"The Group-specific Effect of Monitoring Compliance with the Dutch Compulsory Education Act”

(1)

Acknowledgement: Ministry of Education, Culture and Sciences and Government of the Netherlands

Compliance with the Dutch Compulsory

Education Act”

Sanne Hinderks (s2673363) Supervisors: S. Sovago & R.D. Freriks

June 2019

Abstract: Because of the great benefits of education, the goal to eliminate, or at least decrease, unauthorized absence has been a serious issue over the last decade. This paper examines the effect of a change in monitoring compliance of the compulsory education law on the test scores of elementary school pupils in the second, fifth and last grade. The first stage is based on prior research and evaluation reports of the policy reform. The second stage is examined through a repeated cross-section and uses data of the cohort research named COOL5-18. The repeated cross-section shows different effects between the groups. Students perform significantly better on language and mathematics test scores after the policy reform, while the effect on the end test of the eighth grade actually shows a negative relation.

However, there is no comprehensive evidence that the policy reform will be more effective if it kicks in at an earlier age.

(2)

2 I. Introduction

Education plays an important role in modern world. Children attending school will actually result in more chances further in life. Large series of studies show that if children perform better this significantly increases their chances in life regarding the probability of work, income and healthier behavior (Oreopoulos, 2006, 2007; Bhuller, Mogstad, & Salvanes, 2017; Brunello, Fort, Schneeweis, & Winter-Ebmer, 2016). The researchers agree upon the positive effects, referred to as benefits, of compulsory education on future aspects. Because of these great benefits of education, the goal to eliminate, or at least decrease, unauthorized absence has been a serious matter over the last decade. Cabus & De Witte (2015) compared

unauthorized absent pupils and regular school attendees on their dropout behavior. The results show that unauthorized absentees are associated with a 37.4% risk of dropping out of school (Cabus & De Witte, 2015). Moreover, Douglas & Ross (1965) show that in all social classes, different from the upper middle classes, children who are often absent in the last two years of elementary school will receive lower test scores of the last grade.

Since a lot of research has been focusing on the future, the probability of work, income and healthier behavior and there is a large consensus on this topic, I examine a different side of compulsory education law. Because of the future benefits that education has, many countries are trying to reduce school dropout rates. Dropping out of school is a big issue since we acknowledge the advantages of schooling and consequently schools cooperate more with external organizations nowadays. The goal of this paper is to show whether monitoring compliance of compulsory education in the Netherlands has an effect on the test results of children. Do children who are forced to go to school and children who are supervised on compliance with the education law, perform significantly better? This study will be an

introduction to this kind of reasoning and will create opportunities for extension. Moreover, it will be a contribution to existing literature on compulsory education law. Several studies have examined the relationship between compulsory education and school performance, however, none of them actually paid particular attention to the compliance of compulsory education. It will be compelling to observe whether monitoring of compliance results in better

performance. The outcome may be of importance for future policy implications.

(3)

3

up and till 18, who do not have a so-called start-qualification yet, are obliged to attend education by qualification law. If these pupils for some reason do not attend school, they will be marked as absent. Absence can, however, be entitled as either authorized or unauthorized. Throughout this paper there will be no distinction between the forms of absence.

On January 1st, 2012, the monitoring of compliance with the compulsory education law has been altered. Since the policy change, supervision of compliance with the Dutch Education Law was not provided by municipalities anymore, but the Dutch Inspectorate of Education is in charge now. A change in the monitoring of compliance with compulsory education law on January 1st, 2012, enabled me to examine the effect of compulsory education on test results. If the compulsory education law and especially the tightening and supervision of the law, kicks in at an earlier age does it have a significant effect on the test results of the students? More specifically, do children who are more tightened to education perform better in school?

The context of this analysis is partly based on earlier research of Cabus & De Witte (2011). Cabus & De Witte (2011) focused on an earlier policy reform in the Netherlands and therefore several aspects of this study could be implemented and used for the analysis regarding the policy reform of 2012. Cabus & De Witte (2011) study the impact of an earlier tightening of the compulsory education law in the Netherlands on the dropout rate. That specific change in law mostly alters the compulsory education age. They found evidence that the policy reform does have an effect on the group of students, since the dropout rate after the reform is much lower. However, they argue that this can also be the reason of the economic revival at that point on the labor market. On the contrary, Heers, Van Klaveren, Groot, & Van den Brink (2014) are showing that community schools are as effective as regular schools with respect to dropout reduction. Community schools are especially established to build on students’ needs and reduce the dropout rate with the help of subsidies provided by the government. In contrast to what would be expected, the community schools do not have a significant effect on dropout rate. Research of Christle, Jolivette, & Nelson (2007) shows that the rate of school attendance is not only one of the school characteristics that is negatively correlated with the dropout rate, but it is also showing the strongest relationship. Keeping this research in mind, one can use the work of Cabus & De Witte (2011) and Heers et al. (2014) in this perspective.

(4)

4

focusing on the importance of education and more specifically focusing on the effects of compulsory education, will be provided. Before examining the effect of the extended law, prove of the effectiveness of this law on dropout rates needs to be present. This prove of the first stage will be examined in section III. What follows is the analysis of the two periods, before and after the tightening of the compulsory education law in the different cohorts. To identify the research question, I will use the recent policy reform in the Netherlands. The remainder of this paper is structured as followed. Firstly, the existing literature will be examined and in the section thereafter, I will elaborate on the Dutch education system, the details of the policy reform of the compulsory education law and the first stage. Subsequently, in section IV and V, the data and the used methods will be described and afterwards, the impact of the tightening of the compulsory education law on test results will be discussed.

II. LITERATURE REVIEW

The aim of this section is to provide an overview of existing and related literature. In

example, the first paragraph discusses large series of studies that examine the effect of future benefits (Oreopoulos, 2006, 2007; Bhuller et al., 2017; Brunello et al., 2016). Secondly, the relation between absence and dropout rates will be addressed. Namely, that school

characteristics, especially absence rates, are negatively correlated with dropout rates (Christle, et al., 2007) and that community schools are as ineffective as public schools (Heers et al., 2014). In the third paragraph a particular old studies of Douglas & Ross (1965) on the relation between absence and school performance will be exploited.

(5)

5

better educated individuals may also have healthier jobs or live in a healthier surrounding. Better education may raise income levels, which will at the same time increase health outcomes. A different, more aligned research is that of Fang, Eggleston, Rizzo, Rozelle, & Zeckhauser (2012). They examined the relation between compulsory education and the return of one additional year of schooling in China. First of all, they did find that the Compulsory Education Law of 1986 raised overall educational attainment by about 0.8 years of schooling (Fang et al., 2012). More particular, Fang et al. (2012) found evidence that the return of one additional year of schooling is approximately 20 percent per year.

It is commonly known that education results in better health and wealth. Since this paper is particularly focusing on compulsory education and performance of students, the following papers are of greater importance for the analysis. A study conducted by Christle et al. (2007) shows that the school attendance rate is negatively correlated with the dropout rate. They argue that school characteristics influence the level of dropout rates in high schools. More specifically, this means that failure of schools in early years attribute to a weaker attachment with schools and eventually student drop out. Moreover, they did not just found a negative correlation between attendance rates and dropout numbers, but from all characteristics the, attendance rate, has the strongest relation. Because attendance rates and dropout rates are correlated, other work of dropout rates will be of relevance for this paper. Earlier work of Cabus & De Witte (2011) studies the impact of a change of the compulsory education law on August 1st_{2007 in the Netherlands. Basically, what happened in 2007 is that the age of} mandatory education has increased. Students before the policy reform, compared to students after the policy reform, are more likely to drop out of school as they have legal rights to do so. The authors found evidence that the policy reform does have an effect on the group of

students since the dropout rate after the reform is much lower. However, with some rational thinking they argue that the outcome can also be explained by the economic revival at that point in time on the labor market.

In another paper, Oreopoulos (2006a) examined the effects of raising the compulsory

(6)

6

and health perspectives? One of the explanations may be the high uncertainty of additional earning from staying on (Chen, 2001). Another explanation, dropout exists because students simply detest school and are willing to leave school as soon as possible (Lee & Burkam, 2003). The last reason mainly flows into the work of Heers et al. (2014). They focus on the introduction of community schools which are inclined to establish an environment that is based on students’ needs. However, Heers et al. (2014) are showing that community schools are as effective as regular schools with respect to dropout reduction. Community schools are especially established to build on students’ needs and reduce the dropout rate with the help of subsidies provided by the government. Moreover, they work together with external

organizations, cooperate with parents to improve children’s capacities and offer more extracurricular activities. Heers et al. (2014) focus on pre-vocational education because this track shows highest dropout rates. Stimulating subsidies were not introduced before the beginning of schoolyear 2006/2007. Therefore, the authors can use a difference-in-difference model by taking both students before and after the introduction of community schools into account. In contrast to what would be expected, there is no empirical evidence that

community schools contribute to a decline in dropout rates and better results.

A rather old, but sufficiently helpful research is that of Douglas & Ross (1965) in which they examine the effects of absence on primary school performance in England. In their model they make a distinction between the upper and lower, middle and manual working class. The outcomes of their analyses differ between social classes. For the upper middle class test results are unaffected by the amount of absence. However, for the other social classes this absenteeism resulted in a lower performance. Absence in the first two years of elementary school will be adjusted by students in the accompanying years, but not if these students are from the lower manual working class, the lowest social class.

III. INSTITUTIONAL DETAILS

(7)

7 Dutch education system

Before moving on to the analysis, I will provide the reader with more insight of the Dutch education system. Dutch children are obliged to follow education from the age of five.

However, most children already attend elementary or primary school on their fourth birthday. At the age of four parents are able to subscribe children at elementary school. Before that, most children are going to pre-school or daycare. Primary school consist of 8 years of schooling, including grade 1 up and till grade 8. Special needs children and children with learning difficulties will go to a special schools for primary education. Since August 2014, primary schools need to offer special needs children a program that is appropriate for them (Dekker, 2015). While in elementary school, students need to perform required national tests on language and mathematics. These tests are mandatory for all grades. In the last grade, the eighth grade, students need to do a so-called Cito end test which entails information of the specific track of secondary education the student should follow. This test is also a by the government required test. After finishing primary school, pupils are following secondary education based on their own education level decided by primary education tests. Ministry of Education, Culture and Sciences writes that there are three major levels: pre-vocational secondary education (VMBO), senior general secondary education (HAVO) or pre university education (VWO). These major levels prepare students for, respectively, secondary vocational education (MBO), higher professional education (HBO) or university education (WO). All these tracks are summarized in figure 1 that gives a clear overview of how the Dutch Education system works.

Figure 1: Flow chart of the Dutch Education System.

(8)

8 Compulsory education law

The compulsory education law in the Netherlands prescribes that children, aged from 5 up and till 16 year, are obliged to attend education (Leerplichtwet 1969). Since August 1st_{, 2007, the} compulsory education age has increased to 18 years. However, it is not just the eighteenth birthday that is taken into account. Between the age of 16 and 18 students need to qualify. Qualification means a completed senior general secondary education, pre university education or a secondary vocational education, as the Government of the Netherlands explains. It could be that a student completes the senior general secondary education or pre university education track without reaching the age of 18. In that case students receive their so-called

start-qualification and are not mandated to education anymore. This start- start-qualification has been introduced to avoid or reduce student dropout.

Policy reform

On January 1st, 2012, the compulsory education law changed again. Before the policy reform, supervision of compliance with the education law of schools was provided by municipalities. Afterwards, supervision was no longer provided by municipalities but since then the Dutch Inspectorate of Education has been supervising attendance. The Ministry of Education describes the Dutch Inspectorate of Education as an organization that is responsible for the inspection and review of schools and other educational institutions1_{. Other responsibilities of}

the Inspectorate are stimulating educational institutions and hence municipalities to maintain and improve the quality of schooling. Each school has a policy on how to act in case of absence. In accordance with the compulsory education law schools are obliged to report unauthorized school failures when the 16-hour limit in four school weeks has been reached. Schools will contact the school attendance officer of the municipality, who in turn will speak with the students and will try to reduce the absence. The Dutch Inspectorate of Education works together with the attendance officer of the municipality and they exchange rarities. If there are shortcomings in the schools, the Inspectorate will supervise the schools on a deeper level, while otherwise the supervision is just on the surface. In the end, the compulsory education law itself has not tightened, only the supervision has become stricter and tighter. The Dutch evaluation of the change in compulsory education law supervision concludes less absence due to stricter supervision (Eimers, Jager, & Keppel, 2014).

(9)

9 First stage

Before one can estimate the effect of change in compliance with the compulsory education law, one step needs to be examined first. Namely, the first stage in which the effect of the tightening of the law with respect to absence rates will be examined. According to the literature Christle et al. (2007) state that attendance rates and dropout rates are correlated. Eventually arguing that all dropout rates are interchangeable with attendance rates. Cabus & De Witte (2011) concluded that a change in compulsory education law led to a decline in dropout rates, or interchangeably attendance rates. Next to the literature outcomes, some raw data on attendance rates in the Netherlands is included. According to the raw number of the NJI (Netherlands Youth Institute) right after the implementation of the new supervision system of compulsory education in 2012 the relative absence rates has declined more and more. Figure 2 shows these decline of relative absence rates after 2012. Strikingly, the relative absence rates have declined over time while the luxury absence rates had declined right after the implementation but has been increasing over the last years. Additionally, Eimers et al. (2014) wrote a report after the policy reform about the effects of the implementation and they concluded that absence rates had declined afterwards. Along with the aforementioned

literature the raw numbers of the NJI provide evidence for the negative relationship between the policy reform and absence rates.

Figure 2: Relative absence rates per year

(10)

10

IV. DATA AND DESCRIPTIVE STATISTICS

This section describes the data sample of the Netherlands. Detailed information about the source, years and variables included in the dataset is provided. Moreover, I discuss which variables are included in the model and why those variables are important for the analysis. The last part of the section contains information about the descriptive statistics.

Data source

Data of the primary school students is found at the Netherlands Institute of Social Research (SCP)2. This cohort research is known as COOL5-18 and includes information about children aged 5 up and till 18 years, the compulsory education age. A total of 550 schools is included into this cohort research, which is approximately 8% of the whole population of schools (Driessen, Mulder, Ledoux, Roeleveld, & Van der Veen, 2009). These schools are a mixture of regular and minority schools which are treated equally in the analysis. All students participating in COOL5-18 will remain in this sample even if they change school or repeat a class. The data is a panel data set and includes sets of the years 2007/2008, 2010/2011 and 2013/2014 that in turn include information about students and its characteristics. The focus of the COOL5-18_{research is on primary school students of grade 2, 5 and 8 and on secondary} school students of grade 3. Additionally to this grades, the last two cohorts contain

information on senior general secondary education, pre university education and secondary vocational education. Special point of interest is the development of the students regarding knowledge and knowhow of Dutch, English and Mathematics, social competences and social-emotional status. My focus in the empirical model is on the primary school students. The size of each cohort is approximately 5.000-7.500 students. The sample only includes those

students who performed all tests in primary school. Variables included in the dataset are, in example, age, gender, learning difficulties, disorders, country of birth of both the student and the parents, education background of the parents, type of school, school projects of pre-school children, province they are living in, urbanization grade and different test results of the three grades. For this analysis, age, gender, school, test results, the social-economic status,

urbanization and the province they live in are used (Appendix A, table 1). Gender, social-economic status, urbanization and the province are used as control variables. These variables are held constant throughout the analysis in order to assess the test results over time. It is

(11)

11

important to include these control variables because these may affect the independent variable which in turn can lead to underestimation or overestimation. The test of grade 2 is composed of language skills of students aged 5-6. Throughout this paper, this test will be referred to as test_language. Additionally, test results of grade 5 and 8 are assembled based on technical reading, reading comprehension and mathematics skills which will be referred to as test_techread, test_compr and test_math, respectively. The last test score that needs an explanation is the end test of grade 8. This test provides the student with an insight of the specific track (intermediate vocational education, higher secondary education or

pre-university education) suiting them the next year at secondary education. In the remaining this test will be referred to as test_total. Control variables and test results are not available for all students. Therefore, it is important to drop all the missing values in order to perform all analysis, with and without control variables, under the same estimation sample.

Descriptive statistics

One can see that the test results of test_total are only available for primary students of grade 8. The reason for that is based on the fact that these tests are to verify the level of secondary education the student will attend the year after. The mean of the test_total in 2007/2008, 2010/2011 and 2013/2014 are 533.12, 534.59 and 533.57 respectively (Appendix A, table 4, 7 and 10). The results of the elementary end test vary between 501 and 550, which is the

(12)

12

7, 9 and 10). Reasons for the decline are not mentioned, but it could be that schools are paying less attention to reading comprehension overall. Another striking issue with the test_compr variable concerns the minimum value of the test score. The results of test_compr can have a negative value because the test scores are measured in a way that the average test score of reading comprehension is zero. Only the test_compr variable has that particular kind of assessing. Test_math concerns a test that measures the ability of children regarding

mathematics. Average results of grade 5 for the years 2007/2008, 2010/2011 and 2013/2014 are respectively 82.53, 70.73 and 71.02, while average results of grade 8 for the years

2007/2008, 2010/2011 and 2013/2014 are respectively 116.97, 112.06 and 110.33 (Appendix A, table 3, 4, 6, 7, 9 and 10). For the empirical analyses all the test results will be adjusted as standardized numbers. The different test results are measured in different ways and therefore standardization has been applied.

The dummy variable gender has two obvious segments. For all classes and years the division between male and female differs but is close to fifty percent. As a control variable the social-ethnical background of the students is included in the model. In the dataset the CoolSES is a categorical variable varying from 1 up and till 6, whereas 1 = max lower vocational education for foreigners 2 = max lower vocational education for natives, 3 = max secondary vocational education for foreigners, 4 = max secondary vocational education for natives, 5 = higher professional education/university for foreigners and 6 = higher professional

education/university for natives. For all the years and grades the fourth segment, maximum of higher secondary vocational education for native people, is most represented. On top of the CoolSES control variable, the degree of urbanization and the province are included. Both these variables tells us more about the area they live in. Do people in urban areas perform better or worse in school and is the influence of the policy reform on the test results heavier in a particular living area? Both control variables are dummies in which the degree of

urbanization varies between strong urban, urban, moderate urban, little urban and rural. One can see from Appendix A that the composition of the grades is based on different

representations of the degree of urbanization. Province is just divided into the twelve

(13)

13 V. Empirical strategy

Throughout this section several things will be discussed. Starting with the repeated cross-section and the control versus treatment group. The assumptions needed to perform the analysis will be discussed thereafter and lastly the strategy of the repeated cross-section will be the main talking point.

Repeated cross-section

The straightforward way to examine a structural break is by using panel data. Panel data uses the same groups (cohorts) at different point in times, which is exactly what this paper is analyzing here. Actually, three cohorts will be examined before and after the tightening of the supervision of the compulsory education law.

Over the years, each school will be examined on the same grade. In words, this means that of one particular school grade 2 is examined in 2007/2008, grade 2 is examined in 2010/2011 and grade 2 is examined in 2013/2014. However, for the repeated cross-section only the two cohorts of 2010/2011 and 2013/2014 are used for the analysis. The reason for this is that otherwise the two groups (before and after the policy reform) are too different for comparison. Investigating the repeated cross-section will provide a clear overview of evidence if schools will obtain better test results over time. The focus is especially towards the latest cohort of 2013/2014 to which the new law has been introduced. Additionally, grade 5 and grade 8 are taking into account as well, in which cohort 3 will again be appointed as treatment group. In table 11 (Appendix A), one can find a clear overview of how the cohorts and years are intertwined and which cohort will be the treatment or control group.

(14)

14

The first analysis is a simple regression in which the dummy variable reform will show the achieved results on tests. The dependent variable is different for each grade. In example for grade 2 the dependent variable is only test_language, while for grade 5 and 8 the dependent variable has multiple categories of test results, including test_techread, test_compr, test_math and test_total. In the main text, the focus will be on test_langauge of grade 2, test_math of grade 5 and test_total of grade 8. All the other test scores will be analyzed in Appendix B.

Yisc = β0 + β1 * Reformc + Mi + Reform*Xisc + εisc, (1) i = identification number, s = school, c = cohort, M = control variables

Formula 1 actually consist of multiple analyses. Namely, the first analysis described above, which only includes the first part of the formula. Roughly speaking, the second analysis will be the same as the first one. The second analysis is based on school fixed effects. In this case it will not be tested upon individuals over time, but the schools are of interested. Performing a fixed effect test will account for unobserved school effects. Meaning that all unobserved school effects that could influence the relation between the policy reform and test results are removed from the model which reduce the threat of omitted variable bias. The third analysis is related to the second analysis but includes a couple of control variables. The control variables are introduced because of the large possibility that they would affect the outcomes of this analysis. More specific, the school-specific effects are included to account for potential compositional differences of cohorts.

Interaction effects

The last and fourth analysis is based on the entire formula. It includes fixed effects, control variables and interaction effects. These interaction effects are a multiplication of the reform dummy variable and the gender dummy variable. Additionally, an interaction effects is made based on the multiplication of the dummy variable reform and the social economic status, referred to as CoolSES in the model. Interaction terms are added to check whether the effect of one variable (reform) depends on the value of another variable (gender or CoolSES). Furthermore, the interaction terms of the control variable CoolSES and the dummy variable reform will be tested of heterogeneity. In order to examine if heterogeneity is present, a Wald test is performed. A Wald test assumes the null hypothesis to be homogeneous and

(15)

15

VI. RESULTS

In this section the results of the four empirical tests will be discussed. First, the three tests that will be analyzed individually are elaborated upon. Thereafter, the interaction effects and the homogeneity versus heterogeneity issue will be discussed. Furthermore, a conclusion with a comparison of the effects of the different grades is included.

The first analysis shows us that students after the policy reform actually perform better. Namely, being a student after the policy reform will increase test results of grade 2 by 0.19 of a standard deviation at a 1% significance level (Table 1). The coefficient of test_math, which measures mathematics skills of grade 5, is not significant at the 10% level (Table 2). In contrast with the other test results explained before, the policy reform has a negative effect on the results of test_total. The results of grade 8 decline by 0.10 of a standard deviation at a 1% significance level (Table 3).

The second analysis shows relatively similar outcomes. Namely, being a student after the policy reform will increase test results of grade 2 by 0.21 of a standard deviation (Table 1). However, the fixed effect analysis now shows a positive and significant effect of the policy reform on the results of the mathematics test. To be specific, for students after the policy reform compared to students before the policy reform, test results of grade 5 rise by 0.05 of a standard deviation (Table 2). The negative effect in the first analysis has come forward in the second analysis as well. For students after the policy reform, the results of grade 8 decline by 0.04 of a standard deviation (Table 3). All these coefficients are shown to be significant at the 1% level.

The third analysis is based on the reform variable and control variables. Compared to the regressions before, the third analysis shows quite similar results as well. For students after the policy reform, the results of grade 2 increase by 0.20 of a standard deviation (Table 1). The dummy variable gender can be interpreted by saying that a female is associated with a 0.16 of a standard deviation (Table 1). Moreover, having a lower vocational education native

(16)

16

compared to foreign backgrounds. All these coefficients are shown to be significant at the 1% level. In addition, living in an urban area compared to a strongly urban area increases test results of grade 2 by 0.26 of a standard deviation (Table 1). The urbanization results are a bit ambiguous, since moderate urban compared to urban does not show that big of a change, 0.32-0.26= 0.06 (Table 1). Both effects are shown to be significant at the 5% level. There is evidence for an increase in test results when the degree of urbanization declines, but there is difference in the extent of increasing test scores.

Moving on to grade 5, for students after compared to students before the policy reform, the results increase by 0.04 of a standard deviation at the 5% significance level (Table 2).

Additionally, one can say that being a female is associated with a declining result of 0.35 of a standard deviation (Table 2). This is line with the gender stereotype outcomes. Moreover, having a LBO native background compared to a LBO foreign background increases the test results by 0.11 of a standard deviation (Table 2). From table 3 it is quite clear that having a foreign background will most likely result in similar results as natives with a lower degree of education. All these coefficients are shown to be significant at the 1% level. In addition, living in an urban area compared to a strongly urban area decreases results of mathematics, but no economic interpretation can be given due to no significance (Table 2). The rest of the urbanization results are declining and significant at the 10% and 5% respectively, which means that students living in an urban area perform better in mathematics (Table 2).

Lastly, grade 8 and the results of test_total will be examined. For students after compared to students before the policy reform the results are declining, but no economic interpretation is possible based on absence of significance (Table 3). Additionally, one can say that being a female decreases the results by 0.07 of a standard deviation at the 1% significance level (Table 3). From table 3 it is quite clear that test results are increasing with the education background, independent of foreign or native. Evidence is sufficient for only four out of five coefficients. Remarkably is the decline in test results of test_total by students living in a little urban area compared to those living in a moderate urban area. The decline of 0.87 of a

(17)

17 Table 1: Specification of the model for grade 2

Dependent variable: test_language

*, **, *** indicate significance at the 10, 5 and 1 percent level, respectively

Grade 2

test_language

Linear regression Fixed Effects Fixed effect + control variables

Coefficients (S.E.) Coefficients (S.E.) Coefficients (S.E.)

Reform .19003 (.01528)*** .21405 (.01940)*** .20430 (.01943)***

Gender

Female .16399 (.01354)***

CoolSES

Max. lower vocational education, native Max. secondary vocational education, foreign Max. secondary vocational education, native Higher professional education/university, foreign Higher professional education/university, native

(18)

Dependent variable: test_math

Grade 5

test_math

Reform .01807 (.01478) .05357 (.01837)*** .03720 (.01833)**

Gender

Female -.34726 (.01357)***

CoolSES

(19)

Dependent variable: test_total

*, **, *** indicate significance at the 10, 5, and 1 percent level, respectively

Grade 8

test_total

Coefficients (S.E.) Coefficients (S.E.) Coefficients (S.E.) Reform -.10182 (.02054)*** -.03628 (.03128) -.03511 (.03241) Gender Female -.07160 (.01858)*** CoolSES

(20)

20 Table 4: Specification of the model: Interaction effects

*, **, *** indicate significance at the 10, 5, and 1 percent level, respectively

Interaction effects

The fourth analysis does not only include the control variables, it also focuses on the

interaction effect of the reform and education background. Compared to the other regressions, coefficients and significance are quite similar. Therefore, the focus is on the marginal effect of the reform. Students of grade 2 after the policy reform need to deal with the marginal effect 0.24 of a standard deviation and on top of that the coefficient of the interaction term. At the 10% significance level this only holds for the native education background of secondary vocation education. This means that after the reform students of that specific background had an increase in performance of 0.24 – 0.10 = 0.14 of a standard deviation (Table 4). All these coefficients are interpreted based on the ceteris paribus condition. These interaction effects are also tested on heterogeneity. The interaction terms are jointly significant at the 10% level (F=1.95, p=0.0833), meaning that heterogeneity is present.

For the mathematics test score of grade 5 the coefficients are indistinguishable as well. Interpreting the interaction effects is not possible due to fact that none of the coefficients is

Fixed effect + control variables + interaction effect Fixed effect + control variables + interaction effect Fixed effect + control variables + interaction effect Coefficients (S.E.) test_language grade 2 Coefficients (S.E.) test_math grade 5 Coefficients (S.E.) test_total grade 8 Reform .24257 (.05212)*** .03825 (.04800) -.06624 (.06775) Reform#gender Reform#CoolSES

(21)

21

significant. Another striking issue concerns the reform coefficient. Including the interaction effects resulted in an insignificant reform coefficient. The interaction terms for test_math are not jointly significant (F=0.44, p=0.8243), hence homogeneity. The same issues hold for the scores of test_total of grade 8. These interaction are also not jointly significant (F=0.80, p=0.5484). Additionally, female students after the policy reform are associated with an increase of 0.14 of a standard deviation on the test results of test_total, significant at the 1% level (Table 4).

Comparison of the grades

One of the most compelling outcomes is the fixed effects analysis. Including fixed school effects would normally make the analysis and thereby the coefficients more accurate. However, no precision gain exists moving on from the first to the second analysis.

One of the first interesting comparisons may be the one of the development of the fifth and eighth grade over time. At first, the focus will be on the variable test_math. There is a significant effect of 0.04 of a standard deviation in the fifth grade at the 5% significant level according to the third analysis (Table 2). In contrast, this variable has a negative significant effect of 0.08 of a standard deviation in grade 8 at the 1% significance level (Appendix B, table 5). One can, based on these aforementioned outcomes, conclude that when the reform kicks in at an earlier age it has a positive effect on mathematical test results. Secondly, the technical reading test score test_techread shows a positive and significant effect of 0.05 and 0.22 of a standard deviation on the 1% level in respectively grade 5 and grade 8 (Appendix B, table 1 and 3). These coefficients are derived in the first analysis, the linear regression without fixed effects and control variables. This conclusion is not in line with the one obtained from the test_math variable since it actually says that children of grade 8 are benefitting more from the policy reform. Taking the first analysis into consideration, the variable test_compr shows a negative outcome of 0.07 and 0.11 of a standard deviation on the 1% level for grade 5 and grade 8, respectively (Appendix B, table 2 and 4). Again this is more in line with common knowledge. Namely, that when the policy reform kicks in at an earlier age the effects are positive or less negative in this case. Similarly, all of the coefficients corresponding to the gender variable, is associating a more positive or less negative effect in the fifth grade

(22)

22

The outcomes of the analyses are quite striking because one would expect a positive

relationship between the policy reform and the test scores. A reason for this negative outcome might be that due to the compliance with education law more students are forced to attend school. Often students who are regularly absent are underperforming. Underperforming might be the reason why students are absent. After the policy reform these students needs to comply with the compulsory education law which means that all underperforming students are

included in the sample as well. More concrete, test results might be lower due to inclusion of bad test scores of underperforming students. Since these test are compulsory for all students, it might not may be the case that the reason described above is the right one. Another reason for this declining results in grade 5 and grade 8 might be based upon a third variable that influences both the absence rates and test scores of students or it might just be other control variables that affect the test results.

VII. CONCLUSION AND DISCUSSION

(23)

23

evidence that it will be more effective for test scores of children if the policy reform kicks in at an earlier age.

Room for discussion is present for the reliability of the absence rates. In August 2014 the Dutch government introduced the so-called ‘appropriate’ education plans (Dekker, 2015). Practically, this means that special needs students are more often inclined to follow a ‘normal’ education route, because of the duty that schools have regarding care and support services. However, the attendance rates of special needs students at a normal school have risen. Lots of special needs students do not see a normal school as suitable, while going to special education also is not under discussion. Therefore, it is debatable whether the absence rates of Dutch children are accurate or not. Moreover, the first stage is based on raw numbers of the NJI (Netherlands Youth Institute). These numbers do include both elementary school and secondary school students. This research is focusing on elementary school students, which might have altered the outcomes. Nevertheless, the data source is reliable and next to these raw numbers the first stage is also build upon earlier research.

(24)

24 VIII. REFERENCES

Bhuller, M., Mogstad, M., & Salvanes, K. G. (2017). Life-cycle earnings, education premiums, and internal rates of return. Journal of Labor Economics, 35(4), 993-1030.

Brunello, G., Fort, M., Schneeweis, N., & Winter-Ebmer, R. (2016). The causal effect of education on health: What is the role of health behaviors? Health economics, 25(3), 314-336.

Cabus, S. J., & De Witte, K. (2011). Does school time matter? On the impact of compulsory education age on school dropout. Economics of Education Review, 30(6), 1384-1398.

Cabus, S. J., & De Witte, K. (2015). Does unauthorized school absenteeism accelerates the dropout decision?–Evidence from a Bayesian duration model. Applied Economics Letters, 22(4), 266-271. Chen, S. H. (2001). Is Investing in College Education Risky? State University of New York at Albany, Department of Education Discussion Papers, (01-09).

Christle, C. A., Jolivette, K., & Nelson, C. M. (2007). School characteristics related to high school dropout rates. Remedial and Special education, 28(6), 325-339.

Dekker, S. (2015). Reactie op onderzoeken passend onderwijs [Letter of government]. Retrieved from

https://www.rijksoverheid.nl/onderwerpen/passend- onderwijs/documenten/kamerstukken/2015/09/25/kamerbrief-met-reactie-op-onderzoeken-passend-onderwijs

Douglas, J. W. B., & Ross, J. M. (1965). The effects of absence on primary school performance. British Journal of Educational Psychology, 35(1), 28-40.

Driessen, G., Mulder, L., Ledoux, G., Roeleveld, J., & Van der Veen, I. (2009). Cohortonderzoek COOL 5-18. Technisch rapport basisonderwijs, eerste meting 2007/08.

Eimers, T., Jager, A., & Keppels, E. (2014). Toezicht op naleving: evaluatie wijziging Leerplichtwet 1969. Nijmegen: KBA Nijmegen.

Fang, H., Eggleston, K. N., Rizzo, J. A., Rozelle, S., & Zeckhauser, R. J. (2012). The returns to education in China: Evidence from the 1986 compulsory education law (No. w18189). National Bureau of Economic Research.

Heers, M., Van Klaveren, C., Groot, W., & Van den Brink, H. M. (2014). The impact of community schools on student dropout in pre-vocational education. Economics of Education Review, 41, 105-119. Lee, V. E., & Burkam, D. T. (2003). Dropping out of high school: The role of school organization and structure. American Educational Research Journal, 40(2), 353-393.

Oreopoulos, P. (2006a). Estimating average and local average treatment effects of education when compulsory schooling laws really matter. American Economic Review, 96(1), 152-175.

Oreopoulos, P. (2006b). The compelling effects of compulsory schooling: Evidence from Canada. Canadian Journal of Economics/Revue canadienne d'économique, 39(1), 22-52.

(25)

25 IX. Appendix

Appendix A: Specification of data

Table 1: Data structure

2007/2008 2010/2011 2013/2014

All Grades School number School number School number Student number Student number Student number

Gender Gender Gender

CoolSES CoolSES CoolSES

Urbanization degree Urbanization degree Urbanization degree

Province Province Province

Reform Reform Reform

Grade 2 Test_language Test_language Test_language

(26)

26 Table 2: Descriptive statistics cohort 2007/2008 Grade 2

Variable Obs Mean Std. Dev. Min Max

Schoolnr 12,127 2544.522 1845.362 1001 5314

Student 12,127 2.54e+07 1.85e+07 1.00e+07 5.31e+07

Gender 12,127 1.480086 .4996239 1 2

Male 6,305

Female 5,822

CoolSES 12,127 3.792199 1.688322 1 6

Max. lower vocational education, foreign 1,628 Max. lower vocational education, native 1,671 Max. secondary vocational educational, foreign 934 Max. secondary vocational education, native 4,343

Higher professional education/university, foreign 462 Higher professional education/university, native 3,089

(27)

Variable Obs Mean Std. Dev. Min Max

Schoolnr 3,892 2342.869 1753.846 1002 5291

Student

3,892 2.34e+07 1.75e+07 1.00e+07 5.29e+07 Gender 3,892 1.510021 .4999638 1 2 Male 1,907 Female 1,985 CoolSES 3,892 3.65185 1.667971 1 6

Max. lower vocational education, foreign 521 Max. lower vocational education, native 681 Max. secondary vocational education, foreign 282 Max. secondary vocational education, native 1,431 Higher professional education/university, foreign 102 Higher professional education/university, native 875

(28)

Variable Obs. Mean Std. Dev. Min Max

Schoolnr 6,264 2214.562 1696.036 1001 5311

Student 6,264 2.22e+07 1.70e+07 1.00e+07 5.31e+07

Gender 6,264 1.489943 .4999387 1 2

Male 3,195

Female 3,069

CoolSES 6,264 3.615102 1.713422 1 6

Max. lower vocational education, foreign 951 Max. lower vocational education, native 1,091 Max. secondary vocational education, foreign 453 Max. secondary vocational education, native 2,123 Higher professional education/university, foreign 215 Higher professional education/university, native 1,431

(29)

Schoolnr 9,074 2757.421 1871.511 1001 5500

Student 9,074 2.76e+07 1.87e+07 1.00e+07 5.50e+07

Gender 9,074 1.481926 .4997008 1 2

Male 4,701

Female 4,373

CoolSES 9,074 4.143928 1.626818 1 6

Max. lower vocational education, foreign 826 Max. lower vocational education, native 963 Max. secondary vocational education, foreign 643 Max. secondary vocational education, native 3,248 Higher professional education/university, foreign 435 Higher professional education/university, native 2,959

(30)

Schoolnr 10,624 2692.744 1856.674 1001 5500

Student 10,624 2.69e+07 1.86e+07 1.00e+07 5.5e+07

Gender 10,624 1.495576 .500004 1 2

Male 5,359

Female 5,265

CoolSES 10,624 4.075301 1.665181 1 6

Max. lower vocational education, foreign 1,102

Max. lower vocational education, native 1,205

Max. secondary vocational education, foreign 722

Max. secondary vocational education, native 3,766

Higher professional education/university, foreign 420

Higher professional education/university, native 3,409

(31)

31

Table 7: Descriptive statistics cohort 2010/2011

Grade 8

Schoolnr 4,985 2480.34 1763.865 1001 5500

Student 4,985 2.48e+07 1.76e+07 1.00e+07 5.50e+07

Gender 4,985 1.505517 .5000197 1 2

Male 2,465

Female 2,520

CoolSES 4,985 4.010832 1.671773 1 6

(32)

32 Table 8: Descriptive statistics Cohort 2013/2014 Grade 2

Schoolnr 7,985 3362.63 2042.109 1006 8791

Student 7,985 3.36e+07 2.04e+07 1.01e+07 8.79e+07

Gender 7,985 1.479023 .4995911 1 2

Male 4,160

Female 3,825

CoolSES 7,985 4.24258 1.57936 1 6

Max. lower vocational education, foreign 630 Max. lower vocational education, native 688 Max. secondary vocational education, foreign 650 Max. secondary vocational education, native 2,894 Higher professional education/university, foreign 393 Higher professional education/ university, native 2,730

(33)

Schoolnr 8,044 3367.126 2005.451 1004 8791

Student 8,044 3.37e+07 2.01e+07 1.00e+07 8.79e+07

Gender 8,044 1.495276 .5000088 1 2

Male 4,060

Female 3,984

CoolSES 8,044 4.136002 1.620505 1 6

(34)

Variable Obs. Mean Std. Dev. Min. Max.

Schoolnr 4,501 3473.676 2066.471 1004 8791

Student 4,501 3.47e+07 2.07e+07 1.00e+07 8.79e+07

Gender 4,501 1.510553 .4999442 1 2

Male 2,203

Female 2,298

CoolSES 4,501 4.055988 1.648831 1 6

(35)

35 Table 11: Specification of the cohorts

2007/2008 2010/2011 2013/2014

Grade 2 Control group Control group Treatment group

Grade 5 Control group Control group Treatment group

(36)

36 Appendix B: Specification of the results

Table 1: Specification of the model for grade 5 Dependent variable: test_techread

Grade 5

test_techread

Reform .04791 (.01478)*** .03352 (.01877)* .03197 (.01942)

Gender

Female .07959 (.01438)***

CoolSES

(37)

Dependent variable: test_compr

Grade 5

test_compr

Reform -.06594 (.01477)*** -.07162 (.01807)*** -.08419 (.01813)***

Gender

Female .24067 (.01342)***

CoolSES

(38)

Dependent variable: test_techread

Grade 8

test_techread

Reform .21882 (.02044)*** .19277 (.03246)*** .18820 (.03489)***

Gender

Female .07496 (.02000)***

CoolSES

(39)

Dependent variable: test_compr

Grade 8

test_compr

Reform -.10534 (.02053)*** .00231 (.03091) .01520 (.03218)

Gender

Female .13469 (.01844)***

CoolSES

(40)

Dependent variable: test_math

Grade 8

test_math

Reform -.13685 (.02051)*** -.09631 (.03074)*** -.08454 (.03192)***

Gender

Female -.26377 (.01830)***

CoolSES

(41)

41

Table 6: Specification of the model, interaction effects grade 5

Table 7: Specification of the model, interaction effects grade 8

Grade 5 Fixed effect + control variables + interaction effect Fixed effect + control variables + interaction effect Coefficients (S.E.) test_techread Coefficients (S.E.) test_compr Reform Reform#gender .01945 (.05086) -.02004 (.02901) -.08920 (.04747)* .01652 (.02708) Reform#CoolSES

Max. lower vocational education, native# Max. secondary vocational education, foreign# Max. secondary vocational education, native# Higher professional education/university, foreign# Higher professional education/university, native#

-.04090 (.06740) .06037 (.07046) .05213 (.05452) .02315 (.08259) .00626 (.05546) -.01434 (.06291) .02908 (.06577) -.01037 (.05089) .00768 (.07703) -.00354 (.05176) Grade 8 Fixed effect + control variables + interaction effect Fixed effect + control variables + interaction effect Fixed effect + control variables + interaction effect Coefficients (S.E.) test_techread Coefficients (S.E.) test_compr Coefficients (S.E.) test_math Reform Reform#gender .28483 (.07298)*** .08931 (.04001)** .12064 (.06730) * .03127 (.03690) .02783 (.06673) .04510 (.03659) Reform#CoolSES

Max. lower vocational education, native# Max. secondary vocational education, foreign# Max. secondary vocational education, native# Higher professional education/university, foreign# Higher professional education/university, native#

(42)

42 Results

The first analysis shows us that after the policy reform students perform better. Test scores of technical reading increase by 0.05 and 0.22 of a standard deviation in respectively grade 5 and 8 (Appendix B, table 1 and 3). Moreover, the comprehensive reading skills do have a negative and significant effect of 0.07 and 0.11 of a standard deviation in the fifth and eighth grade respectively (Appendix B, table 2 and 4). The remaining dependent variable test_math of grade 8 shows a negative relation with the policy reform of 0.14 of a standard deviation (Appendix B, table 5). All coefficients are shown to be significant at the 1% significance level.

The second analysis includes the school fixed effects and shows relatively similar results. Test scores of test_techread increase by 0.03 and 0.19 of a standard deviation in grade 5 and 8 at respectively the 10% and the 1% significance level (Appendix B, table 1 and 3). The test score of reading comprehension in grade 5 is associated with a negative relation of 0.07 of a standard deviation at the 1% significance rate (Appendix B, table 2). Moreover the

mathematics test score shows a negative and significant relation of 0.10 of a standard deviation at the 1% level (Appendix B, table 5).

The third analysis included school fixed effects and control variables. It is obviously how to interpret the reform variables and therefore the focus will be on the unusual outcomes. One can conclude that parental background of lower vocational education for natives compared to parental background of lower vocational education for foreigners decreases the test results by 0.11 and 0.16 of a standard deviation in respectively grade 5 and 8 at the 1% significance level (Appendix B, table 1 and 3).