30-01-2017
The effect of family size on the future
income of a child
Marjet Welleweerd – 10649638 Supervisor: Pauline Rossi Amsterdam School of Economics
2 Statement of Originality
This document is written by Student Marjet Welleweerd, who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3 This study examines the relationship between the family size and the future monthly income of a child. In contrast to previous research, a more recent dataset is used in this study. The outcome of previous studies ranges from a negative correlation between family size and the future monthly income of a child or no significant effect when control variables are included. This study is in line with other studies. The outcome shows a negative relationship between family size and the future income of a child until the control variables age and gender are included.
1. Introduction
Educational attainment is an important aspect for policy makers as well as the future income of a child. There is a high correlation between education and productivity, and productivity and wages. One of the main policy goals is to improve the standard of living, which requires high wages, higher productivity and subsequently a higher educational attainment. So for policy makers it is important to know which factors affect the future income of a child. Therefore, it is meaningful to know what the effect of family size is on the future income of a child (Berger & Fisher, 2013).
In 1960 Gary Becker made an economic analysis of fertility. He did research on the factors which affects the quality of a child (Becker, 1960). In 1974 he expanded his theory together with Lewis and found out that there is a trade-off between the quality and quantity of children. There is only a certain amount of resources in a family, when the amount of children increase the resources have to be distributed to more children so the quality will go down (Becker & Lewis, 1973).
There have been a lot of empirical studies about the effect of family size on educational attainment. When the quality-quantity trade-off theory is confirmed, an explanation might be that greater family size may affect child outcomes negatively through resource dilution or because the average maturity level in the household is lower. When the quality-quantity trade-off theory is not confirmed, the reason can be that children stabilize marriages or decrease the probability that both parents work outside the home (Devereux, Black, & Salvanes, 2005). De Haan et al., and Booth et al., did find a negative causal relationship between the amount of children within a family and the educational attainment. Black et al., found a negative
4 correlation between family size and the educational attainment but they did not find a causal relationship between those variables.
In previous studies there has been research on the effect of family size on educational attainment. In this study the effect of family size on the future earnings of the child will be studied. The relationship between the future earnings of the child and the size of its family will be interesting for researchers as well. In this study the effect of family size on the future earnings of a child in the Netherlands will be examined. So far there has not been a research on this topic using a Dutch dataset. The effect will probably be the same as other European countries, such as Norway or Great Britain, with a comparable living standard and purchasing power. It will be interesting to assure this. This effect will be estimated with the ordinary least squares method in STATA. The data that has been used to examine the effect of family size on the future income of the child are taken from NKPS “Netherlands Kinship Panel Study”. This dataset contains information about 2832 randomly selected individuals within private households in the Netherlands, with a minimum age of 18 and a maximum age of 79. Residents of care-institutions, penitentiaries, homes for elderly and holiday homes are excluded from the sample frame. The data that will be used is from the fourth Wave, because this data is most recent and detailed. Another reason to why this research might be intriguing to researchers is that in previous research they used data prior to 2000. In this study data from 2015 is used.
The outcome of this study corresponds to the outcome of Black et al. (2005), there is a negative correlation between the family size and the future monthly income of a child until control variables are included. In this study the level of urbanization is first included. The negative effect of the amount of siblings on the monthly income in euros is still significantly different from zero. When the age and the gender of the observant are included, this effect vanished. The effect of the amount of siblings on the monthly income in euros is not significantly different from zero anymore.
In the next chapter, the literature review, you will read background information about previous research on this topic. I will elaborate on why this topic is relevant and what they did in previous research and their methods. I will synthesize all of the findings and include what is new in my research. In the following chapter, the method, I will describe how I am going to conduct my research and explain which dataset is used. After which I will describe which techniques I used
5 to deduce the results. In the following chapter, the results, I will show my results. Then I will give a short summary and conclusion and in the last chapter I will discuss my findings.
6 2. Literature review
In this chapter I will expand why this research is relevant and describe what has been done in previous research, which datasets are used and which method has been used. Finally, I will describe what is new about my research.
In 1960 Gary Becker made an economic analysis of fertility. He did analyze family decisions within an economic framework such as family size. In his paper he describes that a family must determine not only how many children it has but also the amount spent on them. Becker explains families has to make decisions like whether it should provide separate bedrooms to the children, send them to nursery school and private colleges, give them dance or music lessons, and so forth. He called more expensive children “higher quality” children (Becker, 1960).
In 1973 Becker and Lewis have extended the study on the interaction between the quantity and quality of children. After the study of Becker and Lewis, there has been many studies on this subject. Hanushek did also research on this topic and described that the trade-off between child quantity and quality consists because parents’ time and resources must be spread thinner with more children (Hanushek, 1992).
There have been a lot of empirical studies on the factors that affect the child outcomes. One of the most important factors is the family environment. For example, how many family members do you have? Are your parents married? Are they still together? Where did you grow up? How many siblings do you have? The family environment is widely believed as the primary component, but it is difficult to parcel this out into specific characteristics. One of the perceived inputs in the production of child quality is family size (Devereux, Black, & Salvanes, 2005).
The most important empirical study on this subject is the study by Sandra Black, Paul Devereux and Kjell Savanes “The More the Merrier? The Effect of Family Size and Birth Order on Children’s Education”. Black et al., were able to overcome many limitations of earlier research resulting from small sample sizes or limited information on children’s outcomes after the children have left home. In addition, they also have a plausible exogenous variation in family size (induced by the birth of twins) to identify the causal effect (Devereux, Black, & Salvanes, 2005). They used data from matched administrative files that cover the entire population of a developed country namely Norway, who were aged from 16 to 74 years old at some point during
7 the 1986 to 2000. Black et al., found a negative correlation between family size and children’s educational attainment. This result is in accordance with the quality-quantity trade-off theory of Gary Becker. But when they include indicators for birth order, the effect of family size reduces to almost zero. In this study the evidence suggests that family size itself has little impact on the quality of each child but more likely impacts only the marginal children through the effect of birth order.
Another important study on this topic is the research of Monique de Haan, Erik Plug and José Rosero. This study is extraordinary because the research is done in an undeveloped country. They researched the birth-order effects on human capital development in Ecuador using two datasets. The dataset that is used to examine the effect of birth order on child schooling and child labor was surveyed between 2001 and 2006. The dataset that is used to examine the effect of birth order on preschool cognition come from a recent survey between 2008 and 2010. Some tables are provided with the OLS outcomes of family size on school enrollment and child labor in different age groups in Ecuador. They tested a family size of three children and a family size of four children. The result is that the OLS estimator of four children is smaller than the OLS estimator of three children on school enrollment. This means that the probability of enrolling in school is smaller when a child lives in a family with four children than a child living in a family with three children. In the same table the OLS estimator is given with respect to child labor. The OLS estimator is bigger for the family with four children than the family with three children. This means that the probability of child labor is bigger for a family with four children (De Haan, Plug, & Rosero, 2014). This empirical study confirms the quality-quantity trade-off theory of Gary Becker.
Alison Booth and Hiau Joo Kee researched the effect of family size and birth order on educational attainment as well. They used the British Household Panel as data. The result showed that children from larger families have less education and the family size does not vanish when they control for birth order on educational attainment (Booth & Kee, 2009). So this result differs from the result of Black et al., they found that when they controlled for birth order the effect of family size vanished. So, the study of Booth and Kee does perfectly confirm the quality-quantity trade-off theory of Gary Becker.
The last study I want to refer to is done by Harry Anthony Patrinos and George Psacharopoulos. They did research on the family size, schooling and child labor in Peru. Their analysis is based
8 on the Peru 1991 Living Standard Survey. They found that having a greater number of younger siblings implies less schooling, less higher grades in school and more child labor (Patrinos & Psacharopoulos, 1997). So this research implies that the number of younger siblings has a negative effect on a child. So this research does not corresponds with the quality-quantity trade-off theory of Gary Becker. In this research only the amount of younger siblings do have a negative correlation with the quality of a child, not the total amount of siblings.
Gary Becker did suggest a causal relation between the quality and quantity of children within a family. De Haan et al., did find a causal relationship between the amount of children within a family and the educational attainment as well. They also found a causal relationship between the family size and child labor. Booth et al., did also find a causal relationship between those two variables. Black et al., found a negative correlation between family size and the educational attainment but they did not find a causal relationship between those variables. Patrinos and Psacharopoulos only found a negative correlation with the amount of younger siblings and the quality of a child.
In my results you will find the effect of family size on the later income of the child. Educational attainment and future income is correlated but it is interesting to discover if the effect of family size is the same on both variables. Next to this, in previous research they used datasets from foreign countries. In this study a Dutch dataset will be used. It is interesting if the results will be the same in all those different countries or that the results in the Netherlands will deviate. Furthermore, previous studies used datasets from before 2000, only the study of de Haan, Plug and Rosero used datasets from later on. In this study a recent dataset will be used. The survey is taken in 2015. It might be interesting if the results are still comparable with the older studies or maybe things changed during the years.
In this chapter you have read about the previous studies on this subject. I also described why my research on this topic is different than the studies they did before. In the next chapter, I will describe how I will conduct my research and which method I will use to get my results.
9 3. Method
In this chapter I will describe how I will conduct my research. You will read about the research design, the participants, the sample size, variables and the procedure.
Research Design
This research is a quantitative research. The Netherlands Kinship Panel Study is the Dutch participant in the Generations and Gender Programme (GGP). The GGP is a system of nationally comparative surveys and contextual databases, which aims to improve the knowledge base for policy making in countries of the United Nations Economic Commission for Europe (UNECE) (Netherlands Kinship Panel Study, 2009). The research questions revolve around the theme of solidarity, which is defined as ‘feelings of mutual affinity in family relationships and how these are expressed in behavioural terms’. Four waves of an extensive face-to-face interview have been conducted. Wave 1 in 2002-2004, wave 2 in 2006-2007, wave 3 in 2010-2011, wave 4 in 2014.
The NKPS has four special features that make it highly innovative. It is large (N=9500 at wave 1), it is a panel (prospective longitudinal design), it is multi-method (the data collection involves both structured interviews and in-depth open interviews) and it is multi-actor (the data are from individual respondents as well as from family members) (Netherlands Kinship Panel Study, 2009).
Participants
In this study the dataset of the NKPS fourth wave is used. The participant are members of the general population of the Netherlands aged between 18 to 79 at the time of Wave 1, excluding residents of institutions, nursing homes, old people’s homes and holiday homes and people without permanent residence status (Hogerbrugge, et al., 2014). The participant were first asked to fill in the survey by completing the questionnaire via web. Wave 1 had 8161 participants. Wave 4 only had 2832 participants. So 34.7% did drop out between wave 1 and 4. In the main sample of wave 1 our four largest migrant groups in the Netherlands were oversampled. The four largest migrant groups in the Netherlands are Turks, Moroccans, Surinamese and Antilleans. This Migrant sample was no longer part of wave 3 and 4 because of low to very low response rates. The overall response rate, after removing the oversampled migrant groups, was 65.2%. The main reason is that people refused to fill in another questionnaire (18.9%). Other
10 reasons were that the NKPS cannot reach the individuals (13.5%), the individuals were too ill (1.3%), deceased (0.9%) or moved outside the Netherlands (0.2%) (Hogerbrugge, et al., 2014).
Sample size
The sample size of the fourth wave is 2832 participants. 40% of these participants are male and 60% of these participants are female. In table 1 the age of the participants is shown. In this sample size the age groups 18-39 and 64-79 years old are under identified. The age group 40-64 years old is over identified. In table 2 the urbanization level of the places where the participants live is shown. The sample/population ratio is at every level almost one, so it is a representative sample of the Dutch population. In the most of the regressions the observation number is 1468 because one of the main variables which is used in this study is the income. The sample size decreased to 1468 because 51.8% of the individuals did not fill in their income.
Table 1: Distribution by age for men and women in the population and the realize Wave 4 sample. Note: Table 5.6 Distribution by age for men and women in the population and the realized Wave 4 sample. Reprinted from “Codebook of the Netherlands Kinship Panel Study, a multi-actor, multi-method panel study on solidarity in family relationships, Wave 4. NKPS Working paper No.13.” (Hogerbrugge, et al., 2014)
11 Table 2: Distribution by level of urbanization for the population and the realized wave 4 Sample. Note: Table 5.8 Distribution by level of urbanization for the population and the realized sample. Reprinted from: “Codebook of the Netherlands Kinship Panel Study, a multi- actor, multi-method panel study on solidarity in family relationships, Wave 4. NKPS Working paper No.13.” (Hogerbrugge, et al., 2014)
Variables
In this study the dependent variable is monthly income in euros. The main independent variable is the amount of siblings. A couple of control variables are used in this study.
The final model in this study is:
Y = α + β1*S + β2*Urb + β3*Urb2 + β4*Urb4 + β5*Urb5 + β6*M + β7*A + β8*A2 + e i..
Y is the dependent variable monthly income in euros. S, Urb, M and A are all independent variables which have probably an effect on the monthly income. S stands for the amount of siblings. Urb stands for the level of urbanization. The urbanization level is categorized into five levels and into four dummy variables. The M stands for male and is a dummy variable. The effect of family size on the future income of the child is also measured for the individuals where the place of residence is still the Netherlands.
Procedure
In this study the method of ordinary least squares will be used. With the use of OLS unknown parameters in a linear regression model will be estimated. The log-linear model is used as well to measure the effect of every variable on the monthly income in percentage.
In this chapter you have read how I will conduct my research. In the next chapter the results of this study are shown.
12 4. Results
In this chapter I will describe my findings. 4.1 General regression model.
Table 3: Summary amount of siblings and summary monthly income in euros.
The data from table 3 shows the summary of the amount of siblings and the monthly income in euros. The average monthly income in euros is 1468. The average amount of siblings is 3.07 which means the average family has 4 children. This seems quite high, but even when the outliers are removed the average amount of siblings is still 3. The data from table 4 shows us the summary of the age of the anchor during the interview. The average age of the anchors is 56 years old. This means that the average birth year of the anchors is 1958. During the sixties and seventies it was not irregular to have more than 4 children.
13 Table 5: Regression monthly income in euros on the amount of siblings.
The data from Table 5 show the results of the regression of the amount of siblings on income. The variable income is measured as monthly income in euros and the number of siblings is excluding yourself. So if you have a two-child family, your number of siblings is one. Our first dataset contained 2832 individuals. In table 5 the number of observations is only 1468. The number of observations decreased because the other 1364 individuals did not know their monthly income or they did not want to fill it in. The constant in the model is 2111,378, which means when you do not have siblings the average monthly income of an individual is 2111,378. The slope number is -34.765, so every additional sibling you have, declines your monthly income with 34,765 euros. The T-value is -2.31 and the P-value is 0.021 which means that the outcome is significantly different from 0. So we can conclude that there is an effect of the amount of siblings on monthly income. However, the result is not large because the income declines with only 1.6% when you have one additional sibling.
If we make the assumption that lower income families have more children because back in the days only educated people had knowledge of contraception and furthermore we assume that the 1364 individuals who did not fill in their income come from low-income families. Considering that they are ashamed of their income. The data in table 3 will change, the mean of siblings will be the same and the mean of income will be lower. If we regress this, the constant in the model would have been the same as in table 5. Although, the slope number would have been smaller than -34,765. This means that the average monthly income for an individual without siblings is still 2111,378 but with every additional sibling the monthly income declines with more than
14 34,765 euros. The result will still be significant however, the effect of an additional sibling would be larger than 1.6%.
Table 6: Regression of monthly income in euros on the amount of siblings (All missing observations of income are filled in with 0)
In table 6 the outcome of the regression of monthly income in euros on the amount of siblings is showed. The number of observations is back at the normal level (N=2832). For every individual who did not fill in their income, we assumed their income is 0. The average income is not the same as in table 5 as suggested. The average income is much lower than 2111 euros namely 1282 euros. The slopenumber is indeed much lower than -34.76, namely -77.2. The constant is not the same as in table 5 because some of the individuals who did not fill in their income, do not have any siblings. In table 7a/b the frequency table of the amount of siblings are shown. In table 7a the N=1468, only observants who filled in their income are shown, we see that 50 observants have zero siblings. In table 7b the N=2832 because all the observants are shown, we see that 107 observants have zero siblings. So, the constant is lower because there are 57 observants included with a monthly income of 0 euros and without siblings.
15 Table 7a/b: Frequency table of the amount of siblings in two datasets.
If we still follow the assumption that lower income families have more children although we assume that the 1364 individuals who did not fill their income come from high-income families. Considering that their income is so high that they don’t know the exact number, because they never have to worry about their budget. The constant in the model would have been higher because there are problably some individuals with a high income but without siblings. The slope number would have been smaller than -34,765. This means the average monthly income for an individual without siblings is higher than 2111.378 but with every additional sibling the monthly income declines with more than 34,765 euros. The result will still be significant however, the effect of an additional sibling would be larger than 1.6%
16 Table 8: Regression of monthly income in euros on the amount of siblings (All missing observations of income are filled in with 5000)
In table 8 the outcome of the regression of monthly income in euros on the amount of siblings is shown where the empty observations of income are filled in with 5000 euro. 5000 euros is chosen because this is much higher than the average income. In the table, we see that the constant is indeed higher than in our previous regression. We see that the slopenumber is much higher than -34.76 namely, 80.7. This is different than suspected. So we can conclude that our assumption that lower income families have more children is not true.
Table 9: Summary amount of siblings of the missing observations.
In table 9 the amount of siblings of the missing observations are shown. We see that the average amount of siblings is slightly higher than from the original summary. So, we cannot say that typically large poor families are missing out in the original regression or typically small rich families.
17 4.2 Place of residence
Now we need some variables to control for. The first one is the place of residence. We want to measure the effect of family size on the later income of the child in The Netherlands. So we have to be sure that every individual in our regression is still living in The Netherlands, because the purchasing power and level of income is not comparable with other countries than the Netherlands.
In Table 10 you see that we only have 1458 observations left because 10 individuals do not live in The Netherlands anymore. After this small change you see that the slope number raised a bit, now the slope number is -33.14 instead of -34,77. This means that the correlation between the monthly income and the foreign observations is positive. The correlation between the number of siblings and the foreign observations is negative. In this dataset people who moved abroad have a higher monthly income on average and have less siblings on average. In table 11 the summaries of the monthly income and amount of siblings are shown. In the first two tables the summary of the monthly income is shown. The average income of the foreign observant is 2947.4 and the average income of the observant living in the Netherlands is 2011. The average amount of siblings of the foreign observant is 1.69 and the average amount of siblings of the observant living in the Netherlands is 3.08. The foreign countries where 10 observants moved to are Greece, Germany, US, Canada, Switzerland, Belgium, South-Africa, Australia and Denmark.
Table 10: regression monthly income on amount of siblings but only if the place of residence of the observant is the Netherlands.
18 Table 11: Summaries income and amount of siblings when place of residence is the Netherlands and foreign.
4.3 Urbanization level
The place a person is born and raised is another factor for having a big or small family. CBS is telling us that families are bigger at places not urbanized (CBS, 2001). So this might be a good first variable to control for. CBS tells us the income is lower in places not urbanized (van den Brakel & Ament, 2010).
In our dataset the level of urbanization is measured in 5 different groups. 1 = “Very strongly urbanised” (>=2500 addresses/km2)
2= “Strongly urbanised” (1500-2500 addresses/km2)
3= “Moderately urbanised” (1000-1500 addresses/km2)
4= “Hardly urbanised” (500-1000 addresses/km2)
5= “Not urbanised” (<500 addresses/km2)
To use this data, there are 4 dummyvariables created. Category 3 “Moderately urbanised” (1000-1500 addresses/km2) is omitted and used as a base category in our regression model.
19 Table 12: Regression of monthly income in euros on the amount of siblings and urbanization level.
In table 12 you see the regression of monthly income in euros on the amount of siblings and urbanization level. Our N=1468, all the individuals who filled in their income are part of this regression. The constant in the model is comparable with our previous models. In this model the constant is 2115 euros and in our previous models 2100 and 2111 euros. The urbanization level categories are in accordance with the statement of CBS. CBS told us that income is lower in places not urbanized. In our regressionmodel you see that the slopenumber for variable “urb” is highly positive. The other 3 slopenumber of “urb2,urb4 and urb5” are all negative in order of slightly negative to highly negative. All slopenumbers except urb5 are not significantly different from zero. The P-value is 0.054, when a significancelevel of 10% is used, urb5 is significantly different from zero. So we can conclude that not every urbanization level has a significant effect on income. Living in an area not urbanized (urb5) has a negative effect on the income level. The level of urbanization does affect the level of income and therefore is a good control variable because we want to measure the real effect of the amount of siblings on the future income of a child. We want to remove the effects of the control variables from the equation. The slopenumber of the amount of siblings is equal to -30.83 this is slightly higher than our previous regressionmodels (-33.13, -34.76). This slopenumber is still significantly different from zero, because our t-value is -2.04 and our p-value is 0.041. But still the result is not large because your income declines with only 1.46%, when you have one additional sibling.
20 4.4 Age and gender
Age and gender are both exogenous variables which have definitely effect on the income of a person. Both variables are exogenous because there is no relation between gender or age and the amount of siblings. CBS is telling us that women are still earning less than men. In 2014 women did earn 10% less than men in the public sector and 20% less in the private sector (CBS, 2014). Gender is included as a dummy variable which is called male. Male is equal to one if the sex of the individual is male and zero if the sex of the individual is female. The relation between age and income is not linear but quadratic. This is why age squared is also included in the OLS regression. The income will increase until a certain age and will declines afterwards because people retire or going to work less hours a week.
Table 13: Regression of monthly income in euros on the amount of siblings and the exogenous variables: age, age2 and gender.
In table 13 the regression of monthly income in euros on the amount of siblings and age and gender is shown. The slope number of the amount of siblings is still negative, but it is less negative than in the previous regressions. The amount of siblings is not significantly different from zero anymore. The p-value is 14.9%. The control variables male, age and age2 are all significantly different from zero, so they do have an effect on the future monthly income of a child. The positive slope number for age and the negative slope number for age2 confirms the assumption that the relation between age and income is quadratic. The graph of the relationship between age and income is a parabola. In this model the top of the parabola is around the age
21 of 44. Furthermore, the positive slope number for male in the regression corresponds to the information of CBS about the wage differences between males and females.
4.5 All control variables
In order to get the most complete view of all factors which affect the later income of the child, all variables are put together in one regression. All control variables have been added: urbanization level, place of residence, gender, age and age2.
Table 14: Regression of monthly income in euros on the amount of siblings and all the control variables.
In table 14 the regression of monthly income in euros on the amount of siblings and all the control variables is shown. Our independent variable, the amount of siblings, raised to -16,16. In table 5, the outcome of the first regression is shown, the slope number was -34,76. So 53,5% of the effect of siblings on the monthly income is absorbed into other variables. The independent variable, the amount of siblings, is still not significantly different from zero. When the outcomes of table 5 and table 14 are compared, the adjusted R-squared raised from 0.3% to 16.12%. Thus, the goodness of fit of the model is highly improved.
22 Table 15: Regression of monthly income in euros on the amount of siblings and all the control variables with income expressed in logarithm.
In table 15 the regression of monthly income in euros on the amount of siblings and all the control variables is shown. Table 15 is different from table 14 because the income in table 15 is expressed in logarithm. The effect of every variable on the monthly income is shown in percentage. The average effect of one additional sibling on the monthly income is -0,345%. The average effect of gender on the average monthly income is high. The effect of gender is approximately e0.5078 = 1.6617 = 66.17% in advantage of the male gender. This outcome does not corresponds with the information of CBS. CBS told us that men earn 20% more than women in the private sector and 10% more than women in the public sector.
In this chapter you have read my results about the effect of family size on the future income of a child. In the next chapter you will read the conclusion of this study.
23 5. Conclusion.
In this study the effect of family size on the future income of the child is measured. There has been a lot of previous research on the effect of family size on educational attainment. In this study we want to measure if the effect of family size is the same on the future income of a child as on educational attainment. Besides that, in this study a Dutch dataset is used in contrast to previous research. A dataset from the NKPS is used in this study. The dataset is the fourth wave from 2014-2015. 2832 individuals answered a lot of questions about their family situation. To measure the effect of family size on the future income of a child, a couple of control variables are used. The control variables are the urbanization level, age and gender.
This study refers to the quality-quantity trade-off theory of Gary Becker. This theory suggests that there is a trade-off between the quality and quantity of children. There is only a certain amount of resources in a family, when the amount of children increase the resources have to be distributed to more children so the quality will go down (Becker & Lewis, 1973).
The main study at this subject is done by Black, Devereux & Salvanes. “The more the merrier: The effect of family size and birth order on children’s education.” In this study they found a negative correlation between family size and children’s educational attainment. But when they include indicators for birth order, the effect of family size reduces to almost zero.
The result of the first regression in this study is that there is a significant effect of one additional sibling on the monthly future income. This effect is small, the monthly future income declines with only 1.6% by one additional sibling. When the urbanization level is included as control variable, the effect is still significantly different from zero. The effect became lower namely 1.46%. When the gender and age are included, the effect is not significantly different from zero. The negative relationship between the family size and the future income of a child corresponds with the quality-quantity trade-off theory and the study of Black et al. But when we control for gender and age this negative relationship is not significantly different from zero anymore.
24 6. Discussion
In a subsequent study it might be interesting to examine why 1364 individuals did not fill in their income in the fourth wave. The sample size decreased from 2832 individuals to 1468 individuals. This is a very large drop out. In this study the data is completed by a very low or high income. But this data is not real, so in next study the data can be completed with real data.
Next to this, when the same regression analysis were made of the different waves, the outcomes were different through the years. There is a maximum of three years between the different waves, so the data must be corresponding with each other. For example the general regression of monthly income in euros on the amount of siblings gives these slope numbers from wave 1 till 4 (26.54, -21.7, -73.9 and -34.8). So the slope number in wave 1 is positive and the other three slope numbers are negative. It seems like these numbers differ too much from each other. The time difference between wave 1 and wave 3 is only 8 years. In 8 years, the income level or amount of siblings does not change that much. Besides that, in wave 1 the slope number is positive, how can this effect change that much during a couple of years?
Furthermore, the relationship between age and income has been analyzed. The top of the parabola is around the age of 44. This is not the expected age. In the Netherlands, normally the income grows until one retires (67 years old), but most people work less hours a week when they are 55+ years old. So the expectation of the age, when the income is the highest, is around 55 years old instead of 44 years old.
25 7. References
Becker, G. S. (1960). An economic analysis of fertility. In Demographic and economic change in
developed countries (pp. 209-240). Columbia University Press.
Becker, G. S., & Lewis, G. H. (1973). On the interaction between the Quantity and Quality of Children.
Journal of Political Economy, 81(2), 279-288.
Berger, N., & Fisher, P. (2013). A Well-Educated Workforce Is Key to State Prosperity. Economic policy intitute. Retrieved from http://www.epi.org/publication/states-education-productivity-growth-foundations/
Booth, A. L., & Kee, H. J. (2009). Birth order matters: the effect of family size and birth order on educational attainment. Journal of Population Economics, 22(2), 367-397.
CBS. (2001, november 9). Informatie over grote gezinnen. Retrieved from CBS:
https://www.cbs.nl/NR/rdonlyres/4150A580-F95E-432C-AC4F-A0501F625E2C/0/index1216.pdf
CBS. (2014). Gelijk loon voor gelijk werk? Den Haag: Centraal bureau voor de statistiek.
De Haan, M., Plug, E., & Rosero, J. (2014). Birth Order and Human Capital Development Evidence from Ecuador. Journal of Human Resources, 49(2), 359-392.
Devereux, P. J., Black, S. E., & Salvanes, K. G. (2005). The more the merrier? The effect of family size and birth order on children's education. Quarterly Journal of Economics., 120(2), 669-700. doi:10.1162/0033553053970179
Hanushek, E. A. (1992). The trade-off between child quantity and quality. Journal of political
economy, 84-117.
Hogerbrugge, M. J., de Hoon, S., Dykstra, P. A., Komter, A. E., Liefbroer, A. C., & Mulder, C. H. (2014).
Codebook of the Netherlands Kinship Panel Study, a multi-actor, multi-method panel study on solidarity in family relationships, Wave 4. NKPS Working paper No. 13. The Hague:
Netherlands Interdisciplinary Demographic Institute.
Loozen, S., Pool, M., & Harmsen, C. (2014). Bevolkingstrends 2014: In wat voor gezin worden kinderen
geboren? Den Haag: CBS.
Netherlands Kinship Panel Study. (2009). NKPS. Retrieved from http://www.nkps.nl/NKPSEN/nkps.htm
Patrinos, H. A., & Psacharopoulos, G. (1997). Family size, schooling and child labor in Peru -- An empiral analysis. Journal of population economics, 10(4), 387-405.
van den Brakel, M. , & Ament, P. (2010). Inkomensverschillen tussen en binnen gemeenten. CBS. Retrieved from https://www.cbs.nl/-/media/imported/documents/2010/35/2010-k3-v4-p07-art.pdf
