The estimated effect of parents’ schooling on children’s schooling : a study about the intergenerational transmission of education among three generations

(1)

Thesis Supervisor: Prof. Erik Plug

The Estimated Effect of Parents’ Schooling on Children’s

Schooling: A Study About the Intergenerational Transmission of

Education Among Three Generations

By Derk-Jan Verhaak, 10384340

Bachelor Thesis Economics

February 02, 2016

Abstract

This thesis analyses the intergenerational transmission of schooling among three generations. Next to that, it analyses whether there is a change in intergenerational transmission among these generations. The Netherlands Kinship Panel Study enables this thesis to make the analyses. The main findings are as follows: Firstly, mothers have on average more influence on children than fathers. Secondly, grandparents have an insignificant and smaller effect on children. Thirdly, the intergenerational transmission of

(2)

Der k - Ja n V er h aa k | T h e Es t i m at ed Ef f ec t B et w een p a r en t s ’ a n d c h i l d r en ’ s Ed u c a t i o n | 2

Table

of

Contents

Abstract 1

Table of Content 2

I. Introduction 3

II. Literature Review 4

A: Previous Literature 4 B: Hypotheses 7 III. Methodology 8 A. Data 8 B. Estimation Technique 9 C. Descriptive Statistics 11 IV. Results 14 A. Children’s Estimates 14 C. Parents’ Estimates 18

D. Changes in Intergenerational Transmission 18

V. Discussion 21

VI. Conclusion 22

(3)

I. Introduction

Inequality may be one of the bigger problems countries face. This is what the Organization for Economic Cooperation and Development (OECD) says (2015). In the 34 member countries, the richest 10% of the population earn 9.6 times the income of the poorest 10%. In their report, they find that this is harmful to long-term economic growth. When looking at inequality, the report (2015) states that

education plays an important role as a transmission mechanism. This last statement makes one wonder how much of this inequality of opportunity is responsible for inequality in countries.

Inequality of opportunity can be measured by using the intergenerational transmission of education. By estimating the effects of the intergenerational transmission, one can draw conclusions about the influence of the parents’ schooling on children’s schooling. Think of this in two extremes; one extreme is the situation where there is an absolute equality of opportunity, meaning the intergenerational transmission estimates are zero and there is no relationship between parents’ and children’s schooling. The other extreme is the situation where there is an absolute inequality of opportunity, meaning a perfect relationship between parents’ and children’s schooling. The estimated intergenerational transmission will be somewhere between these extremes and this thesis will research these estimated effects.

In previous studies, the focus was in general on five strategies to find these estimated effects: Twins (e.g. Behrman & Rosenzweig, 2002 and Bingley et al., 2009), adoptees (e.g. Plug, 2004 and

Holmlund et al., 2011), Instrumental Variable (IV) estimations (e.g. Black et al., 2005 and Piopiunik, 2014) and multiple generations (e.g. Lindahl et al., 2015 and Sauder, 2006). These strategies were used to segregate nature from nurture, enabling the researches to make more precise estimations about the intergenerational transmission. This is because nature reflects an inherent inequality of opportunity where nurture is more informative to consider when looking at inequality of opportunity, making it interesting for researches to segregate these channels.

The aim of this thesis is to analyze this effect in the Netherlands. Next to that, this thesis will attempt to answer the following three questions: What is the estimated effect of parents’ schooling on children’s schooling, what is the estimated effect of grandparents’ schooling on children’s schooling and whether the estimated effects of parents’ schooling have changed over the past generations. This will be done by using data from the Netherlands Kinship Panel Study (NKPS). This is a actor, multi-method panel study on family relationships. The NKPS is used before, usually to research sociographic questions (e.g. Kalmijn, 2006). No research using a Dutch dataset about the intergenerational

transmission of schooling is yet done. Studying this effect in the Netherlands could be particularly interesting cause of the changes which happened in the last decades. Multiple incentives have been implemented in the Netherlands to make studying more interesting for children with both relatively less educated and relatively more educated parents (think of free travel and a schol arship for students).

This paper will continue with section two, the Literature Review. This surveys the literature and empirical studies, discussing papers done in the recent years. The other sections will describe the data, regression models, and possible conclusions.

(4)

II. Literature Review

A. Previous Literature

There has been extensive research on estimating the causal effects of parents’ schooling on children’s schooling. In recent work, the focus was always on finding an estimation of these causal effects. To interpret their results correctly, researches need to account for genetic transmissions. To do so, they use four approaches: Identical twins, adoptees, IV and multiple generations. Firstly, the approach with identical twins will be discussed. Here, researches use identical twin parents with differences in followed education. These differences in education between the parents are used to identify the effects on their children. After discussing the identical twin approach, the approaches using adoptees will be discussed. The idea behind the use of adoptees is that there is no genetic transmission between adoptive parents and adopted children, making it easy to segregate nature from nurture. After this, IV is discussed. IV uses an educational reform, like a change in compulsory schooling laws. In this way, similar parents from a generation later had more compulsory schooling than the generation before them, resulting in differences in education. In this way, researches use reforms to instrument for parents’ education. At last, the use of multiple generations will be discussed. If it is assumed that genetic influences are fixed across generations, then differences in the intergenerational transmission across multiple generations reflect changes in inequality of opportunity. To make this more apprehensive: If it is assumed that the equality of

educational opportunities increases over time, then it is expected that the intergenerational transmission and thus the effect of parents’ schooling on children’s schooling falls across time.

Behrman & Rosenzweig (2002), Holmlund et al. (2011) and Bingley et al. (2009) are some of the researchers who use data on pairs of twins to come up with estimations of the causal effects on children’s schooling. The researches will be discussed respectively. Behrman & Rosenzweig (2002) find simple Ordinary Least Squares(OLS) results suggesting a positive and significant effect on a child’s education looking at parent’s education (with a coefficient of 0.33 on mother’s schooling and 0.47 on father’s schooling, both significant). However, these estimates are different once one looks at monozygotic (MZ) twin pairs. Looking at MZ twin pairs will eliminate the genetic influence the mother and father have, which is correlated with children’s IQ and thus schooling. Here, the coefficient of mother’s schooling turns negative and is almost significant. The influence of father’s education remains positive and

significant (with a coefficient of -0.25 on mother’s schooling and 0.36 on father’s schooling). Behrman & Rosenzweig (2002) conclude that, when looking at MZ twins, father’s schooling has a stronger effect on children’s schooling than does the mother’s schooling.

The somewhat same results are found in the larger samples of Holmlund et al. (2011). They also find positive and significant findings for the parents’ schooling on children’s schooling (with a coefficient of 0.28 on mother’s schooling and 0.23 on father’s schooling). The only difference is found when examining twins. Here, Holmund et al. find a positive and significant effect on both the father and the mother (with a coefficient of 0.25 on mother’s schooling and 0.21 on father’s schooling). However, it

(5)

must be taken into account that the twin estimates presented by Holmund et al. are based on a twin sample containing both MZ and dizygotic (DZ) twins. No correction has been made for possible measurement errors, meaning the estimates are probably too high. If that is considered, then the findings of Holmund et al. and Behrman & Rosenzweig are somewhat matching.

Amin et al. (2015) question the previous responses, saying that the previous twin-based studies did not have access to zygosity data to separate out DZ and MZ twins. They find that when separating DZ and MZ twins, mother’s schooling is equally important as father’s schooling (with a coefficient of 0.23 on mother’s schooling and 0.23 on father’s schooling).

The largest sample is from Bingley et al. (2009). They researched the effects with 2,975 twin mothers and 2,713 twin fathers from a Danish dataset on identical twins. They state there is no evidence of mother’s schooling having a significant effect on either boys or girls. Father’s schooling also has no effect on boy’s length of schooling but increases girl’s schooling by 1.2 months for each additional year.

Next to these papers there are other twin-based studies. They conclude that father’s schooling matters more than mother’s schooling (Antonovics and Goldberger, 2005; Hægeland et al., 2010 & Pronzato 2012). In sum, it seems that father’s schooling is equally or more important than mother’s schooling when looking at twins. Both seem to have positive effects as well.

Now that the papers regarding twins are discussed, this paper will continue with the second approach using adoptees. Plug (2004), Sacerdote (2007), Hægeland et al. (2010) and Holmund et al. (2011) are some of the researchers who studied adoptees and the effect of parents’ education. There are other papers, like Sacerdote (2000), but samples from these researchers are relatively small. Therefore the focus is on the other researchers and results from some of these other papers will be summarized at the end. Some papers look at adoptees to eliminate the genetic effect from parents to children. For example, Plug (2004) works with a subsample of 5,582 respondents with 16,481 children of whom 610 were adopted. Here, he finds a positive correlation between parent’s education and the education of their children. Plug finds, for own birth children, that the influence of the mother’s schooling is more important than the father’s schooling (with a coefficient of 0.54 on mother’s schooling and 0.39 on father’s schooling, both significant). After that, he looked at adopted children. This resulted in a positive and significant effect of both father’s and mother’s schooling on children’s schooling (with a coefficient of 0.28 on mother’s schooling and 0.27 on father’s schooling). As noticed, the coefficients are less for adoptees, showing the effect of genetic influences. Plug states there are a few limitations using this approach: Firstly, parents who adopt children are different from other parents, meaning the results may be biased and suffer from omitted variables. Secondly, adoptees are not always randomly assigned to their adoptive parents. This implies that a part of the parent’s schooling may, in fact, be caused by genetics (Plug, 2004).

Sacerdote (2007) solves part of this problem by using randomly assigned adoptees to distinguish the effect of nature and nurture between mother’s schooling and children’s schooling. He analyzed a sample of 1,051 non-adoptees and 1,256 adoptees from South Korea who were quasi-randomly assigned to American adoptive families. He finds a positive and significant effect of the mother’s education on the

(6)

children’s education (for adoptees with a coefficient of 0.09 on mother’s schooling and for the own offspring of 0.32 on mother’s schooling).

Also Björklund et al. (2006) try to control for the selective placement by using additional information on the adoptees’ biological parents. They researched both the biological parents’ schooling and adoptive parents’ schooling. Here, they find a significant and positive effect on the own birth children of 0.17 for the father and 0.16 for the mother. Looking at adopted children, they found two effects. The effect of the biological parents (0.11 for the father and 0.13 for the mother) and the effect of the adoptive parents (0.09 for the father and 0.02 for the mother). All the findings are positive and significant.

Observing these results, one could state that genetics (nature) seems to be more important than the nurture.

Holmund et al. (2011) take a different approach and compare Swedish-born adoptees (where their sample was taken from) and foreign-born adoptees. They find that when looking at Swedish-born adoptees an additional year of education of the father is associated with an increase of 0.10 more years for the child. For the mother, this equals 0.11 per year of education. When looking at adoptees born in foreign countries, Holmund et al. find an effect of 0.04 for both the mother and father. These findings are positive and significant at the 1-percent level (Holmund et al., 2011).

Other papers on adoptees (Dearden et al., 1997; Sacerdote, 2000) report positive and significant effects for the father’s years of schooling on the adopted son’s years of schooling. The results there found are not much smaller than the results found using fathers and their own birth sons. From here, they conclude that environmental factors are an important part of the transmission of years of schooling. In summary, if one examines adoptees, the following can be stated: Although the results do not always accord, it does seem like mothers have a bigger impact on the years of schooling than fathers do. Both usually have significant and positive effects on the children’s schooling.

Next to the twin-based and adoption based studies, there is a strategy based on IV. In this case, the instrument is usually defined as an educational reform, resulting in a change of the schooling system, and thus schooling years. By using a two-stage least squares model, the researchers estimate what effect such a change has on the children of the parents. For example, Black et al. (2005) use a reform in the Norwegian Primary School. Here, the compulsory education in the old schooling system was seven years. The time spent in compulsory education is nine years in the new system. Because of this, there were parents who experienced two extra years of compulsory schooling than other similar parents who were in school before the change. They find a small causal relationship between parents’ education and child’s education. The only exception is among mothers and sons: When the mother’s schooling increases, their sons get significant more education as well (Black et al., 2005) (with a coefficient of 0.03 for the father and 0.08 for the mother and the children’s education, measured in the IV estimates).

Similar results are found by Piopiunik (2014). He uses a compulsory schooling reform that was implemented in all West German states between 1946 and 1969. This reform changed the compulsory schooling from eight to nine years. The results of his research suggest that a causal relationship only

(7)

seems to exist between the mothers’ and sons’ education. In particular, fathers’ schooling is likely to have no causal effect on their children’s education, says Piopiunik (2014) (with a coefficient of 0.14 for mothers-sons, 0.01 for mothers-daughters, 0.02 for fathers-sons and -0.06 for fathers-daughters, measured in the IV estimates).

Holmlund et al. (2011) find the somewhat same results using a Swedish compulsory school reform, gradually implemented from 1949 through the early 1960s. The reform implied an extension of compulsory education by two years. They find small and statistically insignificant results for fathers and for mothers positive and statistically significant results (with a coefficient of 0.07 for mothers and 0.02 for fathers, measured in the IV estimates).

Other researchers find that, when using IV estimates, mother’s schooling affects daughter’s schooling but not son’s schooling (Chevalier, 2004 , Ermisch & Pronzato, 2011). To sum it up, one can say that the results are not always matching. IV estimates usually do not find significant effects. The only exception is found when observing the mother’s schooling on either the son or daughter.

Most studies about the intergenerational transmission of human capital are bound to two generations. Therefore, the effect of grandparents’ schooling on children’s schooling is researched less often. Lindahl et al. (2015) do research this by using a Swedish dataset containing information on lifetime earnings for three generations and educational attainment for four generations. They used this to estimate the effect of great-grandparents’, grandparents’ and parents’ schooling (generation 1, 2 & 3) on children’s schooling (generation 4). They find that both the parents and grandparents have a positive and significant effect on children’s schooling. The effect gets smaller when looking a generation further from the children but remains positive. They show that the persistence in educational attainment is stronger across four generations than they had expected (Lindahl et al., 2015).

Sauder (2006) also looks at this effect and uses three generations using data from the National Child Development Survey of the United Kingdom. His OLS results indicate that grandparents’ schooling is significantly positively related to the grandchildren’s’ schooling.

B. Hypotheses

Looking at the previous literature, the hypotheses of this research can be stated. There are two hypotheses that will be investigated in this paper. The hypotheses are as follows:

Hypothesis 1:

Parents’ years of schooling have a positive effect on children’s schooling.

Hypothesis 2:

Grandparents’ years of schooling have a positive effect on children’s schooling.

Hypothesis 3:

Grandparents’ years of schooling have a bigger effect on parent’s schooling than

parents’ years of schooling have on children’s schooling.

(8)

These hypotheses will be discussed at the results. The section is the Methodology, containing a data description, the estimation technique, and the descriptive statistics. Also, certain assumptions and found difficulties will be discussed.

III. Methodology

A. Data

For the empirical part of the thesis, a dataset is used. This dataset is obtained via the Netherlands Kinship Panel Study. This is a large-scale, nationally representative, multi-panel study on kinship in the Netherlands. Participants answer face-to-face questions revolving around the theme of solidarity and the relationship with the family. Education related questions are also included, of which the data will be used in this thesis. The researchers obtained the data in four waves; the first wave in 2002, the second wave in 2005, the third wave in 2010 and the fourth wave in 2014. Wave 1 contains 9,500 individuals, a number that decreases with every next wave. A reason for this could be the unwillingness of anchors to participate any further. The NKPS is representative of the Dutch adult population with Dutch nationality and covers the period from 1922 to 2002.

For this thesis, only wave 1 will be used. This is because the assumption is made that education does not change after a certain age. Furthermore, wave 1 has the largest N. This makes it, ceteris paribus, the best wave for this thesis. In this dataset, people were asked about the education of their parents, their children and themselves. Here, they could answer in eleven possible ways, varying from elementary school not finished (1) to post-graduate schooling like a Ph.D. (11). These answers were translated into the main variable, years of schooling, by using information from the

education guidelines from the Netherlands. The years of schooling are constructed in the following way: 8 years for primary school, 10 years for lower vocational school (LBO). The secondary school follows after primary school and is divided into three directions; lower general secondary (12 years), medium general secondary (13 years) and upper general secondary (14). Then follows intermediate vocational (15 years), higher vocational (17 years) and/or University or higher (18 years). There was a relatively small group of people who did post-graduate education. Due the sample size, this group has been put for 18 years as well. See Table 1 for more clarity.

(9)

B. Estimation Technique

Ordinary Least Squares (OLS) estimation is used to test if parents’ schooling, the independent variable, has a significant estimated effect on children’s schooling. In the first regression, the empirical model is estimated by regressing the children’s schooling on parents’ schooling. Also grandparents’ schooling will be included in this regression. This will result in a linear effect, estimating the effect of a year extra education of both the parents (and/or grandparents). The same is thereafter done for parents by regressing the parents’ schooling on grandparents’ schooling. In this way, conclusions can be drawn about the change in intergenerational transmission effects.

When one is looking at the years of schooling of a person, there are multiple variables influencing his capacities and decisions. Examples of this may vary from genes to neighborhood (Ermisch &

Francesconi, 2011). The usable information gathered by the Netherlands Kinship Panel Study is too small for this subject. The only data regarding education gathered is parents’ education and respondent’s education. There are probably more variables influencing the years of schooling for an individual. These variables will partly be translated into the error term. Because of this, the model suffers from omitted variable bias. This causes the OLS estimator to be biased and inconsistent. The direction of the bias is probably positive, because the observed effect of parents’ schooling is higher than the true value.

In this thesis, there are three models measured; the intergenerational transmission between children and parents, the intergenerational transmission between parents, grandparents, and the intergenerational transmission between children and grandparents. Firstly, the model between children and parents is shown. The schooling of child The schooling of child i is related to his/her mother’s schooling (𝑆_𝑗𝑚₎_{, father’s schooling}_(𝑆

𝑗𝑓) and other not measured influences / errors, resulting in the error

term (𝜀_𝑖𝑗𝑐₎

𝑆_𝑖𝑐 = 𝛼₀+ 𝛼₁𝑆_𝑗𝑚+ 𝛼𝑆_𝑗𝑓+ 𝜀_𝑖𝑗𝑐 ₍₁₎

After this, model two follows by adding the influences of the grandparents. This becomes the following model with the schooling of child i related to his/her mother’s schooling (𝑆_𝑗𝑚), father’s schooling(𝑆_𝑗𝑓), grandmother’s schooling(𝑆_𝑗𝑔𝑚), grandfather’s schooling (𝑆_𝑗𝑔𝑓) and other not measured influences/errors, resulting in the error term (𝑣_𝑖𝑗𝑐_{) :}

𝑆_𝑖𝑐 = 𝛽₀+ 𝛽₁𝑆_𝑗𝑚+ 𝛽₂𝑆_𝑗𝑓+ 𝛽₃𝑆_𝑗𝑔𝑚+ 𝛽₄𝑆_𝑗𝑔𝑓+ 𝑣_𝑖𝑗𝑐 (2)

Both model 1 and 2 are measured in multiple manners, which can be seen in tables 3 & 4. First, it is measured for the child, not looking at gender. After that, it is measured by son and daughter. The estimated effects on children are measured by using multiple combinations of parental influences. For example, it measures what the influences are when father and grandfather are taken together. See table 3 & 4 for clarification.

(10)

Der k - Ja n V er h aa k | T h e Es t i m at ed Ef f ec t B et w een p a r ent s’ a n d c h i ldr en ’ s Ed u c a t i o n | 10

When looking at the transmission between parents and grandparents, the following is shown: The schooling of parent i is related to his/her mother’s1_schooling_(𝑆

𝑗𝑔𝑚), father’s schooling (𝑆𝑗𝑔𝑓) and

other not measured influences/errors, resulting in the error term (𝜀_𝑖𝑗𝑝): 𝑆_𝑖𝑝 = 𝛽₀+ 𝛽₁𝑆_𝑗𝑔𝑚+ 𝛽₂𝑆_𝑗𝑔𝑓+ 𝑢_𝑖𝑗𝑝 (3)

These effects are again measured in multiple manners. See tables 5 & 6 for clarification. The last estimation is an observation of the difference between on the one side the intergenerational transmissions between generation one and two, and on the other side the intergenerational2_{transmission between}

generation two and three. This will be done by comparing the first value with the second value and drawing conclusions from it. From this, there can be conclusions drawn about possible changes in the intergenerational transmission over time. The estimation is shown by using hypotheses for every coefficient in the following way:

𝐻0 : 𝛼1𝑆𝑗𝑚= 𝛽1𝑆𝑗 𝑔𝑚 𝐻1 : 𝛼1𝑆𝑗𝑚< 𝛽1𝑆𝑗 𝑔𝑚 𝐻₀: 𝛼₂𝑆_𝑗𝑓= 𝛽₂𝑆_𝑗𝑔𝑓 𝐻1 : 𝛼2𝑆𝑗 𝑓 < 𝛽₂𝑆_𝑗𝑔𝑓 𝐻₀: 𝛼₁𝑆_𝑗𝑚_{+ 𝛼} 2𝑆𝑗 𝑓 = 𝛽₁𝑆_𝑗𝑔𝑚 + 𝛽₂𝑆_𝑗𝑔𝑓 𝐻₁: 𝛼₁𝑆_𝑗𝑚_{+ 𝛼} 2𝑆𝑗 𝑓 < 𝛽₁𝑆_𝑗𝑔𝑚 + 𝛽₂𝑆_𝑗𝑔𝑓

A t-test can be used to determine if the coefficients are significantly different from each other and will therefore be used to estimate the effect of gender and generation on the estimated effects of schooling. In this way, conclusions can be drawn whether the coefficients on for example daughters are the same as the coefficients on sons. An example will show how the method works: First, a dummy variable is added which is coded equal to 1 if gender is female and 0 if gender is male. Then, a variable is added which is the product of the parents’ education and the dummy variable. After that, the regression is made with the added variables and hypotheses in the following form are tested:

𝐻₀: 𝛼_{𝐺𝑒𝑛𝑑𝑒𝑟} = 0 𝐻₁: 𝛼_{𝐺𝑒𝑛𝑑𝑒𝑟} > 0

If the t-statistic is significant, then it means that 𝛼𝐺𝑒𝑛𝑑𝑒𝑟 is bigger than 0 and then gender has an influence on the estimated effect on children’s schooling.

1_{Note that here, when looking at the model for parents, by mother is actually meant the grandmother.}

2_{Becker & Lewis (1973) and Becker & Tomes (1976) have replaced the Malthusian growth model (exponential growth model)} with a theoretical framework. This framework sees the number of children and parental investment per child as a choice. In this way, they filtered the presumption that children are a normal good. The models of Becker and Lewis (1973) show that there is a negative causal relation between child quantity and parental investment. By using the model of Becker & Tomes, it can be stated that the estimations found between parents and children are in fact causal.

(11)

C. Descriptive Statistics

Before the descriptive statistics are precisely described, it will be discussed how the data was manipulated before regressing it. Before certain restrictions were made, relevant information was found for a total of 8,147 persons. This data still contained information not usable for this thesis. Therefore, the dataset is restricted in a couple of ways to make sure the data fits best.

- Age restriction.

To control for anchors and children still being in school, the sample is restricted to those respondents that are at least 25 years old when the data was obtained. This is done to take censoring into account. Better educated parents are relatively more likely to postpone having children, making their children on average younger. This causes their children’s average education to be less. (2854 observations deleted)

- Missing data

There are people who don’t know their parents’, children’s or partner’s education, resulting in missing and thus not usable data. (2175 observations deleted)

- Errors in data

There were some errors in the data. People filled in a not possible answer, which would sometimes create large outliers. If these errors were kept in the dataset, the information attained would not be correct. (29 observations deleted)

After the restrictions are made, usable information is found for a total of 3,089 persons, resulting in information about grandparents, parents and children. This results in information about three

generations who are biologically related to each other. Table 2 describes the descriptive statistics for all three generations. Columns 1 and 2 describe the grandfathers and grandmothers. These were respectively born in 1916 and 1925 and have 10.91 and 10.17 years of schooling. Columns 3 and 4 show the second generation, divided by fathers and mothers (born in respectively 1949 and 1948). They respectively acquired 14.10 and 13.22 years of schooling. The last two columns describe the third generation, divided by son and daughter. The columns for son and daughter are actually the first and second children. It is chosen to put them together in one group. Not only were the differences very small, this choice also makes the total number of observations bigger. They are, on average, born in 1966 and have a mean of 15.43 and 15.42 years of schooling. What immediately stands out is that the mean of years of schooling for generation 1 (grandparents) is less than the mean of years of schooling for generation 2 (parents). This is explainable by three reasons: Firstly, compulsory schooling was extended between 1969 and 1975,

(12)

meaning children from later generations would have to go to school longer. Secondly, the generation of parents generally found education more important than the generation of grandparents. Thirdly, education was more accessible to the average person, meaning it was easier for the average person to follow a longer education. When looking further, notice that this trend continues for the third generation, where children have a higher mean than the parents. Also, the standard deviation is among children the smallest across the generations, meaning the data for children is closer to the mean.

Lastly, in figure 1 the education distributions of the three generations are plotted. This is done because it gives a better perspective of the data and changes per generation. For children, only one table is shown because the tables for boys and girls show no visible differences. As one could observe, the years of schooling get higher for every generation. Especially between generation 1 (grandparents) and generation 2 (parents), big differences are visible. Where approximately 65 percent of the grandparents has ten- or fewer years of schooling, this is approximately 25 percent for the parents. For generation 3 (children) this amount is even lower than 10 percent. That makes it clear that this trend continues when looking across all three generations. Also, many more children have attained eighteen or more years of education than their parents and grandparents have. Approximately 20 percent of the children has done eighteen years of schooling or more. For parents, this amount is almost 10 percent and for grandparents this is equal to roughly 4 percent.

(13)

Der k - Ja n V er h a a k | T h e Es t i m a t ed Ef f ec t B et w een p a r en t s ’ a n d c h i l d r en ’ s Ed u c a t i o n | 13

TABLE 2. DESCRIPTIVE STATISTICS

Generation 1 (Grandparent) Generation 2 (Parents) Generation 3 (Children)

Variables Grandfather Grandmother Father Mother Son Daughter

(1) (2) (3) (4) (5) (6) Years of Schooling 10.91a _9.88 _14.10 _13.22 _15.42 _15.43 (3.62b₎ _(2.91) _(3.32) _(3.36) _(2.50) _(2.44) [5,18c_] _[5,18] _{[5,18] [5,18]} _[8,18] _[8,18] Year of Birth 1916.46 1919.25 1949.12 1948.39 1966.30 1965.83 (17.59) (17.35) (15.46) (15.56) (7.43) (7.49) [1868,1959] [1875, 1977] [1923, 1978] [1923, 1978] [1941, 1978] [1938, 1978] Number of Observations 3084 3091 1263 1826 1740 1746

a _{The first figure in each cell is the mean of the variable} b _{Standard deviations are in parentheses and in italic.}

c _{The figures in square brackets are the minimum and maximum values.} All variables are corrected for age.

(14)

IV. Results

Results will be presented using OLS estimations. As already is discussed, there are six different types of models. Three for measuring the estimated effects on children’s education (Child, son and daughter) and three for measuring the estimated effects on parent’s education (parent, father, mother). At last, also the changes in intergenerational transmission will be discussed. Firstly, this thesis discusses the estimated effects on child’s education.

A. Children’s estimates

The estimated effects on children’s education are measured using parents’ and grandparents’ education (for clarification, see table 3 and 4.). All estimations are controlled for the age of the child and in table 4 the estimations are also controlled for gender. These particular effects are not reported in the tables. Table 3 shows that the estimate for the father is 0.256 and for the mother is 0.304. This means that for every year of schooling the father attains extra, the child generates 0.256 extra. This is in line with earlier results and with the hypothesis stated (e.g. Holmlund et al (2011), Björklund et al. (2006) and Bingley et al. (2009)). Notice that the mother has more influence than the father. This is in line with the expectations: The mother is, on average, more around the children. In that way, she can influence the child relatively more, making her possible intergenerational transmission higher than the father can. These estimation effects change when both the mother and father are included in the model. Here, the

estimations change to 0.166 for the father and 0.203 for the mother. At last; the aggregated3_estimated

effect of the parents on children is 0.275. All the estimations are significant at the 1-percent level. When looking at sons and daughters separately (Table 4), the results are somewhat the same. What immediately stands out is that the father has relatively more influence on the son than he has on the daughter for every regression done. For the mother, this is not the case with either the daughter or son. The mother does have more influence when looking at the separate effects on son or daughter. When the parents are taken together to look at estimates on the son, the mother’s and father’s influence diminishes. The estimate of the mother on the son diminishes so much (from 0.307 to 0.164) that she has a smaller estimated effect than the father (from 0.266 to 0.252). This is a remarkable and awkward finding, difficult to explain using economic theory.

Grandparents have much smaller effects on the children’s schooling. Furthermore, all results are insignificant. This is somewhat in line with the expectations. Grandparents are generally less important and this is also what Sauder (2006) and Lindahl et al. (2014) find. The only difference is that they do find some significant results as well for grandparents. The fact that no significant results are found could be due to the dataset or coincidence. On overall, the effect of parents diminishes slightly when introducing grandparents. For example, it is seen that when the grandfather is included, the father goes from an estimated effect of 0.166 to 0.165. This happens as a result of an added explainable variable. The effects

3_{With aggregated, the fourth column in table three is meant. Here, the effect of both parents is shown, when not separating by} gender.

(15)

of grandparents, however, remain very small. So one should not draw too big conclusions out of it. In sum, the results of children’s estimates are in line with what the expectations. At the bottom of Table 4, the t-statistic is mentioned. Here, it is controlled for gender. The t-statistic found is equal to 1.57 (0.116) meaning no significant effect is found. This means that the gender of son and daughter does not play a role when looking at the estimated effect of parents’ schooling. This is something that could be expected. When looking at figure 1, it is seen that the distributions are equal for daughters and sons. Also in Table 1 (Descriptive Statistics) it is seen that sons and daughters have almost the same education levels. See table three and four for further clarification.

(16)

TABLE 3

ESTIMATES OF INTERGENERATIONAL EFFECTS OF SCHOOLING DEPENDANT VARIABLE: YEARS OF SCHOOLING CHILD

Variables Intergenerational Effects on Years of Schooling of the Child

Education Father(1) 0.256*** 0.166*** 0.254*** 0.165*** (0.023) (0.027) (0.026) (0.029) Education Mother(2) 0.304*** 0.203*** 0.283*** 0.202*** (0.022) (0.032) (0.024) (0.033) Education Father 0.275*** 0.256*** + Mother(3) (0.015) (0.018) Education Grandfather(4) 0.009 0.003 (0.028) (0.032) Education Grandmother(5) 0.068 -0.001 (0.030) (0.043) Education Grandfather 0.042* + Grandmother(6) (0.018) N 1494 2288 2920 3782 1490 2286 2910 3782

Notes: Each estimate represents the coefficient from a different regression. Every estimate is controlled for age. Panel A estimates for the first child and panel B estimates for the second child. Father + Mother and Grandfather + Grandmother refers to the sum of either father’s and mother’s schooling or grandfather’s and grandmother’s schooling.

The numbers in brackets and italic are the standard errors.

(17)

TABLE 4

ESTIMATES OF INTERGENERATIONAL EFFECTS OF SCHOOLING DEPENDANT VARIABLE: YEARS OF SCHOOLING CHILD, SORTED BY GENDER Variables Intergenerational Effects on Years of Schooling of the Son

Education Father(1) 0.266*** 0.252*** 0.272*** 0.211*** (0.032) (0.044) (0.037) (0.041) Education Mother(2) 0.307*** 0.164** 0.282*** 0.185** (0.034) (0.038) (0.033) (0.045) Education Father 0.283*** 0.270*** + Mother(3) (0.023) (0.025) Education Grandfather(4) -0.008 -0.013 (0.038) (0.042) Education Grandmother(5) 0.095 -0.005 (0.045)** (0.058) Education Grandfather 0.030 + Grandmother(6) (0.027) N 744 1130 1457 1891 758 1128 1453 1884

Intergenerational Effects on Years of Schooling of the Daughter

Education Father(1) 0.245*** 0.199*** 0.235*** 0.193*** (0.034) (0.040) (0.037) (0.042) Education Mother(2) 0.299*** 0.152*** 0.285*** 0.146*** (0.030) (0.045) (0.033) (0.048) Education Father 0.266*** 0.243*** + Mother(3) (0.022) (0.025) Education Grandfather(4) 0.025 0.019 (0.041) (0.049) Education Grandmother(5) 0.042 -0.001 (0.040) (0.064) Education Grandfather 0.052 + Grandmother (6) (0.025)** N 750 1158 1463 1891 732 1158 1457 1898

Notes: Each estimate represents the coefficient from a different regression. Every estimate is controlled for age. The numbers in brackets and italic are the standard errors. *** significant at 1-percent level, ** significant at 5-percent level, * significant at 10-percent level. t-statistic (Son & Daughter) = 1.57 (0.116)

(18)

B. Parents’ estimates

In table 5 and 6, the intergenerational transmission between parents and grandparents is shown. Firstly, table 5 will be discussed, estimating the effects on parents. Table 6 will be looked upon after, discussing the effects on father and mother.

It is seen that grandmothers have relatively more influence on the parent than grandfathers. The same effect occurs observing the estimated effects on children’s schooling. This is again a logical finding. When observing the combined effects of mother and father, the results show that the grandfather suddenly has more influence than the grandmother. This is something also found when looking at the estimated effects on sons.

Examining the father and mother separately, it is seen that the grandmother has more influence on both the father and mother. This effect diminishes again when looking at the combined estimated effects of grandfather and grandmother. Where this only happened once when looking at the children’s estimates, it happens for every regression when looking at the parent’s estimates. Why this is happening is difficult to explain. The only possible explanation, which is still farfetched, is that grandmothers did not have the same educational access as grandfathers. This is visible using figure 1 on page 12. Because of this fact, it could be that the grandfathers have more influence than the grandmothers when looking at the schooling of children. The found t-statistic is 1.65 for parents with a corresponding p-value of 0.099, meaning it is significant for the 10% level. For the 10% level, this means that gender does matter, looking at the estimated effects of grandparents’ schooling on parents’ schooling.

C. Changes in Intergenerational Transmission

Now that both the children’s and parents’ estimates are discussed, the results found will be compared. The main difference found is that parents’ estimates are higher for almost every regression than children’s estimates. This corresponds with the third hypothesis. The only exception occurs when looking at the combined effect of grandparents and grandmothers. Here, the grandmother gets a small influence on the parent, smaller than when looking at children. Why grandparents seem to have more influence on parents than parents have on children, is explainable using figure 1 on page 12 and an example. Figure 1 shows that the grandparents and parents have, on average, fewer years of schooling than children do. If by example, all the children would have 18 years of schooling, there would be no estimated effects of parents because there are no differences measured in the dependent variable,

children’s schooling. The same as the example above happens in a less extreme way for children from this dataset. Their average education is higher and the educational opportunities are on average better, making the estimated effects of parents smaller. The found t-statistic is equal to 2.71 with a corresponding p-value of 0.007, meaning the found t-statistic is significant for the 1% level. This means generations do have an effect, when looking at the estimated coefficients on schooling. This is a logical result. As is seen, the intergenerational transmission between grandparents and parents is found to be higher than between parents and children. See tables 5 & 6 for clarification.

(19)

TABLE 5

ESTIMATES OF INTERGENERATIONAL EFFECTS OF SCHOOLING DEPENDANT VARIABLE: YEARS OF SCHOOLING PARENT

Variables Intergenerational Effects on Years of Schooling of the Parent

Education Grandfather(1) 0.424*** 0.297*** (0.015) (0.018) Education Grandmother(2) 0.487*** 0.272*** (0.019) (0.022) Education Grandfather 0.474*** +Grandmother(3) (0.015) N 3080 3087 3079 3088

Notes: Each estimate represents the coefficient from a different regression. Every estimate is controlled for age. Panel A estimates for the first child and panel B estimates for the second child. Father + Mother and Grandfather + Grandmother refers to the sum of either father’s and mother’s schooling or grandfather’s and grandmother’s schooling.

(20)

TABLE 6

ESTIMATES OF INTERGENERATIONAL EFFECTS OF SCHOOLING DEPENDANT VARIABLE: YEARS OF SCHOOLING CHILD, SORTED BY GENDER Variables Intergenerational Effects on Years of Schooling of the Father

Education Grandfather(4) 0.405*** 0.312*** (0.024) (0.029) Education Grandmother(5) 0.418*** 0.184*** (0.030) (0.036) Education Grandfather 0.447 + Grandmother(6) (0.023) N 1259 1262 1259 1262

Intergenerational Effects on Years of Schooling of the Mother

Education Grandfather(4) 0.439*** 0.287*** (0.019) (0.022) Education Grandmother(5) 0.536*** 0.331*** (0.024) (0.028) Education Grandfather 0.497 + Grandmother(6) (0.019) N 1821 1825 1821 1826

Notes: Each estimate represents the coefficient from a different regression. Every estimate is controlled for age. The numbers in brackets and italic are the standard errors.

(21)

V. Discussion

The analysis of this thesis shows that parents’ years of schooling do in fact have a significant estimated effect on children’s schooling. Interesting and expected conclusions supporting the literature were found but also remarkable results, which were not expected.

Looking at the analyses done, the biggest problem faced was the size of the data. One of the complications caused by the size of the data is the missing data on twins, adoptees and IV-estimators. Twins and adoptees could not be discussed because twins were not included in the dataset and the sample size of adoptees was very small (42 adoptees in the whole dataset without corrections made). Therefore, no conclusions could be made about the nature and nurture of the intergenerational transmission. Also, IV-estimation could not fully be done because the new compulsory schooling law in the Netherlands was not applied in one step. In 1968, the compulsory schooling was changed to nine years and in 1975 it was changed to ten years, leading to different compulsory schooling systems in a short period of time. In retrospect, it could have been researched for parents before 1968 and after 1975. In that way,

IV-estimates could have been researched. This is something that could be done in a follow-up research using the same dataset.

Next to data size, a concern that should be mentioned is omitted variable bias. Educational attainment is influenced by many factors and parental education is just one of the influences. Although it is probably one of the biggest influences, other factors still influence the educational attainment of an individual. It is possible that the model did not gather all information from the dataset that has influence on educational attainment. Also, there could be factors influencing the educational attainment that were not included in the dataset at all. Lastly, some influences are difficult to interpret but do influence an individual. Examples are the neighborhood someone grows up in. This information was included in the dataset but is hard to interpret. When is a neighborhood ‘’bad’’ for an individual’s educational attainment and when is a neighborhood ‘’good’’? These are questions that could be investigated in future research, researching the influences of certain neighborhoods’ influences on educational transmission. This is already done but not yet with the NKPS dataset (e.g. Sykes & Kuyper, 2009 and Ermisch & Francesconi, 2011). Finally, for other future research, one could think of researching the estimated effects for twins or adoptees if such a dataset comes available in the Netherlands. In that way, empirical evidence could be added to the differences between nurture and nature. It would be especially interesting because most of the papers investigating this come from the Scandinavian countries. A research with a dataset from the Netherlands could in this way contribute to current findings. Also with this dataset, IV-strategy

estimations could be done as mentioned above. If follow-up research would be done and a bigger dataset would be available, the sample groups should also be adjusted in such a way that they look alike even more.

(22)

VI. Conclusion

This paper has analysed the intergenerational transmission using three generations from the Netherlands. It linked grandparents’, parents’ and children’s education by adjusting educational

attainment into years of schooling. These analyses could be done be using a dataset from the Netherlands Kinship Panel Study. Next to the analysis of the intergenerational transmission between parents and children, this thesis took a look at the changes in intergenerational transmission over the past generations. In this way, conclusions were drawn about the development of the influences of parents’ schooling on children’s schooling.

The main conclusion of this thesis is that parents have a significant and positive effect on the children’s schooling, which corresponds with the main hypothesis. When looking more closely, it is seen that mother’s schooling is more important than father’s schooling for children’s schooling. This is a finding that was expected. It is found that mother’s education is more important for every regression except when looking at the combined estimated effect on the son’s schooling. Then the father suddenly has more influence than the mother. This is a finding not corresponding with previous literature and is hard to explain using economic theory. The t-statistic suggests that there is no difference between sons and daughters when looking at the estimated effects.

Grandparents enter the OLS on children with insignificant and very small results. When looking at the grandparents’ estimated effects on parents, again positive and significant effects are examined. The grandmother has more influence on the parents when looking at the separate effects. When combining the effects, the same happens as what happened when looking at the combined estimated effect on the son: The grandfather suddenly has more influence than the mother. This applies to every regression of the grandparents’ effect on parents. The t-statistic suggests there is a difference for the 10% level between fathers and mothers when looking at the estimated effects.

Comparing the estimated effects on parents and on children, it is seen that the estimated effects on parents are bigger for every regression. This corresponds to the hypothesis stated. The t-statistic suggests also that generations do matter when looking at the estimated effects. Found results were significant for the 1% level.

Overall, the regressions were relatively successful and have added to previous literature in providing Dutch empirical evidence. Also, the found differences between the intergenerational transmission over the generations show that expectations are indeed true: The intergenerational

transmission used to be higher than it is now. Two new hypotheses could carefully be drawn out of this: Firstly, parents’ schooling lose a part of its influences, although still being an important factor to

children’s schooling. Secondly, the educational opportunity in the Netherlands seems to be more balanced than it used to be, creating more equal educational opportunities for children.

(23)

References

Amin, V., Lundborg, P., Rooth, D. ‘’ The intergenerational transmission of schooling: Are mothers really less important than fathers?’’ Economics of Education Review, 2015, 47: 100-117.

Angrist, J., Lavy V. & Schlosser A. ‘’New Evidence on the Causal Link between the Quantity and Quality of Children.’’ Central bureau of statistics, July 2005.

Antonovics, K. & Goldberger, A. ‘’Does Increasing Women’s Schooling Raise the Schooling of the Next Generation? Comment’’ American Economic Review, 2005, 95(5): 1738-1744

Becker, Gary S., & Nigel Tomes. “An Equilibrium Theory of the Distribution of Income and Intergenerational Mobility.” Journal of Political Economy, 1979, 87(6): 1153–89

Becker, G. & H. G. Lewis, “On the Interaction between the Quantity and Quality of Children” Journal of Political Economy 81(2) part 2, 1973, pp. S279-S288.

Behrman, Jere R. & Mark R. Rosenzweig. “Does Increasing Women’s Schooling Raise the Schooling of the Next Generation?” American Economic Review, 2002, 91(1), pp 323-334.

Bingley, P., Kaare C. & Vibeke M., “Parental Schooling and Child Development: Learning from Twin Parents.” Danish National Centre for Social Research Social Policy and Welfare Services Working Paper , July, 2009.

Björklund, A., Lindahl M. & Plug E., “The Origins of Intergenerational Associations: Lessons from Swedish Adoption Data.” Quarterly Journal of Economics, 2006, 121(3): 999–1028.

Black S., Devereux P. & Salvanes K. “Why the Apple Doesn’t Fall Far: Understanding Intergenerational Transmission of Human Capital.” American Economic Review, 2005, 95(1): 437–49.

Chevalier, A., Denny, K. & McMahon, D. ‘’A multi-country study of inter-generational educational mobility.’’ In P. Dolton, R. Asplundh, & E. Barth (Eds.), ‘’Education and inequality across Europe. London: Edward Elgar’’, 2009.

Ermisch, J. & Pronzato, C.’’ Causal effect of parents’ education on children’s education.’’ In T.

Smeeding, R. Erikson, & M. Jänti (Eds.), ‘’Persistence, privilege, and parenting. The comparative study of intergenerational mobility.’’ Russell Sage Foundation, 2011.

Ermisch, J. & Francesconi, M. ‘’The Effect of Parental Employment On Child Schooling.’’ Journal of Applied Econometrics, 2013, 28 (1), pp. 987-1013.

(24)

Hægeland, T., Kirkebøen L.J., Raaum O., Kirkebøen, L.J. & Salvanes K.G. ‘’Why children of college graduates outperform their schoolmates: A study of cousins and adoptees.’’ IZA discussion paper no. 5369, 2010.

Hazan, M. & Berdugo B., “Child Labor, Fertility and Economic Growth” The Economic Journal, 112(October), 2002, pp. 810-828

Kalmijn, M. “Educational Inequality and Extended Family Relationships: Influences on Contact and Proximity.” European Sociological Review, 2006, 22(1): pp. 1-16.

Lindahl M., Palme M., Sandgren Massih S. & Sjögren A., ‘’A Testof the Becker-Tomes Model of Human Capital Transmission Using Microdata on Four Generations.’’, 2014, 8(1), pp. 80-96.

Lindahl M., Palme M., Sandgren Massih S. & Sjögren A., ‘’Long-Term Intergenerational Persistence of Human Capital: An Empirical Analysis of Four Generations’’, 2015, 50(1).

Moav, O., “Cheap Children and the Persistence of Poverty,” The Economic Journal, 115 (January), 2005, pp.

88-110.

Organization for Economic Co-operation and Development (OECD), 2015. ‘’In It Together: Why Less Inequality Benefits All.’’

Plug, E. “Estimating the Effect of Mother’s Schooling on Children’s Schooling Using a Sample of Adoptees.” American Economic Review, 2004, 94(1): pp. 358–368.

Plug, E. & Vijverberg W. “Schooling, Family Background, and Adoption: Is It Nature or Is It Nurture?” Journal of Political Economy, 2003, 111(3): pp. 611–641

Piopiunik, M. ‘’Intergenerational transmission of education and mediating channels: Evidence from compulsory schooling reforms in Germany.’’ The Scandinavian Journal of Economics, 2014, 116, pp. 878–907.

Pronzato, C. ‘’An examination of paternal and maternal intergenerational transmission of schooling.’’ Journal of Population Economics, 2012, 25, pp. 591–608.

Holmlund, H., Lindahl, M. & Plug, E. ‘’The causal effect of parents’ schooling on children’s schooling: A comparison of estimation methods.’’ Journal of Economic Literature, 2011, 49, pp. 615–651.

Sauder, Ulrich. “Education Transmission across Three Generations - New Evidence from NCDS Data.” The university of Warwick., 2006, Unpublished.

Sacerdote, Bruce. “The Nature and Nurture of Economic Outcomes.” National Bureau of Economic Research Working Paper 7949, 2000.

(25)

Sacerdote, Bruce. “How Large Are the Effects from Changes in Family Environment? A Study of Korean American Adoptees.” Quarterly Journal of Economics, 2007, 122(1): pp. 119–57.

Sykes, B. & Kuyper H., ‘’Neighborhood effects on youth educational achievement in the Netherlands: can effects be identified and do they vary by student background characteristics?’’ Environment and Planning A, 2009, 41, 2417-2436.

The World Wealth And Income Database, 2013, The wealth gap – inequality in numbers. http://www.wid.world/#Database

Vikesh A., Petter L. & Dan-Olof R.. ‘’The intergenerational transmission of schooling: Are mothers really less important than fathers?’’ Economics of Education Review, 2015, 47, pp. 100-117.