• No results found

The relative over- and under-advising of students for a track level in secondary education in the Netherlands : the influence of parents' education and ethnicity

N/A
N/A
Protected

Academic year: 2021

Share "The relative over- and under-advising of students for a track level in secondary education in the Netherlands : the influence of parents' education and ethnicity"

Copied!
33
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

The relative over- and under-advising of

students for a track level in secondary

education in the Netherlands

The influence of parents’ education and ethnicity

Yan Crabbendam

UvA id: 10270698

Master thesis: Public Policy

University of Amsterdam

Supervisor: Prof. Dr. Hessel Oosterbeek

Amsterdam, 2017

(2)

Abstract

Recent reports by the Dutch Inspection Bureau for Education show an increasing segregation in enrolment in the different track levels in secondary education between students with an immigrant and native background and between students with highly educated parents and students with parents with a low level of education. This thesis aims to find an explanation for this phenomenon by analysing the track advice given by the teachers in primary

education, and to investigate whether some socio-economic groups are relatively wrongly advised. It uses a probit model to regress data from the years 2008 and 2014, including dummy variables for different levels of track advice, and taking variables that could or should affect this advice in account. Evidence for the over-advising and under-advising of different socio-economic groups was found, although this effect did not appear to be increasing.

(3)

STATEMENT OF ORIGINALITY

This document is written by student Yan Crabbendam, 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.

(4)

Table of Contents

1. Introduction ... 5 2. Context ... 8 3. Data . ... 9 3.1 Descriptive data ... 12 4. Empirical method ... 14 5. Results ... 16 5.1 Results 2008 ... 17 5.2 Results 2014 ... 19

5.3 Non-cognitive student profile variables ... 22

5.4 Main results ... 22

6. Conclusion and Discussion ... 25

References ... 28

Appendices ... 32

Appendix A ... 32

(5)

1. Introduction

One of the most important moments of a Dutch student’s school career is the final evaluation talk with the teacher in the final grade of primary school. In the Dutch education system, tracking of students into different levels of education happens after the transition from primary education to secondary education at the age of 12. This decision is made by the teacher(s) in the final grade of primary school, generally in consultation with the school board (Inspectie van het Onderwijs, 2014). The decision is based on both cognitive as well as non-cognitive skills (Driessen & Smeets, 2007). For an objective second measurement of students’ cognitive skills, primary schools use an independent exam; 85% of the primary schools use the Cito exam as an objective measurement (Cito, 2014). The teacher uses personal experience with the student and the Cito exam to provide advice regarding the student’s level of education in high school. Since 2015, this school advice is binding (van Spijker, van der Houwen & van Gaalen, 2017). The teacher’s experiences will not always be in line with the results in the Cito exam. This discrepancy is mostly due to teachers knowing other aspects of the child’s ability, like motivation, working attitude, perseverance and independence (Driessen, 2005). In addition, social factors like parent’s education level and socio-ethnic origin might play a role in the advice of the teacher (Driessen, 2007). If a student receives a higher track advice than the cognitive test indicates, we speak of over-advising and vice versa of under-advising (Inspectie van het Onderwijs, 2011). This thesis will investigate whether certain socio-economic groups are relatively over- or under-advised and if this has increased between the years 2008 and 2014.

Both over- and under-advising are seen as bad outcomes, if the capabilities of the student are really misjudged. Over-advising could lead to low grades, repeating years and drop-outs (Tesser & Iedema, 2001). Under-advising could result in a lower completed track level, compared to students with the same cognitive results (Inspectie van het Onderwijs, 2007).

What factors lead to a certain advice has been a topic of interest since the late ’70s. Starting from the mid ’80s, researchers emphasised the differences between over- or under-advising in different socio-economic groups (Luyten & Bosker, 2004). In the Dutch literature,

(6)

parents both are Dutch. Allochtoon refers to the group who have one parent that is of non-western descent (CBS, 2015). Much of the research on this topic uses this distinction. This paper also uses this distinction and will make use of the English terms native background and immigrant background.

The first papers on this matter, in the late ‘80s and early ‘90s, showed that given the same cognitive achievements, students with an immigrant background received a higher advice (de Jong, 1987). Koeslag and Dronkers (1994) found that 10 to 15% of the students with a Turkish or Moroccan background received a higher advice than expected based on their cognitive skills, compared to 6% of the students with a native background.

De Jong (1987) concluded that teachers weigh the possible potential higher than the actual capabilities. This effect was stronger for immigrant students, because teachers believed that immigrant students had a backlog that was mostly due to a language backlog and this backlog would disappear when the language skills increased later on. Jungbluth (2003) argued that the fear of discriminating caused over-advising. Another explanation is that the composition and the average results of a class could have an effect on the teacher’s advice; in a class full of students with an immigrant background, the slightly higher performing students get a disproportionately higher advice (Mulder & Tesser, 1991). Driessen (1991) found similar results and showed that, if students followed the advice, this group had a higher chance of falling behind during high school or even dropping out.

In the same period, Mulder (1993) showed that students with average cognitive skills and with native parents with a low level of education, got a below average advice. This enhanced the discrepancy between the advice for students with an immigrant and a native background.

After these findings, policy makers advised teachers to emphasise the objective test outcome more in their advice. Another measure that was taken was that secondary schools reported students who they thought were over- or under-advised. Primary schools were judged more strictly by municipality institutions upon these results (Inspectie van het Onderwijs, 2007). Moreover, over time teachers gained experience in advising and forecasting the abilities of a student with an immigrant background (Driessen, 2005).

These policy changes had their effect. In the mid ‘90s until the early 2000s, similar research on new data found different effects. Dagevors, Gijsberts and van Praag (2003) concluded after a multilevel analysis that between 1994 and 1998, the over-advising of

(7)

students with an immigrant background substantially decreased and in 2001, there was almost no sign of over-advising. Moreover, they also conducted a regression where they made a distinction between the different nationalities of the largest minority groups in the Netherlands, no sign of over- or under-advising was found when distinguishing between these nationalities. Luyten and Bosker (2004) did see a minor effect for the education level of parents on advice, but also concluded that this effect decreased over the years and the effect of cognitive achievements increased. Driessen and Smeets (2007) performed a multilevel regression with data from 2008 and concluded there was evidence, although small, for under-advising for native students with parents with a low level of education, but no sign of over- or under-advising of students with an immigrant background.

The latest research report by the Dutch Inspection for Education (Inspectie van het Onderwijs, 2016) found an increasing discrepancy between students with highly educated parents and students with parents with a low level of education in the enrolment in the high track levels of secondary education. They followed students with a comparable IQ (an IQ that corresponds with the track level VMBO G/T, a pre-vocational education track). Students with highly educated parents enrolled relatively more often into the HAVO/VWO track (pre-university education track) compared to students with low educated parents, who in most cases enrolled in VMBO tracks. This is mostly because the advice for students with low educated parents decreased in recent years.

Furthermore, the influence of ethnic background increased regarding the advice for secondary education. Additionally, the gap between students with an immigrant background and a native background in tertiary education in the Netherlands is above the European mean (Inspectie van het Onderwijs, 2016). Moreover, van de Werfhorst, Elffers and Karstens (2015) showed that compared with other Western European countries, fewer students with an immigrant background in the Netherlands start higher education. If they attend higher education, it usually takes them longer to get there than in other Western European countries (Crul, Pasztor, Lelie & Schnell, 2009).

Although, this is also due to the different composition of immigrant groups in different countries, with different language barriers, the European Commission addressed this as a challenge for the Dutch government (Europese Commissie, 2015).

(8)

Steeg & Webbink, 2011; Werfhorst, Elffers & Karstens 2015). However, the early tracking existed before the first wave of mass immigration to the Netherlands in the sixties and cannot entirely explain the recent segregation in Dutch education.

Research showed that in different times, different groups were over- or under-advised (Driessen & Smeets, 2007; Luyten & Bosker, 2004). The hypothesis in this thesis is that the recent segregation may be due to the over- or under-advising. Therefore, my research question is: ‘Do socio-economic background factors contribute to the relative over- or under-advising of students after they finish primary school and is this effect increasing?’ This is researched with probit regressions with different track advice dummies as independent variables, using data from 2008 and 2014. The advice dummies are: a pre-vocational education track or higher advice and a pre-university education track advice.

The main findings are that some socio-economic groups are relatively over- or under-advised, considering a pre-vocational education track or higher. Students with low educated native parents are under-advised, compared with students with native parents with a middle level of education. Students with immigrant parents with a middle level of education, highly educated immigrant parents and highly educated native parents are over-advised, compared to students with native parents with a middle level of education. There were no significant differences between the years 2008 and 2014, and thus no evidence was found of an increasing effect.

The next part of this thesis will explain the Dutch secondary school system. After that, the data collection and data description will be discussed. The methodology section will explain the different probit regressions, followed by the results and conclusion.

2. Context

To get to a better understanding of under- and over-advice, a quick summary of the Dutch secondary education system is given. At a relatively early age, Dutch students are tracked in different levels of education (van Elk, van der Steeg & Webbink, 2011). At age 12 (6th grade), students receive an advice for a certain track level. With this advice, students can apply for a high school that offers the corresponding track. This can either be a comprehensive school with multiple tracks or a categorial school with one track level (van Elk, van der Steeg &

(9)

Webbink, 2011). Most comprehensive schools have the so called bridge years, consisting of the first one or two year(s), in which the track levels are mixed. The education levels are roughly divided into four preparatory tracks for further education: VWO (pre-university education), which takes six years, HAVO (pre-college education), which takes five years and VMBO (pre-vocational education) and PO (practical education), which take four years. VMBO is further divided into four categories: B, K, G and T, where K, G and T have an increasing theoretical focus. After a certain point, there are inflow possibilities for continuing to a higher track. After students finish a track, they can apply for the corresponding tertiary education or can do a switch year for a higher level of tertiary education.

Figure 1: Flow of Dutch education system.

Dutch education system (Korthals, 2015). Retrieved from:Tracking Students in Secondary Education, Consequences for Student Performance and Inequality, on 12/07/2017

3. Data

Most of the research mentioned in the introduction Section, relied on the PRIMA-dataset and the later successor, the COOL5-18 dataset (COOL). COOL5-18 is a joint data collection program by the University of Nijmegen, University of Amsterdam, University of Groningen and Cito.

(10)

One of the aims of the COOL data is to follow the cognitive development of students from 5 to 18 years old.

A survey is distributed to students in pre-school, third grade, sixth grade and ninth grade. The student, teacher and parent(s) have to fill in questions about the (family) background, social competences, attitude toward school and educational performance. Additionally, the students have to take a cognitive test to measure reading, writing and math skills.

The COOL data consists of data from 2008, 2011 and 2014. Every dataset consists of around 50,000 students of different age categories. The schools are selected in accordance with a representative sample of the population regarding religion, degree of urbanization, region and social-ethnic composition of the school. The data used in this research consists of students in their last grade of primary school in 2008 and 2014. Variables of interest are:

- Advice dummies

 VWO: student received a VWO advice

 Higher VMBOT: student received at least a VMBO-t advice

- Cito, continuous variable of test scores ranging from 501 to 550. Below, test scores with their corresponding track level as stated by Cito are given to give an indication of the score values (Cito, 2014).

1 VMBO B 501-523

2 VMBO K 524-528

3 VMBO G/T 529-536

4 HAVO 537-544

5 VWO 545-550

- Socio-economic status dummies

 Low immigrant: students with non-western immigrant parents whose highest education level is pre-vocational secondary education.

 Low native: students with native parents whose highest education level is pre-vocational secondary education.

(11)

 Middle immigrant: students with non-western immigrant parents whose highest education level is upper secondary vocational education

 Middle native: students with native parents whose highest education level is upper secondary vocational education

 High immigrant: students with non-western immigrant parents whose education level is higher professional education level or higher

 High native: students with native parents whose education level is higher professional education level or higher

- Non-cognitive student profile variables  Level of under achievement  Good behaviour  Working attitude  Level of dependence  Conflict  Social interaction  Parental involvement

The student profile variables were developed to measure the teacher’s view on the student’s behaviour, relation with the teacher and other students, and the parental involvement. Each variable consists of three or four questions regarding the topic. The teacher could choose between the following answers: (1) absolutely not true, (2) not true, (3) neither true nor not true, (4) true and (5) absolutely true. Based on these answers, students received a score ranging from 1 to 5 for each of the variables. For example, a 5 for good behaviour means that the teachers think a student displays good behaviour and a 5 for conflict means that a teacher and the student often have conflicts.

As stated in the introduction, the non-cognitive skills may have a large influence on the teacher’s decision on advice. The first four variables in this category are mainly focused on the teacher’s view on motivation, perseverance and working attitude. The last three variables focus on the relation between the student and the teachers. These subjective variables are

(12)

also indicators for the subjective part of the teacher’s advice. For this, it is important to include these variables in the analysis.

3.1 Descriptive data

Table 1 presents the sample statistics for the 2008 and 2014 dataset. There are significant differences at a 1% level in almost all variables between the two years. As the COOL data are collected by selecting schools in accordance with the same ratio of different schools in the population, these differences in characteristics may be due to the changing composition of schools in the Netherlands. This has to be kept in mind when comparing regressions in different years.

Table 1. Sample statistics for the COOL 2008 dataset and COOL 2014 dataset

2008 n= 6967 2014 n=5167 p-value mean SD mean SD Low immigrant 0.15 0.36 0.09 0.29 0.000 Low native 0.17 0.37 0.10 0.30 0.000 Middle immigrant 0.07 0.26 0.07 0.26 0.811 Middle native 0.34 0.47 0.38 0.49 0.000 High immigrant 0.03 0.18 0.04 0.19 0.667 High native 0.23 0.42 0.32 0.47 0.000 Male 0.50 0.50 0.46 0.50 0.000 Cito score 532.89 10.31 533.91 10.50 0.000

Note: the table presents means and standard deviations of the social economic dummies, sex dummy and Cito score in the 2008 and 2014 dataset, unless stated otherwise. The socio-economic variables are determined by ethnicity and educational level of the parents. The Cito score is an objective cognitive test. The p-values in the final column are for tests of the difference of the variables in 2008 and 2014.

Table 2 presents the means of the socio-economic dummies for VMBO-t advice or higher in 2008 and 2014. Almost none of the differences in means are significant between the years. There is only a significant difference between the means of students having native parents with a middle level of education at a 5% level. This is a small indication of segregation between

(13)

students with a native background and students with an immigrant background in Dutch secondary education.

Table 2. Means of the socio-economic groups with VMBO-t or higher advice

2008 2014 p-value mean SD mean SD Low immigrant 0.46 0.02 0.49 0.02 0.347 Low native 0.45 0.01 0.43 0.02 0.643 Middle immigrant 0.59 0.02 0.65 0.03 0.105 Middle native 0.67 0.01 0.65 0.03 0.053 High immigrant 0.72 0.03 0.78 0.03 0.203 High native 0.86 0.01 0.87 0.01 0.652

Note: the table presents the means and standard deviations of the students in socio-economic groups who have received VMBO-t or higher track advice. The socio-economic groups are determined by ethnicity and educational level of the parents. The p-values in the final column are for tests of the difference of the variables in 2008 and 2014.

Table 3 presents the means of the socio-economic dummies for a VWO advice. There are significant differences in means between 2008 and 2014 for students with native parents with middle and high levels of education, and students with highly educated immigrant parents at a 1% level and immigrant parents with a low level of education at a 5% level. There are more students in these groups with a VWO advice in 2014 than in 2008. This may be due to the fact that there was a significantly higher mean Cito score in 2014. Only the change of the mean of the highly educated immigrant group was significantly higher than the other changes. These increases in means do not clearly explain the increasing segregation as described by the Dutch Inspection for Education (2016).

(14)

Table 3. Means of the socio-economic groups with VWO advice 2008 2014 p-value mean SD mean SD Low immigrant 0.04 0.01 0.07 0.01 0.015 Low native 0.04 0.01 0.05 0.01 0.280 Middle immigrant 0.09 0.01 0.10 0.02 0.690 Middle native 0.10 0.01 0.13 0.01 0.001 High immigrant 0.13 0.02 0.23 0.03 0.006 High native 0.26 0.01 0.31 0.01 0.004

Note: the table presents the means and standard deviations of the students in socio-economic groups who have received a VWO track advice. The socio-economic groups are determined by ethnicity and educational level of the parents. The p-values in the final column are for tests of the difference of the variables in 2008 and 2014.

4. Empirical method

The starting point of this research is the report by the Dutch Inspection Bureau of Education (2011), by Driessen, who uses the COOL 2008 data to estimate effects of performance and background variables on advice. Like many papers discussed in the introduction Section, Driessen uses a multi-level analysis. This method is often used in social data analysis to correct for the different levels of variation in the dataset (Rasbash, Steele, Browne & Goldstein, 2009). The methodology in this thesis uses clustered errors on a school level. Not clustering errors will in general cause standard errors of regression coefficients to be underestimated (Rasbash, et al., 2009).

Another factor that is different in this methodology compared with previous work is the dependent variable. Driessen uses an indexed advice variable ranging from 1 to 5, where 1 stands for low advices and 5 for high advices. The assumption, made by Driessen, that the grouped track advices increase linearly in difficulty is open for discussion.

For this analysis, a different approach is chosen. The analysis consists of two different regressions, with two dependent dummy variables: the student received at least a VMBO-t advice and the student received a VWO advice. In this way, the assumption that the given advices are linear in difficulty is bypassed. By conducting two analyses, we can see whether there is a difference in the over- or under-advising for higher or lower track levels.

(15)

With a binary dependent variable, a linear probability model or a logit/probit model must be chosen. The linear probability model has some flaws. Probabilities exceed 1 for high values and drops below 0 for low values, which is problematic considering probability range from 0 to 1 (Stock & Watson, 2012). Moreover, the linear probability regression line has a miss-specified functional form for a dichotomous depended variable (Mood, 2010). Probit and logit are non-linear regressions designed for binary variables that force the predicted values between 0 and 1 (Stock & Watson, 2012). The estimated regression lines in a probit or logit regression are very similar if not identical (Stock & Watson, 2012). The models seek the probability of an event occurring depending on the values of the independent variables, which can be categorical or numerical. This paper uses a probit regression.

The probit regression model has a cumulative standard normal distribution function and the coefficients of the model are estimated by maximum likelihood. In large samples, the maximum likelihood estimators are consistent and normally distributed. The general function is:

Pr(𝑌 = 1|𝑋

1

, 𝑋

2

, . . , 𝑋

𝑘

) = 𝝓(𝛽

0

+ 𝛽

1

𝑋

1

+ 𝛽

2

𝑋

2

+ ⋯ + 𝛽

𝑘

𝑋

𝑘

) = ∫

−∞𝑥′𝛽

𝝓(𝑧)𝑑𝑧

(1) The estimated coefficients and the variable values give us a Z value. Using a cumulative normal distribution table, one can use the Z value and find the probability for a score of 1, so a change in the variables gives an expected change in the probability that Y = 1 (Stock & Watson, 2012).

To interpret the coefficients of the probit model, we look at the marginal effects of the coefficients. The marginal effects reflect the probability change of Y=1 given a unit change of one of the independent variables X. The marginal effects are calculated by taking the derivative of equation (1) with respect to the variable of interest, as follows (Hammer & Kalkan, 2013):

𝜕𝑝

𝜕𝑥𝑗

= 𝝓(𝑋

𝛽)𝛽

𝑗

(2)

The analysis starts with model 1, where only the socio-economic status variables and sex dummy are regressed against the binary advice variables. In the second model, we add the Cito score, as dummies for every possible score. In the third model, the non-cognitive

(16)

student profile variables are included as well. By adding variables in each model, we can analyse the change in magnitude and significance of the coefficients of the socio-economic variables. If the coefficients of the socio-economic variables in the third model are still significant, we can speak of evidence of the relative over- or under-advising in the different socio-economic groups. As reference group, we take the students with a native background and parents with a middle level of education. This analysis will be conducted for the two years. As mentioned, for every year, two dependent dummy variables are analysed.

First model:

= 𝝓(𝛽

0

+ 𝛽

1

𝐿𝑜𝑤

𝑖𝑚𝑚𝑖𝑔𝑟𝑎𝑛𝑡

+ 𝛽

2

𝐿𝑜𝑤

𝑛𝑎𝑡𝑖𝑣𝑒

+ 𝛽

3

𝑚𝑖𝑑𝑑𝑒

𝑖𝑚𝑚𝑖𝑔𝑟𝑎𝑛𝑡

+

𝛽

4

𝐻𝑖𝑔ℎ

𝑖𝑚𝑚𝑖𝑔𝑟𝑎𝑛𝑡

+ 𝛽

5

𝐻𝑖𝑔ℎ

𝑛𝑎𝑡𝑖𝑣𝑒

+ 𝛽

6

𝑏𝑜𝑦)

(3) Second model:

= 𝝓(𝛽

0

+ 𝛽

1

𝐿𝑜𝑤

𝑖𝑚𝑚𝑖𝑔𝑟𝑎𝑛𝑡

+ 𝛽

2

𝐿𝑜𝑤

𝑛𝑎𝑡𝑖𝑣𝑒

+ 𝛽

3

𝑚𝑖𝑑𝑑𝑒

𝑖𝑚𝑚𝑖𝑔𝑟𝑎𝑛𝑡

+

𝛽

4

𝐻𝑖𝑔ℎ

𝑖𝑚𝑚𝑖𝑔𝑟𝑎𝑛𝑡

+ 𝛽

5

𝐻𝑖𝑔ℎ

𝑛𝑎𝑡𝑖𝑣𝑒

+ 𝛽

6

𝑏𝑜𝑦 + ∑

50𝑘=1

𝛽

𝑘

𝐶𝑖𝑡𝑜𝑑𝑢𝑚𝑚𝑦)

(4) Third model:

= 𝝓 (𝛽

0

+ 𝛽

1

𝐿𝑜𝑤

𝑖𝑚𝑚𝑖𝑔𝑟𝑎𝑛𝑡

+ 𝛽

2

𝐿𝑜𝑤

𝑛𝑎𝑡𝑖𝑣𝑒

+𝛽

3

𝑚𝑖𝑑𝑑𝑙𝑒

𝑖𝑚𝑚𝑖𝑔𝑟𝑎𝑛𝑡

+

𝛽

4

𝐻𝑖𝑔ℎ

𝑖𝑚𝑚𝑖𝑔𝑟𝑎𝑛𝑡

+ 𝛽

5

𝐻𝑖𝑔ℎ

𝑛𝑎𝑡𝑖𝑣𝑒

+ 𝛽

6

𝑏𝑜𝑦 + ∑

50𝑘=1

𝛽

𝑘

𝐶𝑖𝑡𝑜𝑑𝑢𝑚𝑚𝑦 +

𝛽

7

𝑈𝑛𝑑𝑒𝑟 𝑎𝑐ℎ𝑖𝑒𝑣𝑒𝑚𝑒𝑛𝑡 + 𝛽

8

𝐺𝑜𝑜𝑑 𝑏𝑒ℎ𝑎𝑣𝑖𝑜𝑢𝑟 + 𝛽

9

𝑊𝑜𝑟𝑘𝑖𝑛𝑔 𝑎𝑡𝑡𝑖𝑡𝑢𝑑𝑒 +

𝛽

10

𝐼𝑛𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑐𝑒 + 𝛽

11

𝐶𝑜𝑛𝑓𝑙𝑖𝑐𝑡 + 𝛽

12

𝑆𝑜𝑐𝑖𝑎𝑙 + 𝛽

12

𝑃𝑎𝑟𝑒𝑛𝑡 𝑖𝑛𝑣𝑜𝑙𝑣𝑒𝑚𝑒𝑛𝑡)

(5)

5. Results

This section presents the results in 2008 (Section 5.1) and 2014 (Section 5.2). Followed by a discussion on the student profile variables results (Section 5.3) and main results (Section 5.4).

(17)

5.1 Results 2008

Table 4 presents the marginal effects for the two dependent variables in the three models for 2008. The 50 Cito dummies in models 2 and 3 are kept from the regression table to keep it uncluttered.

In the first model with the VMBO-t or higher advice dependent variable, all socio-economic dummies are significant, at least at the 10% confidence level. Compared to students with native parents with a middle level of education (this applies to all socio-economic variable results in every model), students with immigrant parents with a middle level of education, native parents with a low level of education and students with immigrant parents with a low level of education have an 8.0, 22.7 and 21.8 percentage points lower chance respectively to receive VMBO-t or higher advice. Students with highly educated parents with an immigrant or native background have a 5.0 and 21.1 percentage point higher chance respectively to receive VMBO-t or higher advice.

The only conclusion we can draw from these results is whether students from a certain social group have more or less chances of receiving higher advice than VMBO-t, compared to students with native parents with a middle level of education.

When adding the Cito dummies, the difference in the chances decreases; in the case of immigrant parents with a low level of education, the significance disappears. This means that Cito score is truly a factor that influences the advice. The direction of the coefficient for students with immigrant parents with a middle level of education changes. In other words, although they have a smaller chance of receiving VMBO-t or higher advice, given the Cito score they are over-advised compared to the reference group. When the non-cognitive variables are added in the third model, students with native parents with a low level of education still have a 3.9 percentage point lower chance of receiving VMBO-t advice, compared to students with native parents with a middle level of education. Students with immigrant parents with a middle level of education, highly educated immigrant parents or highly educated native parents still have a 3.0, 3.8, and 4.1 percentage point higher chance respectively of receiving VMBO-t or higher advice.

For a VWO advice, going from model 1 to 2 all the socio-economic variables lose significance. In model 3, when the non-cognitive variables are added, none of these variables are significant, thus indicating that the chance of receiving a VWO advice in this model is solely

(18)

Table 4 Regressions in the 2008 dataset

(1) (2) (3) (1) (2) (3)

VARIABLES Higher VMBO-t Higher VMBO-t Higher VMBO-t VWO VWO VWO Low immigrant -0.212*** -0.000627 0.00431 -0.0688*** 2.69e-05 3.49e-05

(0.0232) (0.0162) (0.0154) (0.00913) (9.48e-05) (7.10e-05) Low native -0.227*** -0.0524*** -0.0393** -0.0742*** -9.39e-05 -3.85e-05

(0.0197) (0.0173) (0.0157) (0.00839) (7.77e-05) (3.95e-05) Middle immigrant -0.0802*** 0.0324** 0.0304** -0.0144 0.000300 0.000182 (0.0257) (0.0160) (0.0147) (0.0151) (0.000310) (0.000215) High immigrant 0.0504* 0.0445*** 0.0383** 0.0218 0.000231 0.000121 (0.0302) (0.0168) (0.0159) (0.0223) (0.000265) (0.000172) High native 0.212*** 0.0444*** 0.0412*** 0.135*** 0.000162 8.17e-05

(0.0164) (0.0128) (0.0122) (0.0154) (0.000114) (7.96e-05) Sex (male = 1) -0.0121 -0.00966 0.00897 -0.00441 3.32e-05 2.89e-05

(0.0120) (0.00825) (0.00849) (0.00734) (4.03e-05) (3.57e-05)

Cito dummies no yes yes no yes yes

Under achievement 0.0158** -3.40e-05 (0.00618) (2.50e-05) Behaviour -0.0144 -2.13e-05 (0.00893) (1.83e-05) Working attitude 0.0436*** 2.63e-05 (0.00962) (3.28e-05) Dependence -0.00303 -5.86e-05 (0.00666) (5.23e-05) Conflict -0.000659 4.05e-05 (0.00885) (5.01e-05)

(19)

(1) (2) (3) (1) (2) (3) VARIABLES Higher VMBO-t Higher VMBO-t Higher VMBO-t VWO VWO VWO Social interaction 0.00734 -6.12e-06 (0.00730) (1.58e-05) Parental involvement 0.0139** 2.91e-05 (0.00552) (3.47e-05) Pseudo R2 0.0819 0.5789 0.5880 0.0887 0.5667 0.5759 Observations 6,967 6,967 6,967 6,967 6,967 6,967 Note: each column reports results from a separate regression. The first three regressions have the dependent variable VMBO-t or higher track advice, the last three regressions have the dependent variable VWO track advice. The first five variables are socio-economic dummies, determined by ethnicity and educational level of the parents. Cito dummies are dummies for the 50 possible Cito scores, a cognitive test. The remaining variables are non-cognitive variables. Standard errors (between parentheses) are clustered at the school level. * significant at the 10% level, ** significant at the 5% level, *** significant at the 1% level.

5.2 Results 2014

For the 2014 regression with the dependent variable VMBO-t or higher advice, Table 5 displays the following results. In the first model with VMBO-t or higher advice as dependent variable, all socio-economic variables are significant at a 1% level except for an immigrant group with parents with a middle level of education. Compared to students with native parents with a middle level of education, students with parents with a low level of education have less chances of receiving VMBO-t or higher advice. There is no significant difference between the two groups with parents with a middle level of education in receiving VMBO-t or higher advice. The both socio-economic groups with highly educated parents have relatively higher chances of receiving VMBO-t or higher advice.

When the Cito score dummies are added, the group with immigrant parents with a low level of education has no significantly different chances of receiving VMBO-t or higher advice compared to the reference group. Given the Cito score, students with native parents with a low level of education are under-advised, compared to the reference group. In the first model there were no significant differences between the two groups with parents with a middle level of education. When the Cito score dummies are added, the students with immigrant parents with a middle level of education have a higher chance of receiving VMBO-t advice, meaning that given the Cito score they are relatively over-advised. The group with highly educated native parents is still over-advised compared to students with native parents with a middle

(20)

level of education, although the influence of the socio-economic background is less strong, compared to the first model. Students with highly educated immigrant parents, on the other hand, do not have a relatively higher chance anymore when the Cito score is added. Given the Cito score they receive the same advice as students with native parents with a middle level of education.

In model three, when all the variables are added, most of the socio-economic variables lose the magnitude of the effect and significance compared to model 2, except for the group with highly educated immigrant parents, which in this model is again significant. This means that given the non-cognitive variables, they are over-advised compared to the reference group. In short, students with native parents with a low level of education have a 4.1 percentage point lower chance of receiving VMBO-t advice or higher and are under-advised compared to students with native parents with a middle level of education. Students with immigrant parents with a middle level of education, highly educated immigrant parents and highly educated native parents have 2.8, 2.8 and 4.2 percentage point higher chance respectively of receiving VMBO-t advice or higher, and are over-advised compared to students with native parents with a middle level of education.

For the models with the VWO advice dummy as dependent variable, similar results are found as in the 2008 models. In the first model, compared to students with native parents with a middle level of education, the students with parents with a low level of education and students with immigrant parents with a middle level of education have less chances of receiving VWO advice. Students with highly educated parents have a relatively higher chance of receiving VWO advice. When the Cito score variables are added, all variables lose significance. This implies that given the Cito score, none of the socio-economic groups are over- or under-advised compared to the reference group. Although some of the student profile variables are significant in the third model, these effects are negligibly small.

(21)

Table 5. Regressions in the 2014 dataset

(1) (2) (3) (1) (2) (3)

VARIABLES Higher VMBO-t Higher VMBO-t Higher VMBO-t VWO VWO VWO

Low immigrant -0.151*** -0.00437 0.00179 -0.0795*** 2.91e-05 0.000173 (0.0314) (0.0167) (0.0148) (0.0168) (0.000456) (0.000301) Low native -0.203*** -0.0573*** -0.0407** -0.107*** -0.000520 -0.000197 (0.0261) (0.0203) (0.0180) (0.0130) (0.000343) (0.000153) Middle immigrant 0.00385 0.0285** 0.0282** -0.0437** -0.000346 -0.000120 (0.0302) (0.0128) (0.0115) (0.0203) (0.000336) (0.000157) High immigrant 0.119*** 0.0287 0.0284* 0.103*** 0.000940 0.000664 (0.0299) (0.0184) (0.0159) (0.0353) (0.000996) (0.000614) High native 0.232*** 0.0501*** 0.0416*** 0.163*** 0.000689 0.000282 (0.0165) (0.0101) (0.00901) (0.0155) (0.000464) (0.000203) Sex (male = 1) -0.0107 -0.00223 0.00354 0.0123 0.000261 8.46e-05

(0.0142) (0.00832) (0.00831) (0.00997) (0.000221) (9.78e-05)

Cito dummies no yes yes no yes yes

Under achievement 0.0117** -0.000151* (0.00520) (8.47e-05) Behaviour -0.0200*** -0.000266* (0.00724) (0.000151) Working attitude 0.0253*** 0.000138 (0.00724) (9.80e-05) Dependence -0.00676 -0.000178 (0.00657) (0.000117) Conflict -0.00130 -5.55e-07 (0.00748) (9.89e-05)

(22)

(1) (2) (3) (1) (2) (3) VARIABLES Higher VMBO-t Higher VMBO-t Higher VMBO-t VWO VWO VWO Social interaction -0.00845 -0.000259* (0.00741) (0.000139) Parental involvement 0.0241*** 0.000343** (0.00640) (0.000172) Pseudo R2 0.0830 0.5723 0.5817 0.0725 0.5777 0.5940 Observations 5,167 5,167 5,167 5,167 5,167 5,167 Note: each column reports results from a separate regression. The first three regressions have the dependent variable VMBO-t or higher track advice, the last three regressions have the dependent variable VWO track advice. The first five variables are socio-economic dummies, determined by ethnicity and educational level of the parents. Cito dummies are dummies for the 50 possible Cito scores, a cognitive test. The remaining variables are non-cognitive student profile variables. Standard errors (between parentheses) are clustered at the school level. * significant at the 10% level, ** significant at the 5% level, *** significant at the 1% level.

5.3 Non-cognitive student profile variables

Some results stand out in the third models in both years with the dependent variable VMBO-t, considering the student profile variables. Parental involvement and working attitude both have a significant positive effect, meaning that a one-unit higher score on these scales contributes to a higher percentage change (given by the coefficients) for receiving VMBO-t or higher advice.

One striking result is that the variable for underachievement also has a positive effect in both cases. According to the technical report of the COOL data collection, a high score for the underachievement variable corresponds to a higher degree of underachievement. Intuitively, this will lead to a lower advice. According to these models, this is not the case.

Another result is that the sex of the student is insignificant in every model.

5.4 Main results

In both years there is similar evidence of over- and under-advising compared to students with native parents with a middle level of education. The over- or under-advising of students compared to the reference groups does not happen for the highest track advice VWO.

For a VMBO-t or higher advice, in both years, students with immigrant parents with a low level of education are not relatively over- or under-advised. Students with native parents with a low level of education are under-advised. Students with immigrant parents with a middle level of education, highly educated immigrant parents and highly educated native

(23)

parents are all over-advised, compared to students with native parents with a middle level of education.

Table 7 presents the significant coefficients from both third models in 2008 and 2014, with the dependent variable VMBO-t or higher advice. The starting point of the research was the possible increase of over-advising or under-advising in different groups as an explanation for the increasing segregation in secondary education between students with a native and immigrant background, and the segregation between students with highly educated parents and parents with a low level of education. The previously discussed results present an indication of over- and under-advising of different groups compared to students with native parents with a middle level of education. When the coefficients for the socio-economic group variables in the different years were tested for differences, no significant results were found. This indicates that there is no sign of an increase in over- or under-advising students of different socio-economic groups compared to the reference group.

To analyse the possible different outcomes further, Yun’s (2004) decomposition method for a binary dependent variable, based on Blinder-Oaxaca decomposition methods, is used. Decomposition methods divide the change of the effect in a part that is attributed to a change in characteristics (for example in demographics) and a change in coefficients (the difference in behavioural response to a variable). This is done by an equation based on one group’s characteristics and the estimated coefficients of the other group’s equation. The general function for a detailed decomposition equation for a probit regression may be written as follows (Yun, 2004):

𝑌̅1− 𝑌̅2 = ∑𝑖=𝑘𝑖=1𝑊∆𝑥𝑖 [

𝝓

̅̅̅̅̅̅̅̅̅̅̅ − (𝑋1𝛽1) ̅̅̅̅̅̅̅̅̅̅̅] + ∑

𝝓

(𝑋2𝛽1) 𝑖=𝑘𝑖=1𝑊∆𝛽𝑖 [

𝝓

̅̅̅̅̅̅̅̅̅̅̅ −(𝑋2𝛽1)

𝝓

̅̅̅̅̅̅̅̅̅̅̅̅] (𝑋2𝛽2) (6)

The above bar represents the average of the samples, 𝑋1,2 and 𝛽1,2 are matrices of the variables and a vector of the coefficients in the regressions, respectively. To estimate the detailed decomposition of the coefficients and characteristics, one must weigh the contribution of each variable to the characteristics and coefficients. This is done by 𝑊∆𝑥,𝛽𝑖 ; detailed calculations are presented in Appendix A. Furthermore, ∑𝑖=𝑘𝑖=1𝑊∆𝑥𝑖 [

𝝓

̅̅̅̅̅̅̅̅̅̅̅ −(𝑋1𝛽1)

𝝓

(𝑋2𝛽1)

(24)

The results of the detailed decomposition are presented in Table 7. In the column that presents the overall results, we can see that the difference in mean outcome, attributed by the change in coefficients, is not significant. The detailed decompositions of the change in the socio-economic coefficients are also insignificant. Changes in the effect are not due to the change in behavioural response to a variable, but rather due to changes in composition.

Table 6. Significant coefficients in the third model with dependent variable VMBO-t

2008 2014 p-value Coefficients Coefficients Low native -0.0393** -0.0407** 0.756 (0.0157) (0.0180) Middle immigrant 0.0304** 0.0282** 0.839 (0.0147) (0.0115) High immigrant 0.0383** 0.0284* 0.889 (0.0159) (0.0159) High native 0.0412*** 0.0416*** 0.626 (0.0122) (0.0090)

Note: Table 6 presents the significant coefficients and the robust standard errors (in parentheses) of the socio-economic variable dummies in the regression with the dependent variable VMBO-t or higher for the years 2008 and 2014. The socio-economic variables are determined by ethnicity and educational level of the parents. The p-values in the final column are for tests of the difference of the coefficients in 2008 and 2014. Standard errors (between parentheses) are clustered at the school level. * significant at the 10% level, ** significant at the 5% level, *** significant at the 1% level.

(25)

Table 7. Decomposing changes in VMBO-t or higher advice between 2008 and 2014

VARIABLES overall characteristics coefficients

𝑌( ̅̅̅2008) 0.640*** (0.0120) 𝑌( ̅̅̅2014) 0.686*** (0.0148) difference -0.0465*** (0.0178) Characteristics -0.0428*** (0.0135) Coefficients -0.00373 (0.00997)

Low immigrant 0.000236 -1.29e-05

(0.000535) (0.000250)

Low native -0.00194*** -5.80e-05

(0.000580) (0.000292)

Middle immigrant 2.98e-05 3.41e-05

(0.000231) (0.000193)

High immigrant -4.39e-05 -1.72e-05

(0.000134) (0.000130) High native -0.00329*** 0.000220 (0.000995) (0.000850) Constant 0.0931 (0.283) Observations 12,134 12,134 12,134

Note: Table 7 presents the results of the decomposition of the overall results and the detailed decomposition of the socio-economic variable dummies in the regression with the dependent variable VMBO-t or higher for the years 2008 and 2014. The socio-economic variables are determined by ethnicity and educational level of the parents. 𝑌(̅̅̅2008) presents the mean outcome of the regression in 2008 and 𝑌(̅̅̅2014) the mean outcome in 2014. Standard errors (between parentheses) are clustered at the school level. * significant at the 10% level, ** significant at the 5% level, *** significant at the 1% level.

6. Conclusion and Discussion

The research question in this thesis is a broadly researched topic that has gained new interest due to recent signs of segregation in Dutch secondary and tertiary education. This paper investigated over- and under-advising with data from 2008 and 2014 from the COOL dataset, as a possible explanation for this fact. In addition, a different approach was used than in previous research. Instead of a linear regression model with the different advice as dependent variables, two different analyses for two different binary dependent variables for a certain

(26)

track advice (or higher) were conducted by computing a probit regression and using the marginal effects to interpret the results. In this way, the assumption that the different advice is linear in difficulty was bypassed, and it was shown whether over- or under-advising occurred more often in lower or higher track advice.

The hypothesis was that when the socio-economic variables are insignificant in model 3, we cannot speak of over- or under-advising, and we have evidence that the decision on track level made by the teacher is based on cognitive and non-cognitive factors, and not influenced by socio-economic background. With significant socio-economic variables in model 3, there is evidence of over- or under-advising students with a certain background compared to the reference group: students with native mid-level educated parents.

The main finding is that there is no evidence for over- or under-advising compared to students with native parents with a middle level of education in both years for the highest track level VWO. There is however evidence for relative over- and under-advising in both years for receiving VMBO-t or higher advice, and the results are similar in both years.

Students with native immigrant parents with a low level of education are under-advised compared to students with native parents with a middle level of education. Students with immigrant parents with a middle level of education, highly educated immigrant parents and highly educated native parents are over-advised compared to students with native parents with a middle level of education. In both years, students with immigrant parents with a low level of education are neither relatively under- or over- advised in receiving VMBO-t advice or higher.

There is no evidence that over- or under-advising has increased over the years. Therefore, the current segregation in Dutch secondary education cannot be explained by the possible over- or under-advising of students of different socio-economic groups, compared to the reference group. Moreover, in the used data, no increase in segregation between students with highly educated parents and parents with a low level of education, or increase in segregation between students with immigrant and native parents was found. The percentages of socio-economic groups who received VMBO-t advice or higher were all equal over the years, except for students with native parents with a middle level of education. Furthermore, there is evidence of an increase in percentages of socio-economic groups with VWO advice; this increase was found for students with immigrant parents with a low level of education as well as for students with native parents with a middle level of education, highly educated

(27)

immigrant parents and highly educated native parents. Only the increase of the highly educated immigrant group is significantly different from the others. These mixed results cannot be used as an argument for the recent segregation, as described by the Dutch Inspection for Education (2016).

Yet, there is evidence for relative over- and under-advising. Over-advising can lead to high dropout rates (Tesser & Iedema, 2001). Under-advising can result in students not achieving their full potential and completing lower secondary education tracks, compared to students with the same cognitive results (Inspectie van het Onderwijs, 2007). A suggestion for further research is to investigate whether the over-advised socio-economic groups found in this study experience higher dropout rates, when considering track levels higher than VMBO-t. Considering the students with native parents with a low level of education, a further research possibility could be to investigate whether they enrol more often in a higher track level after the first completed secondary education track. When this is not the case, this can be seen as an indication that under-advising leads to students with native parents with a low level of education not reaching their full potential.

There are some limitations to this research. The main one is the sample. Although the COOL dataset collected data by selecting schools based on characteristics and in the same ratio as the population, we could not control which students handed in a completed survey. The selected schools are a representative sample of the population; this is maybe not the case for the students.

The COOL data set had variables on math, reading and writing test scores. These tests were used and development by the researches of the COOL data set and were not part of the ordinary curriculum of the students. Part of the teacher’s decision is based on previous cognitive results, other than the Cito score. These results are not included in the COOL data set, but could be an interesting factor to include in future research.

When comparing coefficients changes between two probit regressions with marginal effects over time, there is a strong assumption that the unobserved heterogeneity is the same across the compared points in time (Mood, 2010). Mood (2010) argues that when dealing with binary dependent variables there is no simple all-purpose solution to the problems of interpretability and comparison of effect estimates in regressions. Probit regressions capture the non-linearity of the relation, but one must make strong assumptions to compare effects

(28)

but are better comparable across samples. Mood (2010) proposes to report more than one type of estimate. Table 8 in Appendix B reports the significant coefficients of the third model with the dependent variable VMBO-t in 2008 and 2014, when using a linear probability model. In the probit and linear probability model, the same coefficients are significant and have the same direction. Moreover, none of the coefficients differ between 2008 and 2014 in the linear probability model.

References

Boer, H. D., Bosker, R., & Werf, M. van der (2007). De gevolgen van onder-en overadvisering. In: Inspectie van het Onderwijs, Onderadvisering in beeld. Utrecht: Inspectie van het Onderwijs.

CBS (2015). Allochtonen. Retrieved from

http://www.cbs.nl/nlNL/menu/-themas/dossiers/allochtonen/methoden/begrippen/default.htm?ConceptID=37, on

[18/07/2017].

CITO (2014). Cito dossier Eindtoets 2014. Arnhem: CITO.

Crul, M., Pasztor, A., Lelie, F., Mijs, J., & Schnell, P. (2009). De lange route in internationaal

vergelijkend perspectief: Tweede generatie Turkse jongeren in het onderwijs in Nederland, België, Duitsland, Frankrijk, Oostenrijk, Zwitserland en Zweden. Amsterdam: Instituut voor

Migratie en Etnische Studies.

Dagevos, J., Gijsberts, M., & Praag, C. van (2003). Rapportage minderheden

2003: Onderwijs, arbeid en sociaal-culturele integratie. Den Haag: Sociaal en Cultureel

Planbureau.

Driessen, G. (1991). Discrepanties tussen toetsresultaten en doorstroomniveau. Positieve discriminatie bij de overgang basisonderwijs—voortgezet onderwijs? Pedagogische Studiën. 68(1), 27-35.

(29)

Driessen, G. (2005). De totstandkoming van de adviezen voortgezet onderwijs: Invloeden van thuis en school. Pedagogiek, 25(4), 279-298.

Driessen, G. (2007). ‘Peer group’ effecten op onderwijsprestaties. Een international review van

effecten, verklaringen en theoretische en methodologische aspecten. Nijmegen: ITS.

Driessen, G., & Smeets, E. (2007). De relatie tussen prestaties en advies. Onder- of

overadvisering bij de overgang van basis- naar voortgezet onderwijs? In: Inspectie van het

Onderwijs, Onderadvisering in beeld. Utrecht: Inspectie van het Onderwijs.

Elk, R. van, Steeg, M. van der, & Webbink, D. (2011). Does the timing of tracking affect higher education completion? Economics of Education Review, 30(5), 1009-1021.

Europese Commissie (2015). Onderwijs- en opleidingenmonitor 2015. Nederland. Luxemburg: Bureau voor publicaties voor de Europese Unie.

Hanmer, M. J., & Ozan Kalkan, K. (2013). Behind the curve: Clarifying the best approach to calculating predicted probabilities and marginal effects from limited dependent variable models. American Journal of Political Science, 57(1), 263-277.

Inspectie van het Onderwijs (2007). Onderadvisering in beeld. Utrecht: Inspectie van het Onderwijs.

Inspectie van het Onderwijs (2011). Onderadvisering van allochtone leerlingen? Utrecht: Inspectie van het Onderwijs.

Inspectie van het Onderwijs (2014). De kwaliteit van het basisschooladvies. Utrecht: Inspectie van het Onderwijs.

(30)

Jong, M. J. de, (1987). Herkomst, Kennis en kansen. Allochtone en autochtone leerlingen tijdens

de overgang van basis- naar voortgezet onderwijs. Lisse: Swets & Zeitlinger.

Jungbluth, P. (2003). De ongelijke basischool. Etniciteit, social milieu, sekse, verborgen

differentiatie, segregatie, onderwijskansen en schooleffectiviteit. Nijmegen: ITS.

Koeslag, M., & Dronkers, J. (1994). Overadvisering en de schoolloopbanen van migrantenleerlingen en autochtone leerlingen in het voortgezet onderwijs. Tijdschrift voor

Onderwijsresearch, 19 (3), 240-258.

Korthaas, R., A. (2015). Dutch Education System [Figure]. Retrieved on 12/07/2017, from:

Tracking Students in Secondary Education, Consequences for Student Performance and Inequality. Maastrich: ROA.

Luyten, H., & Bosker, R. J. (2004). Hoe meritocratisch zijn schooladviezen. Pedagogische

Studiën, 81 (2), 89-103.

Mood, C. (2010). Logistic regression: Why we cannot do what we think we can do, and what we can do about it. European sociological review, 26(1), 67-82.

Mulder, L., & Tesser, P. (1991). De schoolkeuze van allochtone leerlingen. Nijmegen: ITS.

Mulder, L. (1993). Secondary school recommendations in relation to student and school characteristics. Tijdschrift voor Onderwijsresearch, 18(2), 111-119.

Rasbash, j., Steele, F., Browne, F., & Goldstein, H. (2009). A user’s guide to MLsiN. Bristol: University of Bristol.

Spijker, F. van, Houwen, K. van der, & Gaalen, R. van (2017). Invloed ouderlijk opleidingsniveau

(31)

Stock, J. H., & Watson, M. W. (2012). Introduction to econometrics (3rd edition). Boston: Addison Wesley.

Tesser, P., & Iedema, J. (2001). Rapportage minderheden 2001. Vorderingen op school. Den Haag: SCP.

Werfhorst, H. van de, Elffers, L., & Karsten, S. (2015). Onderwijsstelsels vergeleken. Leren,

werken en burgerschap. Den Haag: NRO.

Yun, M. S. (2004). Decomposing differences in the first moment. Economics letters, 82(2), 275-280.

(32)

Appendices

Appendix A

𝑊

∆𝑋𝑖

=

(𝑋̅1𝑖−𝑋̅2𝑖)𝛽1𝑖𝑓(𝑋̅1𝛽1) (𝑋̅1−𝑋̅2)𝛽1𝑓(𝑋̅1𝛽1) (7)

𝑊

∆𝛽𝑖

=

𝑋̅2𝑖(𝛽1𝑖−𝛽2𝑖)𝑓(𝑋̅2𝛽2) 𝑋̅2(𝛽1−𝛽2)𝑓(𝑋̅2𝛽2) (8)

(33)

Appendix B

Table 8. Significant coefficients in the third model using a linear probability model with

dependent variable VMBO-t

2008 2014 p-value Coefficients Coefficients Low native -0.0446*** -0.0563*** 0.615 (0.0107) (0.0148) Middle immigrant 0.0334*** 0.0403** 0.805 (0.0147) (0.0169) High immigrant 0.0390* 0.0477** 0.798 (0.0200) (0.0159) High native 0.0340*** 0.0430*** 0.537 (0.0099) (0.0101)

Note: Table 6 presents the significant coefficients and the robust standard errors (in parentheses) of the socio-economic variable dummies in the regression with the dependent variable VMBO-t or higher for the years 2008 and 2014. The socio-economic variables are determined by ethnicity and education level of the parents. The p-values in the final column are for tests of the difference of the coefficients in 2008 and 2014. Standard errors (between parentheses) are clustered at the school level. * significant at the 10 percent level, ** significant at the 5 percent level, *** significant at the 1 percent level.

Referenties

GERELATEERDE DOCUMENTEN

Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers) Please check the document version of this publication:.. • A submitted manuscript is

When making financial choices under risk, individuals thus do not significantly alter their choices, when they are in the presence of peers and they are provided

Tom Braams en Marisca Milikowski schrijven in hun voorwoord: “In deze bundel hebben wij de argumenten op een rij proberen te zetten die we in de afgelopen jaren hebben gehoord in

On another note, in the context of home-( un )making, the impact of caregiving the parents experience are less strong than exemplified in other studies where the patient

This may present challenges related to academic achievement and communication between the parents and teachers because the attitudes and expectations of education professionals are

(2011) about different semantic word representations in adults with and without dyslexia and much knowledge stays unclear, we will examine semantic word representations in another

Such a mid-estuary dissolved phosphate maximum was also observed by van Beusekom and de Jonge (1998). In general, nutrient gradients in the Ems estuary during summer are