being an only child is seen to have only negative effects (Falbo, 1977). Even though their stereotype is negative, research has shown that only children are not doing worse at a lot of aspects than children from bigger families. Falbo and Polit (1986), who have conducted research on only children, show that only children in the United States do even better than children with at least two brother or sisters at achievement and intelligence and have a better relationship with their parents than children with siblings. Now that China hopes to boost the economy by abandoning its one-child rule, it is interesting to see how having siblings impacts the qualities of a child.

Economists are eager to understand the components that influence children’s qualities, but these qualities is hard to measure. One could look at the children’s achievements in life and one aspect that is often used in economic research is educational

attainment. An influence that seems of importance when it comes to future

educational attainment is family size. Family size may negatively influence children’s achievement because children will have to share parental resources. Family size could positively influence children’s achievements if children stabilize marriages or when siblings compete with each other on educational level.

In the Netherlands it is very common to have more than one child, only six percent of the people in the Netherlands do not have siblings and the most popular family size in the Netherlands is two children (Statistics Netherlands, 2003). Multiple studies show a negative relationship between child quantity and children’s qualities, but few to none of the studies were conducted using data on Dutch families. Every country has its own unique culture, therefore results from one country do not necessarily answers the same question for another country. One relevant example concerning the

difference between countries is their educational system together with the cost of studying. In the Netherlands children are obligated to follow education till the age of 18 and most of the education is subsidized and therefore available for everyone.

In conclusion, only children have a negative stereotype while research shows that they often do better at some aspects than children from bigger families. Children from small families may benefit because they do not have to share parental resources, while children from bigger families may benefit if they stabilize marriages or when there is positive competition between siblings. Research done on family size and educational attainment so far has not used data on Dutch families.

In this paper I aim to research the relationship between family size, birth order and educational outcomes of children using data on Dutch families. I expect family size to have a negative correlation with child quantity due to the fact that children will have to share parental resources. Furthermore I expect birth order to have a negative correlation with educational outcomes as is suggested by previous studies. More theory concerning this will be discussed in chapter two. The hypotheses about the influence of family size and birth order on educational attainment will be tested using a regression-based approach. Educational attainment is measured in ordinal data

5

using ten different levels of education and the data used for this analysis is Wave one of the Netherlands Kinship Panel Study (KNPS). The research method will be further discussed in chapter three. In the fourth chapter the results of the regression will be described and used to answer the research question and in chapter five a conclusion of the research will be given.

2. Literature review

A lot of research concerning family size, birth order and educational attainment has been done. However research done on this topic using data on Dutch families is very limited. In explanation Belmont and Marolla (1973), two researchers specialized in psychology, did research on the effect of family size and birth order on intelligence using Dutch data. They only used 19-year-old males so their research was not aimed at highest educational attainment. Besides, since they only included 19-year-old males their sample was not a good representation of the whole population of the Netherlands. In this section relevant findings of other papers concerning the effect of family size and birth order on future educational attainment done in other countries will be

discussed. First relevant findings of the effect of family size on educational attainment will be discussed followed by a discussion of the relevant literature on the effect of birth order on educational outcomes.

Family size

An influential article regarding the effect of family size is by Becker and Tomes (1976). They introduced the quantity quality model, in which they suggest that there is a negative relationship between the quantity and quality of children. They define child quality as educational outcomes and family size by the amount of siblings. Becker and Tomes find that children from bigger families in general have lower educational

outcomes, due to the tradeoff between quantity and quality in the budget constraint leading to an increasing marginal cost of quality of the child with respect to sibship size.

The dilution model is a much used argument for the negative correlation between family size and educational attainment. This model states that parental resources will be diluted with an increasing family size. Examples of parental resources are divided by Blake (1981) into three categories which are (1)"types of homes, necessities of life, cultural objects (like books, pictures, music and so on)," (2) "personal attention, intervention, and teaching," and (3) "specific chances to engage the outside world or, as kids say, 'to get to do things'” (p. 422). Blake’s research supports the dilution model by finding that the number of siblings have a negative effect on the quality of the child. In her analysis, done in the United states, siblings have shown to have a negative impact on intervening variables that affect college plans. Blake shows that with bigger family sizes, the availability of educational objects like books will decrease. Other research tells us that as sibship size increases, college expenses funded by parents (Steelman and Powell 1989, 1991) and time spend with parents (Hill and Stafford 1974) also decreases. Research done on eight graders in the United States show that parental resources explain most or all of the negative correlation between family size and educational outcomes (Downey, 1995).

6

Salvanes (2005). They used data on the entire population of Norway which makes the sample identical to the population. They did their research on an extended period of time and had information on all siblings and parents. Their research showed at first a negative correlation between family size and educational attainment, but after

including indicators for birth order the effect of family size became negligible. The research suggests family size itself has no significant effect on child quality and the chance is bigger that it impacts marginal children through the influence of birth order. They found that birth order has a significant negative effect on the educational

attainment of the child (Black et al., 2005).

Birth order

As discussed in the previous section Black et al. (2005) suggest family size itself has no significant effect on child quality and the chance is bigger that it impacts marginal children through the influence of birth order. They found that after adding birth order to the regression, the effect of family size became negligible.

The dilution model also supports the inverse relationship between birth order and educational outcomes due to the correlation between family size and birth order. Someone who is firstborn has the chance to be an only child, come from a medium sized family or from a big family. Having birth order six, excludes the possibilities of being in a small family and so parental resources will have to be shared with at least five siblings. But even firstborn children from big families can still have an advantage over children with higher birth orders due to not having to share parental resources up until siblings are born. This emphasizes the role of the gaps between the birth of siblings. When there is a large gap, the argument can also be argued the other way where younger siblings benefit more because their older siblings leave the parental house and the lastborn will not have to share resources anymore (Birdsall, 1991).

Therefore first and lastborn children will have an advantage over middle born children. Booth and Kee (2008) found a way to solve for the correlation between birth order and family size by introducing the birth order index. In their research they find a negative relationship between family size and educational attainment and that shares of

parental resources are decreasing by birth. Also they find that the effect of family size does not vanish after adding this birth order index.

In conclusion, Becker and Tomes (1976) introduced the quantity quality model which suggests a negative relationship between family size and educational attainment. The dilution model as explained by Blake supports this argument by saying that children will have to share parents resources with their siblings (1981). The research done in Norway shows a negative correlation between family size and educational attainment but this effect becomes negligible after adding birth order to the regression (Black et al, 2005). Booth and Kee (2008)introduced the birth order index to their regression to purge the family size effect from birth order and found that the negative correlation between family size and educational attainment did not vanish after including this index to their regression.

In this research I expect both family size and birth order to have a negative correlation with educational attainment as well supported by the dilution model. Whether the effect of family size will vanish after adding the birth order index, as introduced by booth and Kee, will follow from the results.

7

3. Data

Data

The data in this thesis is drawn from wave one of the Netherlands Kinship Panel Study (NKPS). The NKPS is a large-scale nationally representative study on Kinship in the Netherlands. This database is largely used for research on sociological or micro economical fields. Wave one was conducted between 2002 and 2004 and has the biggest sample size and thus is more reliable in predicting than the later waves. This wave focuses on basic questions on kinship like family size, educational attainment and information about, and relationships with, family members. Later waves focus more on changed situations since the first wave was conducted. The first wave

consisted of two different samples. The first sample is a random sample of individuals within private households and the second sample is the migrant sample consisting of the four biggest migrant groups. For this research the first sample was used because migrants did not have the same educational possibilities and educational system as individuals born in The Netherlands. This sample was completed by 8161 individuals living in private homes in the Netherlands between the age of 18 and 79.

This data set is excellently suited for this research because it provides not only information about the respondent itself regarding educational attainment, but also about family members of the respondent and family background attributes. However this data set also has some drawbacks. Unlike study done by Black et al. (2015) our data set does not cover the whole population of one country and thus is not as representative as their study. Also income of the parents of the respondent is not observed which is likely to be a big influence on their future education.

Educational attainment in the NKPS is not measured by amount of years studied but by different levels attained. In the data set used, individuals could choose between ten options ranging from no certificate to having obtained a master’s degree. Therefore the dependent variable in most of the analysis is educational attainment made up by ten categories as shown in appendix A.

852 cases of individuals younger than 28 when answering the survey were deleted. This is to make sure that they are finished with their education. Furthermore 2759 individuals born before 1950 were deleted to make sure that the second world war had no influence on their schooling decisions. Nine cases were dropped of individuals who did not know their educational attainment and/or amount of siblings and seven observations were dropped where the respondent did not know the years of birth of their parents. A few cases where the individual had stepchildren were deleted, because parents could treat their biological children differently than children from remarriages. Furthermore 25 cases of individuals with adopted siblings were deleted, because there was no information about the age at adoption. If the child was adopted at a later age, the adopting family had no influence on prior education. Besides

adopted- and stepchildren will disturb the relationship between birth order and family size since they would not be accounted for in birth order but they would be accounted for in family size. The removing of these special cases resulted in a sub sample of 4397 individuals.

Variables

A model using similar explanatory variables as used in Booth and Kee’s (2008) research is estimated. Dummy variables were created for the age of the respondent,

8

ordered in five categories: 28-33, 34-39, 40-45, 46-50 and 51-55. This variable is included because age could correlate with educational attainment. People that are older studied a longer ago and things were different. For example study grants were introduced in 2000 which made studying more affordable for the younger respondents. The dataset already had a dummy variable for sex which is one if the respondent is female. Moreover this variable is relevant since the sex of a person correlates with educational attainment. Especially a longer time ago it was less likely for woman to follow higher education than men because often they were expected to care for the children. Therefore it is necessary to include these control variables to make the estimated coefficient of the independent variables more reliable. Furthermore a dummy variable for the ethnicity was created. All respondents except the individuals from the following countries are labeled nonwestern: The Netherlands, Canada, Denmark, Germany, France, Ireland, Austria, Poland, Russia, UK, USA and Sweden. This is included because this variable also may correlate with educational attainment. Furthermore dummy variables were created for the parent’s age at birth, the amount read during childhood, the level of urbanization in the place the respondent lived during childhood, whether the mother worked during childhood and the amount the father worked. A dummy variable for the degree of the mother and father was also created which will be one if they obtained at least an intermediate vocational degree and zero otherwise. All of these dummy variables were included because they could correlate with educational attainment and/or family size and by including them the relationship between the independent and dependent variable can be measured more securely. Furthermore a very important thing when doing regressions is to make sure to have a proper control group. Concerning this particular research it is most reliable when families are apart from the dependent and independent variables identical. This is not the case because families with certain characteristics might end up with bigger family sizes. This is an important reason to include these control variables. By

including for example age, gender and educational attainment of the parents as control variables these different characteristics of the families are accounted for and the estimated coefficients for the independent variables will be estimated more securely.

Individuals were asked (Question AD101) “How many siblings do you have?” and (Question AD201) “How many half siblings do you have?”. The answers of these questions were used to construct the variable family size by creating

famsizeall=ad101+ad201+1. After creating the family size, a new variable famsize was created which put all family sizes from at least ten together in one category. The

respondents were asked all birth years of their (half)siblings which allowed me to construct the birth order of the respondent. I did not specifically account for twins, but when siblings were born in the same year the respondent would be treated as the younger one of the two. This is because the argument for family size influencing educational attainment mainly is based on the fact that parents have to divide resources. In the case of two siblings born in the same year (or at the same time) the parents immediately have to divide their resources by both children so both children should be accounted for as being second born when it comes to dividing the parent's resources. Furthermore a second variable for birth order was made which put birth orders of ten and above together in one category.

Appendix A summarizes these variables used for my analysis together with their description, mean and standard deviation. 21.9% is aged between 28 and 33, 25.8% is aged 34 to 39, 24.7% are between 40 to 45, 19.1% are between 46 and 50 and 8.5% is older than 50. Of the 4388 respondents, 58.7% is female, 10.5% have a university

9

degree and 43.1% read a lot during their childhood. The average number of children in a family in this sub sample is 3.82 with a standard error of 0.032. In this data set 33.7% of the individuals were firstborn, 27.4% were second born, 19.9% third born, 8.8%

fourth born and the others as stated in the appendix.

In the Netherlands in 2012 28% of the population had at least obtained a higher vocational degree (Statistics Netherlands, 2013). In the sample used in this research the percentage of individuals that obtained at least higher vocational is approximately 27%. Furthermore the most popular family size in the Netherlands is two children (Statistics Netherlands, 2003). In this sample the biggest group in family size is also two children. These examples suggest that the sample is a good representation of the population of the Netherlands.

Table 1 shows the distribution of birth order across family sizes. Each cell represents the amount of individuals with a certain birth order in different family sizes. For example in the table can be seen that in two-child families there are 631 individuals that were firstborn and 556 individuals that are the youngest. As shown in the table, larger family sizes have more categories for birth order and thus the cell sizes for bigger families are smaller than for small families. Besides in the table it can be seen that the number of observations is the biggest for two-child families, and this declines as family size increases which in turn also results in smaller cell sizes for bigger

families. Apparently bigger family sizes are less popular in the Netherlands. The table shows that the number of observations of nine-child families is only 86, and of these 86 observations only two were firstborn. The reason that there are more observations for later born children in the bigger families might be because everyone in the sub sample that was born before 1950 was deleted. Moreover in general people with eight younger siblings are expected to be older and thus likely to be born before 1950 than

respondents with only one younger sibling. From this table it can be seen that family size and birth order correlate and it is important to find a way to correctly represent this data. This will be done by creating a birth order index as explained later.

Table 1: Distribution of Birth Order Across Family Size

Family Birth Order

Size 1 2 3 4 5 6 7 8 9 ≥10 Total 1 186 0 0 0 0 0 0 0 0 0 186 2 631 556 0 0 0 0 0 0 0 0 1.187 3 370 340 374 0 0 0 0 0 0 0 1.084 4 162 170 189 188 0 0 0 0 0 0 709 5 73 73 87 86 105 0 0 0 0 0 424 6 26 26 52 57 56 78 0 0 0 0 295 7 16 16 17 25 30 30 35 0 0 0 169 8 6 8 8 17 17 15 25 27 0 0 123 9 2 8 8 6 9 14 14 13 12 0 86 ≥10 10 6 6 10 12 11 9 12 17 41 134 Total 1.482 1.203 741 389 229 148 83 52 29 41 4.397

In table 2 family size is cross-tabulated by educational attainment of the respondent. The full table can be found in appendix B, but for a better overview family sizes of 6 and bigger were put together and different levels of education were divided in seven different groups. The table shows the amount of individuals completing different levels of education across different family sizes. In between brackets the percentage of a certain family size completing that level of education is shown. In total the most

10

attained level of education is completing at least higher vocational. From this table it can be seen that respondents from bigger family sizes tend to have lower educational attainment as the proportions of the highest attained education get slightly smaller the bigger the family size. As shown in the table, two-child families have the highest proportion of respondents that completed at least higher vocational with 44.42%. As family size gets bigger this proportion decreases to 30.22% for families with at least six children. This table suggests a negative correlation between family size and educational attainment.

Table 2: Distribution of Educational Attainment Across Family Size

Educational Attainment Family Size 1 2 3 4 5 6 Total Incomplete Elementary School 1 3 2 4 4 8 22 (0.5%) (0.3%) (0.2%) (0.5%) (0.9%) (1.0%) (0.5%) Elementary School only 8 28 29 27 29 52 173 (4.4%) (2.4%) (2.7%) (3.8%) (6.8%) (6.5%) (3.9%) Lower Vocational and Lower General Secondary 53 209 214 175 109 268 1,028 (28.8%) (17.6%) (19.7%) (24.6%) (25.6%) (33.3%) (23.4%) Medium General Secondary 18 76 57 41 24 38 254 (9.8%) (6.4%) (5.3%) (5.8%) (5.6%) (4.7%) (5.8%) Upper General Secondary 7 45 40 24 10 19 145 (3.8%) (3.8%) (3.7%) (3.4%) (2.4%) (2.4%) (3.3%) Intermediate Vocational 37 298 284 199 109 176 1,103 (20.1%) (25.1%) (26.2%) (28.0%) (25.6%) (21.9%) (25.1%) Higher Vocational, University and Master 60 527 459 242 141 243 1,672 (32.6%) (44.4%) (42.3%) (34.0%) (33.10%) (30.2%) (38.0%) Total 184 1,186 1,085 712 426 804 4,397 (100%) (100%) (100%) (100%) (100%) (100%) (100%)

-Number in between brackets is column percentage -Percentages may not add to 100% due to rounding

Table 3 cross-tabulates birth order by educational attainment of the respondent. Again families of six and bigger were put together and educational attainment was

categorized in seven categories. The full table can be found in appendix C. The columns shows the amount of individuals completing each level of education

distributed by birth order and the percentage of individuals completing the level of education for a certain birth order is in between brackets. Again in this table a pattern can be seen where the proportion of the two highest attained levels of education get slightly smaller the higher the birth order gets starting from two-child families. In explanation the percentage of only children completing at least higher vocational is approximately 33%. This percentage increases to approximately 44% for two-child families and slowly decreases by family size to 30% for families consisting of at least six children. This suggests a negative correlation between birth order and educational attainment.

11

Table 3: Distribution of Educational Attainment Across Birth Order

Educational

Attainment _{1 2 3} Birth Order _{4 5 6 Total}

Incomplete Elementary School 7 5 1 2 2 5 22 (0.5%) (0.4%) (0.1%) (0.5%) (0.9%) (1.4%) (0.0%) Elementary school only 40 47 23 23 20 20 173 (2.7%) (3.9%) (3.1%) (5.90%) (8.7%) (5.7%) (3.9%)

Lower Vocational and Lower General Secondary 310 237 179 122 52 128 1,028 (21.0%) (19.7%) (24.1%) (31.3%) (22.7%) (36.3%) (23.4%) Medium General Secondary 95 70 38 17 17 17 254 (6.4%) (5.8%) (5.1%) (4.4%) (7.4%) (4.8%) (5.8%) Upper General Secondary 54 45 22 16 4 4 145 (3.7%) (3.7%) (3.0%) (4.0%) (1.8%) (1.1%) (3.3%) Intermediate Vocational 350 312 217 93 55 76 1,103 (23.7%) (25.9%) (29.3%) (23.8%) (24.0%) (21.5%) (25.1%) Higher Vocational, University and Master 622 489 262 117 79 103 1,672 (42.0%) (40.6%) (35.3%) (30.0%) (34.5%) (29.2%) (38.0%) Total 1,478 1,205 742 390 229 353 4,397 (100%) (100%) (100%) (100%) (100%) (100%) (100%)

-Number in between brackets is column percentage -Percentages may not add to 100% due to rounding

4. Method

In this paper the relationship between family size, birth order and educational attainment is investigated. Previous literature showed a negative relationship between family size and educational attainment and between birth order and educational attainment. To investigate this relationship it necessary to conduct regressions. One problem in estimating these relationships is that family size and birth order correlate, this will be solved by using the birth order index.

Black et al. showed in their research that the effect of family size became negligible after including birth order, while Booth and Kee’s research showed that family size, after adding the birth order index, still has a significant effect.

To test whether family size is still significant after adding birth order for the data used it is necessary to first conduct a regression including family size and excluding birth order and then add birth order. In this way the effect of adding birth order to the regressions and what happens to the coefficient of family size can be seen.

Another difficulty is the fact that the dependent variable is ordinal and therefore an ordinary least squares regression cannot be used. Without doing a Linear regression it is hard to interpret the coefficients and therefore a linear regression will be conducted. Birdsall (1991) suggests first and lastborn children might both have an advantage over middle born children when older siblings leave the house and the last born will have to share parental resources with less siblings. This suggests that the effect of birth order might be non-monotonic. To test this a separate regressions for different family sizes will be conducted using dummy variables for birth order.

12

Setting up a birth order index

Family size and birth order correlate with each other, because people from smaller families are always more likely to have a higher birth order than people from bigger families. Therefore both including family size and birth order as dependent variables could cause for unreliable estimates. To solve for this correlation Booth and Kee’s method of creating the birth order index is followed. This birth order index effectively purges effect of family size from birth order. The correlation between family size and birth order in my data is 0.7535 and the correlation between the new birth order index and family size is only 0.1795. The way the Birth order index is made up is as follows:

_{( ) ⁄} (1)

Where B is the birth order index and is the absolute birth order of the respondent which can take the values one to ten. N is the total amount of children in the family including the respondent which also can take values one to ten. The denominator in this equation is the average birth order for the family, so the birth order is specified as the absolute birth order divided by the average birth order.

This ratio of birth order to average birth order will take data values between 0.18 and 1.82 for the data used. BOI=1 will be the within-family mean for all families and respondents with value BOI=1 are a middle child. My estimating equations are:

(2)

and

(3)

in this equation stands for educational attainment of the respondent represented by 10 categories ranging from no qualification to having a master’s degree. Demographics and family background attributes are collected in . Four regressions will be done. First equation (2) will be estimated and only the demographics of the respondent will be included in . Demographics are essential control variables and are unlikely to correlate with the independent variables and therefore are suited to include in the first regression to measure the relationship between family size and educational attainment. In the second regression birth order will also be included to see the effect of including this on the estimated probability of family size thus equation (3) will be estimated with again only demographics included in . As mentioned before the two researches done by Black et al. and Booth and Kee show contrary results when adding birth order to the regression and this regressions will reveal what happens with the estimated coefficients after adding the birth order index. The third regression will be the same as the second regression but then parental cohorts will also be included in . These are the age of the parents at birth of the respondent and whether they obtained a degree. In the fourth and last regression of this part equation (3) will be estimated and all control variables will be included in .

I expect α to be negative as previous research has shown a negative correlation

between family size and educational attainment, due to dilution of family resources. I also expect to be negative as is fond both in the papers of Black et al. and Booth and Kee. Furthermore the regression will also reveal whether adding the birth order index will make the effect of family size negligible or not.

Linear regression using amount of years studied

13

is difficult to interpret the regression results. Therefore another regression will be done with amount of years studied as the dependent variable. This dependent variable allows us to use a linear regression. The way the amount of years studied is

constructed is done by using the minimum amount of years studied to reach a certain level of education in the Netherlands. 22 cases were dropped of respondents that had incomplete education. They were left out of this regression because it was not known after how many years of studying they stopped before getting their qualification. By giving all of them zero years of studying these 22 respondents become outliers because the difference between no qualification and finishing elementary school would be eight years. Background information regarding the Dutch schooling system is given in appendix A. The following equations will be estimated:

(4),

(5) and

(5)

In these equations represents the amount of years studied. In equation (4) and (5) the other coefficients are the same as in equation (3). In equation (5) the coefficient of birth order will be estimated instead of the birth order index so the result can be better interpreted.

Regressions per family size

To see the effect of different birth orders and check whether birth order has a

monotonic relationship with educational attainment, as suggested by Birdsall (1991), another type of regression will be done. Five regressions will be done for five different types of family sizes. The families with one child will be excluded from this section, because all the children will be firstborn and therefore including this will be useless. The five remaining family sizes used are two-child families, three-child families, four-child families, five-four-child families and families with at least six four-children. The

dependent variable in this regression will still be educational attainment in ordinal categories and the independent variable is birth order made up of six dummy

variables: firstborn, second born, third born, fourth born, fifth born and a birth order of at least six. The base will be first born. Since the dependent variable is ordinal in this regression an ordered probit regression will be done. The estimation equations will be: (6) in which, ( ) ( ) ( ) ( ) ( ) ( ) (7) and (8)

These are the dummy variables for different birth orders and birth orders of at least six are put together.

Family size is left out of the equations, because we will do separate regressions for each kind of family size. In other words these regressions will be estimated using five

14

different samples, each sample consisting of respondents with the same family size. In this way the effect of birth order is also purged from family size. In equation (6) it can be seen whether the relationship between birth order and educational is monotonic and equation (8) helps comparing the results to the previous regressions done estimating equation (3).

5. Results

Regression estimates

Ordered probit regression

Table 4 represents the coefficients of an ordered probit regression of educational attainment using equation (2) and (3). Educational attainment is made up of 10 categories where 1 represents no qualification and is the lowest and 10 is a master’s degree and is the highest.

Specification [1] includes demographics ,which are dummy variables for the age of the respondent (with 28-33 used as the base), the sex and the ethnicity. Besides

demographics it contains the independent variable family size. Specification [2] is the same as specification [1] but also includes the birth order index. Specification [3] adds parental cohorts which are dummy variables for the age of father and mother at birth(both with as base <21), mum degree and dad degree. Specification [4] also adds in the rest of the variables which are the degree of reading (base is no reading), area of parental home (with base not urbanized), dummy variables for the work of dad (with less than part-time as the base), mom worked till the age of 15 and lived with both biological parents till the age of 15.

The estimates of specification [1] show a negative coefficient for family size significant at a 1% level. This means the correlation between family size and educational

attainment controlled for demographics is negative. In other words children from bigger families will on average attain less educational attainment than children from smaller families. As expected educational attainment is decreasing in family size.

Specification [2] adds in birth order to the regression and this coefficient is also significantly negative. This is in line with my hypothesis. The addition of birth order to the regressions showed different effects on the coefficient of family size in the literature and therefore the effect on family size is an interesting aspect to look at. It can be seen from the table that the coefficient for family size does not change much by adding the birth order index: it increased only slightly from -0.066 to -0.062. This is a difference of 0.004 which is smaller than the standard errors of both coefficient and it is therefore not a significant change.

Specification [3] adds in parental cohorts like the age of the parents at birth of the respondent and the education of the parents. The coefficient of family size barely changes but the coefficient of birth order noticeably decreased from -0.088 to -0.317. This is a big difference and therefore important to find an explanation for. A possible reason for this big decline can be due to correlation. Birth order and the age of parents can be correlated because individuals with a higher birth order in general have older parents, so this correlation is not surprising. The question is whether including these dummy variables makes the regression better or worse than not including these

15 variables.

One reason for this big change in the coefficient of birth order could be

multicollinearity. Multicollinearity can be a problem because it can make the estimates of the coefficients unstable due to an increase in the variance of these estimates. When these two variables are highly correlated it is difficult to determine what part of the effect is caused by what variable. The correlation between birth order and the age of the mother and the age of the father are respectively 0.5431 and 0.5180. even though these correlations are quiet high they are not extremely high and no reason yet to exclude these variables from the regression. Furthermore the standard error did not increase considerably (increased from 0.038 to 0.048) which makes multicollinearity unlikely.

Another reason for this change in the coefficient could be that older parents have a positive influence on the child’s educational attainment.

If the age of parents at birth and birth order strongly correlate and adding the age of parents at birth causes the coefficient of birth order to decline, this might be the case because the age of the parents have a strong positive influence on educational

attainment. The age of parents, controlling for family size and birth order, might have a positive influence on their children because younger parents may be less patient and less amenable to give up a career or social concerns to spend time with their children which in turn might help with developing their learning potential. By not including this variable this positive influence is reflected in the coefficient of birth order and thus is not as negative as it should be. The literature available does not provide us with a proven argument why the age of parents have a positive influence on

educational attainment, but there is also no clear prove that the variable must be removed due to multicollinearity. It is important to include parental cohorts like parental age and parents education because as can be seen from the regression they have a significant relationship with children’s education and it is not what we want to estimate. We want to estimate the effect of birth order and not the effect of being born to older parents. Therefore I think specification [3] is more suited in this case than specification [2] to evaluate the effect of birth order.

Specification [4] was done to see the effects of adding more control variables, and by adding these variables the coefficients of birth order and family size compared to specification [3] do not change considerably.

When looking at the control variables, it can be seen that in specification [1] all the demographic control variables have negative coefficients. Moreover educational attainment is declining in age, which could be due to the study grants introduced in 2000 that made studying easier accessible. Furthermore the coefficients suggest that females attain less education than males and nonwestern people attain less education than western people.

Specification [2] adds in birth order and as can be seen from this regression the coefficients of all control variables stay approximately the same.

Specification [3] adds in parental cohorts and it can be seen that for the demographics nothing changes, these coefficients are still significantly estimated to be negative and thus all have a negative relationship with educational attainment. From the parental cohorts it can be seen that the estimates of the age of both the mother and father at birth of the respondent are significantly positive and thus educational attainment of

16

the child is decreasing in age of the parents. As explained before younger parents may be less patient and less amenable to give up a career or social concerns to spend time with their children which in turn might help with developing their learning potential. Another reason for this positive effect is that higher educated people on average start later with having children than lower educated people. Specifically higher educated woman postpone maternity. For men the relationship between fertility and education only persist with the firstborn child (Statistics Netherlands, 2012). As we will turn to next children with high educated parents are likely to get high educated as well. The regression shows that when parents have a degree, this significantly increases the probability of higher education of their children. This could be due to family wealth effects or because it provides a supportive educational background.

In specification [4] all other variables are added. From the demographics and parental cohorts none of the coefficients significantly change. It can be seen that the

urbanization of the place the respondent lived during childhood has a positive

relationship with educational attainment compared to someone who lived in place that was not urbanized. However the coefficients are bigger for less urbanized cities which states the contrary and the coefficients are all not significant and therefore not

relevant.

The next variabls in the regression is the amount of reading that was done during childhood. These dummy variables also have positive coefficients, so children that read a lot during childhood have a higher probability of attaining more education than children that did not read during childhood. Therefore there is a positive relationship between the amount of reading done and educational attainment but not all

coefficients are significant and this does not suggest a causal relation. The dummy variable made for a mother that worked is also positive but not statistically significant. I included this dummy variable as a proxy for parents income, since a mother that worked during childhood probably had a bigger income. But an opposite argument can also be made, one could also say that poorer families will need the mother to work. These opposite arguments could be a reason for this dummy variable to have a negligible coefficient. Furthermore the amount worked by the father of the child is statistically significant. If the father worked fulltime or part-time the probability of the child attaining a higher level of education is bigger, this can be seen from the statistically significant positive coefficients. The last family attribute is the dummy variable for the situation where the child lived with both biological parents during childhood. This does not seem to have a significant relationship with educational attainment as the coefficient is not significant. Even though not all control variables are significantly important they can be kept in the regression without difficulties. They do not interfere considerately with the explainable variables since the

coefficients estimates of family size and birth order stay roughly the same with the same significance. Also the total fit of the model improved slightly with the pseudo increasing from 0.0444 to 0.0483.

Since children from one child families all have the same birth order index they could be intervening with the estimates. One child families could also be intervening since they could have characteristics that are different from bigger families. I checked this by doing an extra regression. This regression can be found in appendix D. The results of this regression suggests no big change in the estimates compared to table 4 and therefore the regression estimates in table [4] seem reliable. The coefficient for family size is -0.077 which, compared to -0.067, is not significantly changed and the birth

17

order index is in specification [4] of table 4 -0.313 and when excluding only children the coefficient is -0.306. Later on in table 6 and table 7 the relationship between birth order and educational attainment will be shown without the effects of different family sizes due to different characteristics.

Table 4: Ordered Probit Regression (Categorical Educational Attainment as Dependent Variable)

Spec[1] Spec[2] Spec[3] Spec[4]

Coeff SE Coeff SE Coeff SE Coeff SE

Family Composition famsize -0.066 (0.007)*** -0.062 (0.008)*** -0.066 (0.008)*** -0.066 (0.008)*** BOI -0.088 (0.038)* -0.317 (0.048)*** -0.313 (0.049)*** Demographics age3439 -0.182 (0.045)*** -0.184 (0.045)*** -0.165 (0.046)*** -0.161 (0.046)*** age4045 -0.234 (0.046)*** -0.242 (0.046)*** -0.238 (0.047)*** -0.239 (0.048)*** age4650 -0.213 (0.050)*** -0.225 (0.050)*** -0.237 (0.052)*** -0.243 (0.052)*** age5155 -0.297 (0.064)*** -0.310 (0.064)*** -0.293 (0.065)*** -0.303 (0.066)*** female -0.130 (0.031)*** -0.130 (0.031)*** -0.148 (0.032)*** -0.163 (0.032)*** nonwestern -0.303 (0.114)** -0.313 (0.114)** -0.270 (0.115)* -0.234 (0.117)* Family Attributes mum2125 0.246 (0.075)*** 0.201 (0.075)** mum2630 0.334 (0.080)*** 0.280 (0.080)*** mum3135 0.407 (0.089)*** 0.346 (0.090)*** mum3640 0.400 (0.101)* 0.342 (0.101)*** mum41up 0.335 (0.132)* 0.286 (0.132)* dad2125 0.258 (0.119)* 0.236 (0.120)* dad2630 0.352 (0.122)** 0.321 (0.122)** dad3135 0.435 (0.127)*** 0.409 (0.127)*** dad3640 0.487 (0.133)*** 0.469 (0.133)*** dad41up 0.618 (0.140)*** 0.616 (0.140)*** mum_deg 0.491 (0.065)*** 0.461 (0.066)*** dad_deg 0.711 (0.065)*** 0.655 (0.046)*** kidvsurban 0.006 (0.054) kidsurban 0.032 (0.048) kidmurban 0.051 (0.051) kidhurban 0.050 (0.049) kidreadlot 0.212 (0.036)*** kidreadfew 0.025 (0.043) workmum 0.012 (0.039) workdadfull 0.343 (0.077)*** workdadpart 0.345 (0.125)** fambio 0.136 (0.127) N 4397 4397 4397 4397 Pseudo R2 0.0097 0.0100 0.0444 0.0483 *p < 0.05. **p < 0.01. ***p < 0.001

18

Linear regression

Table 5 represents the coefficients of a linear Regression with amount of years studied as the dependent variable. The coefficient of family size is -0.074, this means that for every extra sibling a child has, the amount of years he or she will study is reduced by 0.074 years. A child with four siblings will study (4×-0.074) 0.296 years less than a child without siblings.

Specification [2] adds in the birth order index and the coefficient of family size does not change much. The coefficient for family size decreased from -0.074 to -0.071, but this difference of 0.003, when compared to the standard errors, is not significant. Therefore the inclusion of birth order does not make the effect of family size almost negligible. Interpreting the coefficient of the birth order index is not as straight forward. If a family with three children is considered, the middle child will have a birth order index of one. The eldest child will have a birth order index of 0.5

(1/((3+1)/2)) and the youngest will have a birth order index of 1.5 (3/((3+1)/2)). In this case the eldest will follow 0.219 more years of education than the middle child ((0.5-1)×-0.438) and the youngest will follow 0.219 years less of education than the middle child. To solve for this problem one regression was done estimating equation (5). In this regression birth order was used instead of the birth order index and in this way coefficient of birth order can be easily interpreted. The results of this regressions is seen in specification [3]. The coefficient for birth order is -0.150. This means that when birth order is increased by one, the amount of years studied decrease by 0.150 years. The downside of this regressions is that it correlates with family size and this can be seen in the table since the coefficient for family size is now almost negligible.

Table 5: Linear Regression (Amount of Years Studied as Dependent Variable)

Spec[1] Spec[2] spec[3]

Coeff SE Coeff SE Coeff SE

Family Composition

famsize -0.074 (0.017)*** -0.070 (0.017)*** 0.012 (0.026)

BOI -0.445 (0.100)***

birthorder -0.150 (0.034)***

Demographics yes yes yes

Family Attributes yes yes yes

Constant 13.806 14.098 13.702

N 4366 4366 4366

R2 0.1065 0.1106 0.1105

Adjusted R2 0.1006 0.1045 0.1044

*p < 0.05. **p < 0.01. ***p < 0.001

Birth order effects by family size

Table 6 summarizes regressions made with the dummy variables for birth order instead of the birth order index using equation (6) to better show the effects of different birth orders. To make sure it does not correlate with family size a separate regression for all family sizes was done with families with at least 6 children put together. In all five regressions all control variables were included.

19

child is -0.206, thus the probability that a second born child will follow a higher level of education is smaller than of the firstborn. In a three-child family, the coefficient for second born is -0.170 and for lastborn -0.278. This means that a second born will less likely follow higher education than the firstborn and for the lastborn this probability is even smaller. For the other family sizes similar result can be seen: the coefficients are negative and decreasing in birth order. An increase in birth order decreases the

probability of higher educational attainment. In other words firstborn have a higher probability of attaining higher educational attainment than later born children. This table suggest a monotonic effect of birth order and firstborn and lastborn children do not both have an advantage over middle born children as suggested by Birdsall (1991). Only the estimates for the bigger families are not giving similar results as the rest, but these statistics are also not as significant as the coefficients measured for smaller families. The reason these statistics do not match the statistics of the other family sizes can be due to the fact that families of six children and more were put together. In addition there are not as many observations of larger families as there are for the smaller family sizes since they are not very common in The Netherlands and with a smaller sample size the estimates are less reliable.

Table 7 is done using equation (8) and shows the birth order index per family size. In specification [4] of table 4 it could be seen that the estimated coefficient for the birth order index was -0.313. The estimated coefficients per family size in table 7 show small differences in the coefficients for different family sizes. The smallest coefficient is -0.281 for three-child families and the highest coefficients is -0.419 for four-child families. There is no clear pattern in the different birth order indexes as seen in the table. However the two highest birth order indexes are from the two groups of family sizes with the smallest sample. Therefore the noticeable differences are probably due to a smaller sample size and thus a bigger standard error which can also be seen in the table. Therefore it can be concluded that in general the birth order index is the same for different family sizes. The birth order indexes found in this regression are in line with the birth order index found when doing a regressions on all families

20

Table 6: Ordered Probit Regression Across Different Family Sizes Using Birth Order Dummy Variables (Categorical Educational Attainment as Dependent Variable)

Family size

2-children 3-children 4-children 5-children ≥6 children

Birth order 2nd -0.206 (0.067)** -0.170 (0.083)* -0.202 (0.122) -0.104 (0.176) -0.315 (0.176) 3rd -0.278 (0.096)** -0.350 (0.134)** -0.494 (0.183)** -0.219 (0.147) 4th -0.510 (0.155)*** -0.420 (0.201)* 0.006 (0.127) 5th -0.530 (0.219)* 0.117 (0.116) ≥6th -0.163 (0.042)***

Demographics yes yes yes yes yes

Family attributes yes yes yes yes yes

Observations 1162 1059 697 418 798

Pseudo R2 0.052 0.053 0.05 0.0526 0.0438

*p <0 .05. **p <0 .01. ***p <0 .001

Table 7: Ordered Probit Regression Across Different Family Sizes (Categorical Educational Attainment as Dependent Variable)

Family size

2-children 3-children 4-children 5-children ≥6 children

Birth order index -0.309 (0.101)** -0.281 (0.096)** -0.419 (0.125)*** -0.416 (0.159)** -0.380 (0.131)**

Demographics yes yes yes yes yes

Family attributes yes yes yes yes yes

Observations 1162 1059 697 418 798

Pseudo R2 0.053 0.0546 0.051 0.0516 0.042

21

6. Discussion

Accomplishments

This research investigated the effect of family size and birth order on educational attainment. I expected family size and birth order to have a negative effect on educational outcomes. The ordered probit regression found a negative relationship between family size and educational attainment. When adding birth order to the regression the coefficient of family size stayed almost the same and the estimated coefficient of birth order showed a negative relationship with educational attainment. By using the birth order index as introduced by Booth and Kee family size could effectively be purged from birth order. A significant negative coefficient was found for the birth order index suggesting that siblings do not get equal shares of educational resources. Moreover it was found that the eldest get more shares and these shares are declining with birth order.

The Linear regression showed similar results as the ordered probit regressions and confirmed the negative relationship between family size and educational attainment and between birth order and educational attainment. This regression also showed that the coefficient of family size, after adding the birth order index, stayed roughly the same. The linear regression was repeated with birth order as a dependent variable and this showed a negative correlation between birth order and educational

attainment and made the coefficient of family size negligible.

A few regressions were done using different samples consisting of different family sizes. These regressions showed that the negative relationship between birth order and educational attainment was monotonic. This regression also showed that the birth order indexes were roughly the same for different family sizes.

The negative relationship between family size and educational attainment as shown in the ordered probit regressions is consistent with the dilution model. As family size increases, siblings will have to share parental resources with more siblings. After adding birth order to my analysis the effect of family size stayed the same and is therefore not consistent with the findings of Black et al. (2005), where the effect of family size on educational attainment became almost negligible after adding birth order to the regression. The reason my findings are different is probably because they did not use the birth order index in their research. Moreover my findings are

consistent with the findings of Booth and Kee (2009) who did use the birth order index. They found a coefficient of -0.263 for the birth order index which is similar to -0.313 as fond in this research. Moreover their coefficient of family size was -0.101 compared to -0.066 in this research. Their coefficient also stayed roughly the same after adding the birth order index to their regression. The coefficient for family size found in my analysis is a little bit smaller than the coefficient found in the research of booth and Kee but the coefficient found for birth order is bigger. This could be due to differences in the culture of the country or differences in families between England and the Netherlands. One explanation for the difference could be the fact that studying in England is known to be more expensive than studying in The Netherlands. Therefore sharing resources of parents could have a bigger impact in England on educational attainment than in The Netherlands.

22

Shortcomings and further research

As the subsample used for this analysis was quiet small with 4396 respondents I feel that the use of a bigger sample will give us better insights on the relationship between family size, birth order and educational attainment. A research as done by Black et al. is more reliable as it includes information on all the residents of Norway. Even though the NKPS states to be a good national representation there could still be a certain selection error when participating in panels attracts people with certain

characteristics. Furthermore a shortcoming of this analysis is the exclusion of instruments for twin births.

Another shortcoming of research done in this field is the fact that it is difficult to use a proper control group. To effectively research the relationship between family size, birth order and educational outcomes it is necessary to make sure that the small

families and the larger families are identical apart from the size. This in general is not the case. People that choose to have a large family might have other characteristics than people who only choose to have one or two children. Now that china has

abandoned its one child rule it is perhaps an interesting sample for further research regarding the effects of siblings.

Another interesting subject for further research is a proper linear regression. A Linear research is easier to interpret and therefore could be more interesting to look at. The linear research done in this paper was done by using either the birth order index or birth order, but when using birth order this effect was not purged from family size. A more in-depth analysis using a linear regression and a way to purge birth order from family size would be interesting.

23

References

Belmont, L. & Marolla, F. (1973). Birth Order, Family Size, and Intelligence. Science, 182(4117), 1096-1101.

Birdsall N (1991) Birth order effects and time allocation. In: Schultz TP (ed)

Research in population economics: a research annual. JAI Press, Greenwich, 191–213.

Black, S., Devereux, P., & Salvanes, K. (2005). The More the Merrier? The Effect of Family Size and Birth Order on Children's Education. The Quarterly Journal of Economics, 120(2), 669-700.

Blake, J. (1981). Family Size and the Quality of Children. Demography, 18(4), 421-442.

Booth, A., & Kee, H. (2009). Birth Order Matters: The Effect of Family Size and Birth Order on Educational Attainment. Journal of Population Economics, 22(2),

367-397

Downey, D. (1995). When Bigger Is Not Better: Family Size, Parental Resources, and Children's Educational Performance. American Sociological Review, 60(5), 746-761.

Falbo, T. & Polit, D.F. (1986). Quantitative review of the only child literature:

Research evidence and theory development. Psychological Bulletin,100(20. 176-189.

Falbo, T. (1977). The only child: A review. Journal of individual psychology ,33 no. 1: 47-61.

Hanushek, E.A. (1992). The trade-off between child quantity and quality. Journal of Poltical Economy, 100(1):84-117.

Hill, C. Russell & Frank P. Stafford. 1974. Al- location of Time to Preschool Children and Educational Opportunity. Journal of Human Resources, 9:323-41.

Statistics Netherlands. (2003). One out of eleven children is an only child [data file]. Retrieved from https://www.cbs.nl/nl-nl/nieuws/2003/06/een-op-de-elf-kinderen-is-enig-kind

Statistics Netherlands. (2012). Fertility of men and woman considering their educational background[data file]. Retrieved from https://www.cbs.nl/nl- nl/achtergrond/2012/37/vruchtbaarheid-van-mannen-en-vrouwen-naar-opleidingsniveau

Statistics Netherlands. (2013). Educational level of population increased [datafile]. Retrieved from https://www.cbs.nl/nl-nl/nieuws/2013/40/onderwijsniveau-bevolking-gestegen

Steelman, Lala Carr & Brian Powell (1989). Acquiring Capital for College: The Con- straints of Family Configuration. American Sociological Review, 54:844-55. Thompson, V.V. (1974). Family size: Implicit policies and assumed psychological

24

Appendix

Appendix A:

List of variables

Variable Description Mean Std. Dev. female Dummy=1 if respondent is female 0.587 0.492 age2833 Respondent is aged 28-33 0.219 0.414 age3439 Respondent is aged 34-39 0.258 0.438 age4045 Respondent is aged 40-45 0.247 0.431 age4650 Respondent is aged 46-50 0.191 0.394 age5155 Respondent is aged 51-55 0.085 0.279 edu1 Incomplete elementary 0.005 0.071 edu2 Elementary school only 0.039 0.194 edu3 Lower vocational 0.130 0.336 edu4 Lower general secondary 0.104 0.306 edu5 Medium general secondary 0.058 0.233 edu6 Upper general secondary 0.033 0.179 edu7 Intermediate vocational 0.251 0.434 edu8 Higher vocational 0.260 0.439

edu9 University 0.105 0.306

edu10 Master 0.015 0.123

mum20 Mother was <20 when respondent was born 0.044 0.204 mum2125 Mother was 21-25 when respondant was born 0.242 0.429 mum2630 Mother was 26-30 when respondant was born 0.329 0.470 mum3135 Mother was 31-35 when respondant was born 0.210 0.408 mum3640 Mother was 36-40 when respondant was born 0.115 0.318 mum41up Mother was >41 when respondent was born 0.034 0.180 dad20 Father was <20 when respondent was born 0.013 0.112 dad2125 Father was 21-25 when respondant was born 0.134 0.341 dad2630 Father was 26-30 when respondant was born 0.318 0.466 dad3135 Father was 31-35 when respondant was born 0.271 0.444 dad3640 Father was 36-40 when respondant was born 0.149 0.356 dad41up Father was >41 when respondent was born 0.106 0.308 mum_deg Mother had degree 0.076 0.265 dad_deg Father had degree 0.180 0.385 kidvsurban Lived in very strong urbanized 0.153 0.361 kidsurban Strong urbanized 0.240 0.427 kidmurban Mildly urban 0.192 0.394 kidhurban Hardly urbanized 0.217 0.412 kidnurban Lived not in a urbanized place 0.134 0.340 kidreadlot Read a lot during childhood 0.431 0.495 kidreadfew Read few books during childhood 0.199 0.399 kidreadhardly Read hardly during childhood 0.289 0.453 nonwhite Ethnic group is nonwhite 0.019 0.136 workmum Mother worker when respondent was 15 0.207 0.405 workdadfull Father worked fulltime when respondent was 15 0.930 0.254 workdadpart Father worked part time when respondent was 15 0.024 0.152 workdadless Father worked less than pt when 15 years old 0.030 0.171 famnorm Lived with both biological parents till age 15 0.983 0.128 famsize Number of children in respondents family 3.821 2.097 birthorder Birth order respondent 2.586 1.828 firstborn Dummy=1 if respondent is the eldest 0.337 0.472 onlychild Dummy=1 if respondent is only child 0.059 0.237 birthorder2 Birth order is second 0.274 0.446 birthorder3 Birth order is third 0.169 0.375 birthorder4 Birth order is fourth 0.088 0.284 birthorder5 Birth order is fifth 0.052 0.222 birthorder6 Birth order is sixth 0.034 0.180 birthorder7 Birth order is seventh 0.019 0.136 birthorder8 Birth order is eighth 0.012 0.108 birthorder9 Birth order is ninth 0.007 0.081 birthorder10 Birth order is tenth 0.009 0.096 BOI Birth order index 1.040 0.417 aeduyrs Amount of years studied 14.428 2.434

25

Appendix B:

Educational system in The Netherlands

The translations of the different levels of educational attainment are as specified by the NKPS:

1= Incomplete elementary lagere school niet afgemaakt 2= Elementary school only lagere school incl vglo

3= Lower vocational IBO, huishoudschool

4= Lower general secondary mavo, ulo, mulo 5= Medium general secondary havo, mms

6= Upper general secondary vwo, atheneum, gymnasium

7= Intermediate vocational mbo, kmbo

8= Higher vocational hbo, kandidaats

9= University Universiteit

10= Post-graduate Master

The educational system in The Netherlands is summarized in the following diagram as described Nuffic, which is the Dutch organization for international cooperation in higher education.

26

From this scheme we can conduct the minimum amount of years needed to attain a certain level of education. The amount of years I converted for each level is as follows:

Elementary school 8 years

Lower vocational and lower general secondary 12 years General secondary and intermediate vocational 13 years

Upper general secondary 14 years

Higher vocational and University 17 years

27

Appendix C:

Full table distribution educational attainment across family size

Family size Educational attainment 1 2 3 4 5 6 7 8 9 10 Total 1 1 3 2 4 4 3 0 3 1 1 22 0.54 0.25 0.18 0.56 0.94 1.02 0.00 2.46 1.16 0.76 0.50 2 8 28 29 27 29 18 8 7 7 12 173 4.35 2.36 2.68 3.81 6.84 6.12 4.73 5.74 8.14 9.16 3.94 3 24 118 107 99 57 49 38 32 20 26 570 13.04 9.95 9.88 13.96 13.44 16.67 22.49 26.23 23.26 19.85 12.99 4 29 91 107 75 52 35 20 14 11 23 457 15.76 7.67 9.88 10.58 12.26 11.90 11.83 11.48 12.79 17.56 10.41 5 18 76 57 41 24 16 5 5 3 8 253 9.78 6.41 5.26 5.78 5.66 5.44 2.96 4.10 3.49 6.11 5.77 6 7 45 40 24 10 11 4 2 0 2 145 3.80 3.79 3.69 3.39 2.36 3.74 2.37 1.64 0.00 1.53 3.30 7 37 298 282 199 108 71 42 18 23 22 1.1 20.11 25.13 26.04 28.07 25.47 24.15 24.85 14.75 26.74 16.79 25.07 8 39 356 289 160 107 70 38 31 19 31 1.14 21.20 30.02 26.69 22.57 25.24 23.81 22.49 25.41 22.09 23.66 25.98 9 19 148 156 68 22 19 13 8 2 6 461 10.33 12.48 14.40 9.59 5.19 6.46 7.69 6.56 2.33 4.58 10.51 10 2 23 14 12 11 2 1 2 0 0 67 1.09 1.94 1.29 1.69 2.59 0.68 0.59 1.64 0.00 0.00 1.53 Total 184 1.186 1.083 709 424 294 169 122 86 131 4.388 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00

28

Appendix D:

Full table distribution educational attainment across birth order

Educational Birthorder attainment 1 2 3 4 5 6 7 8 9 ≥10 Total 1 7 5 1 2 2 3 0 1 0 1 22 0.47 0.42 0.13 0.52 0.87 2.04 0.00 1.92 0.00 2.44 0.50 2 40 47 23 23 20 12 3 3 1 1 173 2.71 3.91 3.10 5.93 8.73 8.16 3.61 5.77 3.45 2.44 3.94 3 158 135 104 67 25 33 18 15 6 9 570 10.70 11.24 14.04 17.27 10.92 22.45 21.69 28.85 20.69 21.95 12.99 4 151 102 75 55 27 18 14 6 3 6 457 10.22 8.49 10.12 14.18 11.79 12.24 16.87 11.54 10.34 14.63 10.41 5 95 70 37 17 17 7 5 3 1 1 253 6.43 5.83 4.99 4.38 7.42 4.76 6.02 5.77 3.45 2.44 5.77 6 54 45 22 16 4 2 0 1 0 1 145 3.66 3.75 2.97 4.12 1.75 1.36 0.00 1.92 0.00 2.44 3.30 7 350 310 217 92 55 31 18 7 11 9 1.1 23.70 25.81 29.28 23.71 24.02 21.09 21.69 13.46 37.93 21.95 25.07 8 409 329 183 77 60 33 20 13 5 11 1.14 27.69 27.39 24.70 19.85 26.20 22.45 24.10 25.00 17.24 26.83 25.98 9 181 139 70 36 18 7 4 2 2 2 461 12.25 11.57 9.45 9.28 7.86 4.76 4.82 3.85 6.90 4.88 10.51 ≥10 32 19 9 3 1 1 1 1 0 0 67 2.17 1.58 1.21 0.77 0.44 0.68 1.20 1.92 0.00 0.00 1.53 Total 1.477 1.201 741 388 229 147 83 52 29 41 4.388 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00

29

Appendix E:

Ordered Probit regression excluding only children

Spec[5] coeff SE Demographics age3439 -0.167 (0.047)*** age4045 -0.229 (0.049)*** age4650 -0.225 (0.054)*** age5155 -0.280 (0.069)*** asex -0.161 (0.033)*** nonwestern -0.253 (0.122)* Family attributes mum2125 0.153 (0.080) mum2630 0.243 (0.085)** mum3135 0.300 (0.094)*** mum3640 0.302 (0.106)** mum41up 0.263 (0.138) dad2125 0.243 (0.131) dad2630 0.309 (0.133)* dad3135 0.401 (0.138)** dad3640 0.466 (0.144)*** dad41up 0.639 (0.151)*** mum_deg 0.441 (0.068)*** dad_deg 0.648 (0.047)*** kidvsurban -0.014 (0.056) kidsurban 0.019 (0.050) kidmurban 0.033 (0.052) kidhurban 0.040 (0.050) kidreadlot 0.210 (0.037)*** kidreadfew 0.042 (0.045) workmum 0.007 (0.041) workdadfull 0.387 (0.084)*** workdadpart 0.381 (0.133)** famnorm 0.370 (0.159)* family composition famsizeall -0.077 (0.009)*** BO index -0.306 (0.050)*** N 4134 Pseudo R2 0.0486 *p < 0.05 **p < 0.01 ***p < 0.001