Educational inequality in the Netherlands and Belgium
A comparison of the relationship between socioeconomic status and reading literacy
Name: Ruth van de Pol
Course: Research Master Child Development and Education, Thesis 1
Supervisors: Mw. J. Lenkeit Dr (first supervisor), mw. prof. dr. M.L.L. Volman (second supervisor) Studentnumber: 10004512
Date: 30-07-2014
Abstract
This study takes a closer look at the relationship between socioeconomic status (SES) and academic achievement relationship in the Netherlands, and the Flemish and French part of Belgium. Five hypotheses are tested, which are derived from Willms’ analytical framework (2006) and the context of the Dutch and Belgium educational systems, which have freedom of education as a central principle. The hypotheses go into the individual SES-academic achievement relationship, the role of the school SES mean in this relationship and the distinction between public and private schools. To test these hypotheses, data from the PISA-cycle of 2009 is used. It has been found that the SES-academic achievement relationship is especially strong in the French part of Belgium. In the Netherlands, this explained the least part of the variance. No differences were found between schools in the public and private sector. However, the three educational systems show different patterns that might be interesting to investigate in further research.
2 Introduction
The meritocratic ideal is widely held in Western society: People should be judged by their merits, thus on what they have accomplished. In this ideal, success is reached through ambition and ability (Young, 1994). However, an individual’s accomplishments are not independent from his or her family background. When studying family background, most researchers refer to socio-economic status (SES), which captures an individual’s or family’s ranking in a social hierarchy, based on the access to or control over a combination of valued commodities, like wealth, power, and social status (Mueller & Parcel, 1981). Research has shown a positive relationship between SES and academic achievement (Sirin, 2005). This relationship already shows at a young age. By the time children enter kindergarten, children with a low SES significantly lag behind in rudimentary skills (Lee & Burkam, 2002). This gap increases over time (Caro, McDonald, & Willms, 2009).
A relationship between SES and academic achievement can be found in many countries. For example, in the Progress in International Reading Literacy Study (2011) of students in grade four (aged nine or ten years old), a positive relationship between SES and reading literacy was found in all 63 participating countries (International Association for the Evaluation of Educational Achievement, 2012). The same holds for all 65 participating countries of the Programme for International Student Assessment (PISA) of 15-year old students by the Organisation for Economic Co-operation and Development (OECD, 2012c). However, the strength of the SES-academic achievement relationship differs per country. The most extreme countries with regard to the amount of variance explained by SES are the Slovak Republic and Iceland. Respectively 25% and 8% of the total variance in academic achievement within these countries can be explained by the SES of the student (OECD, 2012c). Although cross-national studies might give us some clues about why the SES-academic achievement relation differs per country (for example the share of private schools in a country, Ammermüller, 2005), more detailed studies are needed to interpret these differences. This study tries to do so by comparing the SES-academic achievement relationships in two countries, which offers the opportunity to take a detailed look at the similarities and differences and take the national contexts into account.
3 In this study, a comparison is made between the Netherlands and Belgium. These countries have a lot in common: A partly shared history, language, and similarities in their educational system. Freedom of education is a key principle in both systems. Public funding is not only granted to schools that are governed by the state, but also to schools that are governed by private parties. Therefore, schools can have a relative autonomous position (Mentink & Vermeulen, 2011, De Rynck, 2005). Despite these similarities, the relationship between SES and academic achievement is quite different in both countries. In Belgium, this relationship is stronger than in the Netherlands (OECD, 2012c).
The question this study aims to answer is: What are the similarities and differences between the Netherlands and Belgium with respect to the relationship between socio-economic status and academic achievement? In order to answer this question, an analytical framework proposed by Willms (2006) is used. This framework examines the way individual SES is related to academic achievement. It also includes the compositional effect of SES on academic achievement. Next to that, two country specific hypotheses are formulated about the difference between public and private schools.
Theoretical framework
Family background: Socioeconomic status
Socioeconomic status is often used as a family background variable in studies about the relationship between family background and academic outcomes. It tends to be stable across a child’s life (Gottfried et al., 2003). The measurement of SES usually includes parental income, parental education and parental occupation. These indicators are supposed to reflect the social and economic resources that are available to a child (Sirin, 2005). Next to that, indicators of cultural capital are sometimes included as well. Cultural capital refers to cultural competence and familiarity with cultural practices (Bourdieu, 1986). It is strongly associated with other indicators of SES, but is also able to explain a unique part of the variance in educational achievement (Sullivan, 2002).
Students with lower SES tend to have lower academic achievement, as shown in a meta-analysis by Sirin (2005). In the early years of schooling, the learning gains between children with a lower and higher SES do not significantly differ. Instead, studies have showed a seasonal pattern of learning: During the summer break, differences between children with a lower and higher SES tend to
4 increase again (Alexander, Entwisle, and Olson, 2007). However, if students with a low SES attend a school with an lower school mean SES, they are even more likely to have lower achievements. This is also known as the compositional effect (Willms, 2006). The relationship between SES and academic achievement does not vary noticeably between different subjects (OECD, 2010a). However, research also shows that within schools, there are differences between the SES-science/mathematics on the one hand and the SES-reading/writing on the other hand. Thus, in one school, the relationship between SES and mathematics might be strong, while the relationship between SES and reading weak, or vice versa (Ma, 2010).
In most studies, the relationship between SES and academic achievement is studied as a linear relationship. However, research that focuses on the mechanisms underlying the SES-academic achievement relationship, shows that several moderators and mediators play a role. Examples are the cognitive ability of children, family processes and the family investment in offering a stimulating learning environment (Ganzach, 2000, Yeung, Linver, & Brooks-Gunn, 2002). Also without including other variables, the SES-academic achievement relationship might not be a linear one. Willms (2003) found that the SES-academic achievement relationship varies among different levels of SES: The relationship converges at higher levels of SES. Thus, above a certain threshold, the effect of an increase in SES becomes less. However, other studies have not found this result or only for certain countries (Caro & Lenkeit, 2012, Willms, 2006, Willms & Somers, 2001).
The case of the Netherlands and Belgium: SES-academic achievement relationship
With regard to the two countries that are studied in this paper, the relationship between SES and academic achievement seems to be stronger in Belgium than in the Netherlands. This showed, for example in the PISA-assessment of 2009, in which reading literacy of 15-year old students was tested. An increase of one standard deviation on SES, raises the predicted reading literacy with 47 points in Belgium, and 37 point in the Netherlands (SD=100, OECD, 2010a). However, the most recent PISA-study, in which mathematic skills were assessed, shows a different picture. A one-unit increase in SES leads to an increase in predicted mathematic performance of respectively 40 and 37 points in the Netherlands and Belgium. However, in Belgium, the SES-mathematic performance relationship
5 explains a larger part of the variance: In the Netherlands, 11.5% of the variance in mathematics is explained by SES. In Belgium, this is 15% (OECD, 2012c). The two different PISA-assessment show that it matters which measurement of academic achievement is used.
Within the PISA-study, a possible explanation of the differences in the SES-academic achievement relationship is the Gini-coefficient. There is a weak association between the strength of the SES-academic achievement relationship and the wealth distribution in a country (measured by the Gini-coefficient1, OECD, 2010a). However, the wealth distribution in the Netherlands and Belgium is comparable: They have Gini-coefficients of respectively .29 and .26 (OECD, 2010b). Next to that, in countries where the average education level (in years of education) is higher, the relationship between SES and educational achievement decreases (Willms, 2003, Thomas, Wang, & Fang, 2001). With regard to the average years of education of citizens above the age of 25, the difference between the Netherlands (11.6 years) and Belgium (10.9 years) are also limited (United Nations Development Programme, 2012). In the age group of 25-64, 72% (the Netherlands) and 71% (Belgium) have followed at least upper secondary education (OECD, 2013a). Also, the money invested in education is similar: Per student enrolled in secondary education, the Netherland government invests 11838 dollars, while Belgium spends 11004 dollars (OECD, 2013a). On a more broad level, the Human Development Index, which captures measurements regarding life expectancy, education and income, also shows that the Netherlands and Belgium do not differ that much. On a range from 0 to 1, the Netherlands scores .92, and Belgium .90 (United Nations Development Programme, 2012).
The role of schools: Segregation
Since the Netherlands and Belgium are quite comparable on characteristics at country level, the differences in the relationship between SES and academic achievement might find its origin in factors at school level. At school level, segregation between schools is important in terms of the relation between SES and academic achievement. Segregation occurs because students are not randomly assigned to schools. Students with a low SES tend to go to the same schools, as well as students with
1 The Gini-coefficient can take values between 0 and 1, where 0 indicates a totally equal distribution of wealth (everybody owns the same amount of wealth) and 1 a totally unequal distribution of wealth (one person owns all the wealth in the country).
6 a high SES. Therefore, students with a low SES are more likely to end up in schools with a low mean SES and vice versa. This increases inequality, because students with a low SES tend to do worse in schools with a low SES mean (Montt, 2011, Willms, 2006). Thus, the mean SES of a school is positively related with its students’ academic achievement, after taking into account students’ individual SES (Willms, 2006). In Belgium, the compositional effect explains a larger part of the variation than in the Netherlands (OECD, 2009, OECD, 2012c).
The interpretations of the compositional effect is a topic of discussion. Some researchers use the school SES mean as a proxy for peer effects. Peer effect are related to, for example, engagement in school life and thus academic achievement (Lee, Bryk, & Smith, 1993, Zimmer & Toma, 2000). Other researchers point at the relationship between mean SES and contextual factors, such as financial resources and quality of instruction (Harker & Tymms, 2004). Willms (2006) suggests that both peer effects and contextual factors contribute to the compositional effect. However, there are also researchers who do not agree that the compositional effect has any value. For example, Nash (2010) suggests that the found effects of mean SES are caused mostly by the selection processes of schools. This is especially interesting for the Netherlands and Belgium, since students are selected for different school tracks, based on their ability in both school systems (tracking, Eurydice, 2013, Steenssens et al., 2008).2 Cross-national studies indeed show that tracking at a young age increases inequality in education (Ammermüller, 2005, Hanushek & Wößmann, 2006, Horn, 2009, Marks, Cresswell, & Ainley, 2006, Van de Werfhorst & Mijs, 2010). However, Dumay and Dupriez (2008) show that it is not likely that the compositional effect is caused by the selection processes of schools (which are based on ability), since they have included cognitive skills in their analysis of the French part of Belgium and still found a compositional effect based on SES.
The role of schools: Freedom of education
Next to tracking, freedom of education could also contribute to segregation and thus to educational inequality (Harker, & Tymms, 2004, Hirtt, 2013). Freedom of education is at the core of educational
2 NB: In Flanders and the Netherlands, students are tracked at the age of twelve. In the French part of Belgium, this selection officially takes place at the age of fourteen, but in practice students can attend separate schools from twelve years onwards (Eurydice, 2013, Steenssens et al., 2008).
7 policy in the Netherlands and Belgium. Due to freedom of education, the Netherlands and Belgium both have a high ratio of private schools (De Rynck, 2005, Mentink & Vermeulen, 2011). Private schools are managed by non-governmental organization, while public schools are directly or indirectly managed by a governmental organization. Most cross-national research shows that a high ratio of private schools increases inequality in educational outcomes (Ammermüller, 2005). Orfield and Frankenberg (2013) specify this claim: Private schools, driven by market theory, increase inequality. Schütz, West and Wöβmann (2007) support this claim, by concluding that public funding for private schools decreases the relationship between SES and school achievement. Montt (2011) suggests that the SES-academic achievement relationship might be less strong in private education, based on the idea that private schools have more opportunities to anticipate on the needs of their students, since they are more autonomous. However, the influence of school autonomy on the SES-academic achievement relationship is not straightforward: Only certain aspects of school autonomy contribute to equality, like influence on staffing decisions (Schütz, West, & Wöβmann, 2007).
In the Netherlands, 68% of the schools are private, and in the Flemish and French part of Belgium respectively 75% and 59% (Hirtt, 2014). There are small differences in the implementation of freedom of education in the Netherlands and Belgium. For example, in Belgium, the financing of private schools is a bit lower than that of public schools, while the funding is equal in the Netherlands (Opdenakker & Van Damme, 2007, Mentink & Vermeulen, 2011). In the Netherlands and Flemish part of Belgium, schools have more autonomy than in the French part of Belgium (De Rynck, 2005, Schütz, West, & Wöβmann, 2007). In the Netherlands, private schools are mostly protestant and catholic schools. The difference between SES of students at these schools is very small. Only small differences were found in the outcomes of the schools, which disappeared when school composition was added to the model (Driessen & Van der Slik, 2001). In Belgium, private schools are mostly catholic schools, who recruit students with a SES above average. Differences between the private and public sector are mediated by the school composition (Opdenakker & Van Damme, 2007).
8 Based on this background, five hypotheses are formulated that will be tested for the Netherlands, the Flemish part of Belgium and the French part of Belgium. The hypotheses refer to the individual SES-academic achievement relationship (hypothesis one and two), differences between public and private education (hypothesis three and four) and the compositional effect (hypothesis five).
H1: There is a positive relation between a students’ SES and his or her academic achievement. H2: The relation between SES and academic achievement diminishes at higher levels of SES.
H3 (a): There is a relation between a schools’ legal status (private/public) and the academic achievement of its students.
H3 (b): The schools’ mean SES mediates a relationship between private/public schools and academic achievement.
H4: The relationship between SES and academic achievement is similar in private and public schools. H5: There is a compositional effect of schools’ mean SES, meaning it is positively related to academic achievement of students, after taking into account the student’s individual SES.
Data
For the analysis, data of a PISA-assessment, which is designed and implemented by the Organisation of Economic Co-operation and Development (OECD), is used. The PISA-dataset of 2009 contains first of all information of individual students. Next to that, school principals also filled out questionnaires, providing for example information about school leadership. Students were selected by clustered sampling. Thus, in the first phase, the participating schools were randomly selected. After that, the participating students were randomly selected from the student population in these schools. In the Netherlands, 4760 students from 186 different schools participated. In total, 8501 Belgium students from 278 schools participated. Belgium has a higher sample size, equally distributed over the Flemish part and the German/French part. This makes it possible to also analyse the data about the Flemish on the one hand and the German/French part on the other hand separately. In the Flemish part, 4596 students from 159 schools participated. Unfortunately, with regard to the German/French part of Belgium, the dataset does not indicate whether schools are in the French or German part.
9 Therefore, a less ideal procedure is used. Only the participants who filled out their questionnaire in French, were included in the sample. Therefore, a relative small part of the respondents in the French part of Belgium, 757 students (8,9%), was excluded from the analysis. After this, 3148 French-speaking students from 114 schools were included from the French/German part,.
Method
Since students are clustered in schools, multilevel modelling is used to test the hypotheses. Students in the same school share this context, which is likely to cause dependency among the observations. This dependency is expressed in intra-class correlation, which can be seen as the proportion of variance that exists at the school level (Kreft & De Leeuw, 1998). In multilevel modelling, the linear models for each school are fitted first, and these models are linked together by a second-level model (Kreft & De Leeuw, 1998). The different models are run separately for each educational system.
Operationalization
Academic achievement. There are different domains of academic achievement. In this paper, it is operationalized as reading literacy. Reading literacy is highly correlated with the other academic domains, namely science (r=0.84) and maths (r=0.87) (OECD, 2012). Also, there is a high correlation between socioeconomic inequality in different academic subjects (reading, writing, mathematics and science), ranging from .54 to .82 (Ma, 2000). However, SES is stronger related to reading literacy than to mathematics. Therefore, this is chosen as a dependent variable for this study.
Reading literacy is defined as follows in PISA: “Reading literacy is understanding, using, reflecting on and engaging with written texts, in order to achieve one’s goals, to develop one’s knowledge and potential, and to participate in society” (OECD, 2009, p. 23). In order to measure these competencies, PISA designed a broad range of tasks. The scores for reading literacy were scaled to have a mean of 500, and a standard deviation of 100, across all participating countries. The mean score of the Dutch students was 517 (SD=86). In the Flemish part of Belgium, students had a mean score of 522 (SD=93). The mean score of the French students in Belgium was 490 (SD=111). The reliability of the reading tasks was very high (0.93 in the Netherlands and 0.94 in Belgium).
10 PISA converted the outcome of the different reading tasks in five plausible values per student. These plausible values are comparable to scores with an error marge included. One of these values is used for this analysis, to which will be referred as the language literacy score.
Socioeconomic background. Socioeconomic background is summarised in an index that is calculated by taking into account parents’ education, parents’ occupation, and an array of household possessions. Parents’ education was measured by asking students about their parents’ highest educational qualification and then recoded in an estimation the accompanying years of schooling. Students were also questioned on their parents’ occupation by means of open-ended questions. The responses were coded and mapped to the international socio-economic index of occupational status (OECD, 2009).
Home possessions are measured by asking students if their family owns certain assets. The possessions are summed up in composite variables that indicate wealth/educational resources and cultural resources. This distinction is based on theoretical considerations, but the outcome of factor analysis also shows that the two variables can best be separated. The selection of the included items is based on a principal component analysis: All variables with a factor loading above .30 are included. For the indicator of wealth/educational resources, the following dichotomous variables are included: Possession desk to study at, own room, a quiet place to study, internet, dishwasher, DVD-player, educational software and technical reference books (0=no, 1=yes). Also, students were asked how many cellular phones, computers and cars they have (0=none, 1=one, 2=two, 3=three or more). The reliability of this scale is .58. For the scale cultural possessions, the following dichotomous variables are included: possession of classical literature, poetry and art (0=no, 1=yes). Also, the number of books in the house are included (0=0-10, 1=11-25, 2=26-100, 3=101-200, 4=201-500, 6=more than 500). The reliability of the cultural possessions scale is .69.
Approximately 30% of the respondent have a missing value on one of the variables included in the SES-index. These missing values are not at random (Little’s MCAR test: χ2=1417, DF=1, p<.001). Especially students that score relatively low on reading literacy have missing values. The missing values are replaced with a predicted value based on regression analysis (Tabachnick & Fidell, 2001). As in the PISA report, the variables that are included in the SES-index are used for this
11 regression analysis (OECD, 2012). The SES-scale is standardized and thus has an average of zero and a standard deviation of 1. In table 1, the mean scores, standard deviations and Cronbachs’ alpha before and after imputation can be found. The mean scores and standard deviations in the different educational system do not differ very much. Cronbachs’ alpha is satisfactory in all educational systems. Because variables from the SES-construct were used to predict missing values, Cronbachs’ alpha is a bit inflated after imputation of the missing values.
Table 1
Mean scores, standard deviation and Cronbachs’ alpha’s of SES
Country Mean score SD Cronbachs’ alpha
before imputation
Cronbachs’ alpha after imputation
The Netherlands .02 .92 .70 .72
Flemish part of Belgium -.01 1.02 .77 .79
French part of Belgium -.001 1.1 .77 .79
Legal status of the school. Principal of schools indicated the legal status of their school. Public schools, which are governed by a governmental organization, are scored 0. Private schools, which are governed by a non-governmental organization are scored 1.
Segregation. Segregation based on SES is studied by looking at the mean SES-status in a school. In the Netherlands, the mean SES status of schools is -.01, with a standard deviation of .38. In the Flemish and French part of Belgium, the means are respectively .001 and .02, and the standard deviations .42 and .51.
Results Null model
12 The null model has a random intercept and does not include any predictor variables. A null model is fitted for two reasons. First of all, it serves as a baseline for the following models. Next to that, it provides an estimate for the intra-class correlation in reading literacy. The intra-class correlation can be seen as the proportion of variance that exists at the school level. The formula of the null-model is:
Yij = β0j+rij (1)
β0j = γ00 + u0j (2)
In equation (1), Yij is the reading performance of student i in class j and rij is the within-group error
term (level 1). Parameter β0j is the expected reading score for a student attending class j with an SES
average to the school mean SES. β0j includes the overall intercept (γ00) plus the between-group error
(u0j , level 2).
For the Netherlands, the null model has an overall intercept of 515 (SD=5.1), which is close to the overall intercept in Flanders: 513 (SD=5.9). The French part of Belgium has a lower intercept, namely 480 (SD=8.3). The variance estimates between and within schools (respectively level 2 and 1), together with the intra-class correlation, are given in table 2. To illustrate the intra-class correlation: In the Netherlands, 62% of the variance exists on the school level. In the French part and especially the Flemish part of Belgium, there is less variance on the school level.
Table 2
Variance and intra-class correlation based on the null model
The Netherlands Flemish part of Belgium French part of Belgium
Level 2 variance 4698 4968 7305
Level 1 variance 2817 4115 5517
13 Hypothesis one: There is a positive relation between SES and academic achievement.
This hypothesis is tested by the following model:
Yij = β0j +β1jSESij +rij (3)
β0j = γ00 + u0j (4)
β1j = γ10 (5)
Compared to equation (1), parameter β1j is added. A one-unit increase in SES leads to an increase in
reading performance of γ10and captures the inequality in academic achievement. For hypothesis one,
it is expected that γ10 has a significantly positive value.
As can be seen in table 3, there is a positive relation between SES and academic achievement in all three educational systems. SES has the biggest effect on the predicted scores of students living in the French part of Belgium. A one-unit increase in SES leads to an increase on reading literacy of almost 19 points. In the Netherlands and the Flemish part of Belgium, this is respectively about 8.5 and 14 points.
Table 3
Relationship family SES and academic achievement in the three educational systems
Parameters The Netherlands Flemish part of Belgium French part of Belgium
Intercept (γ00) 515** (SD=4.9) 515** (5.3) 482** (7.2)
SES (γ10) 8.5** (1.0) 14.2** (1.1) 18.9** (1.5)
** indicates a p-value lower than .001.
This model, with only SES to predict reading literacy, explains 6% of the overall variance on reading literacy in the Netherlands. For the Flemish and French part of Belgium, respectively 12% and 16% of the overall variance is explained. The improvement of the prediction based on this model compared to the null model is significant in all educational systems (the Netherlands: χ2=73, df=1, p<.0001, the Flemish part of Belgium: χ2=154, df=1, p<.0001, the French part of Belgium: χ2=216, df=1, p<.0001).
14 Hypothesis two: The relation between SES and academic achievement diminishes at higher levels of SES.
This hypothesis is tested by adding the squared SES term to the equation (3). Therefore, the tested model is:
Yij = β0j + β1jSESij + β2jSES²ij+ rij (6)
β0j = γ00 + u0j (7)
β1j = γ10 (8)
β2j = γ20 (9)
Compared to equation (3), only the parameter β2j is new. It consists of the grand mean slope γ20,
indicating the increase or decrease of the predicted reading scores, that goes with a one-unit increase of SES squared. A negative value of γ20 indicates that the relationship between SES and reading
literacy decreases at higher levels of the SES. A positive value, however, indicates that the relationship becomes stronger. For this hypothesis, it is expected that γ20 has a significant negative
value.
As can be seen in table 3, Belgium and the Netherlands show different patterns. In Belgium, the effect of SES on the predicted reading literacy performance increases at higher levels of SES. In the Netherlands, the effect decreases. However, the effect are very small and not significant. Therefore, this hypothesis is rejected.
Table 3
Relationship family SES, squared SES and academic achievement in the three educational systems The Netherlands Flemish part of Belgium French part of Belgium
Intercept (γ00) 515** (SD=4.9) 514** (5.4) 481** (7.3)
15 Squared SES (γ20) -.09 (.66, p=.89) .40 (.76, p=.60) .26 (.95, p=.78)
16
Hypothesis three: There is a relation between a schools’ legal status (public/private) and the academic achievement of its students.
This hypothesis is tested by adding a dichotomous variable for the legal status of the school to the equation (public=0, private=1):
Yij = β0j + β1jSESij+ rij (10)
β0j = γ00 + γ01Priv. j + u0j (11)
β1j = γ10 (12)
This equation is the same as equation (3), except for the fact that the parameter β0j now includes the
overall intercept, the effect of the dummy variable legal status of schools and the between-group error. A positive value ofγ01 indicates that attending a private school has a positive effect on the predicted
reading performance. For this hypothesis, it is expected that γ01 has a significant value.
The results in table 4 show that this hypothesis must be rejected. Adding the legal status of the school to the model does not significantly contribute to the prediction of reading literacy in any of the school systems. What stands out, is that the results of the Netherlands and the Flemish part of Belgium look quite similar: Attending a private school has a negative influence on the predicted reading literacy. The results of the French part of Belgium show the opposite result: Attending a private school, leads to a predicted reading literacy score that is about 21 points higher. The standard deviation of the effect of the legal status is quite high in all educational systems (varying from 10.9 to 14.8), indicating that there are big differences between the different private schools.
Table 4
Relationship private/public school and academic achievement in the three educational systems The Netherlands Flemish part of Belgium French part of Belgium
Intercept (γ00) 522** (SD=8.5) 523** (10) 469** (11.7)
17 Private school (γ01) -10.4 (10.9, p=.32) -12.5 (12.0, p=.29) 21 (14.8, p=.17)
** indicates a p-value lower than .001.
Hypothesis four: The relationship between SES and academic achievement is similar in private and public schools.
In order to test this hypothesis, an interaction effect between SES and the legal status of the school is added.
Yij = β0j +β1jSESij + rij (13)
β0j = γ00 + γ01Priv. j + u0j (14)
β1j = γ10+ γ11Priv.j + u2j (15)
Compared to equation (10), only parameter γ11 is added to equation (13). γ11 is the effect of being in a
private school on the slope of SES-reading performance (β1). A positive value of γ11 indicates that the
effect of SES is bigger for students in a private school, while a negative value indicates the opposite (a lower effect of SES for students in private schools). For this hypothesis, it is expected that γ11 is not
significant.
The results show that relationship between SES and academic achievement is not significantly different in private and public schools. As table 5 shows, adding this interaction-effect has the biggest impact on the prediction of reading performance of students in the Flemish part of Belgium. While the interaction-effect is absolutely not significant in the Netherlands (p=.64) and the French part of Belgium (p=.73), the p-value of the Flemish part (p=.13) comes closer to the significance level.
18 Table 5
Relationship SES, legal status, their interaction effect on academic achievement in the three educational systems
The Netherlands Flemish part of Belgium French part of Belgium
Intercept (γ00) 522** (SD=8.5) 523** (10.0) 468** (11.7)
SES (γ10) 7.9** (3.7) 11.2** (2.2) 18.1** (2.6)
Private school (γ01) -10.5 (10.4, p=.31) -12.1 (11.8, p=.31) 20.7 (14.8, p=.16)
Public/private*SES (γ11) 1.0 (2.1, p=.64) 3.9 (2.6, p=.13) 1.1 (3.2, p=.73)
* indicates a p-value between .05 and .001, ** indicates a p-value lower than .001.
Hypothesis five: There is a compositional effect of schools mean SES, meaning it is positively related to a students’ academic achievement of students, after taking into account the student’s individual SES.
For this hypothesis, the following model is tested:
Yij = β0j +β1jSESij + rij (19)
β0j = γ00 + γ01Priv. j + γ02MeanSES. j + u0j (20)
β1j = γ10 (21)
Compared to equation (10), only β0j is changed. Equation (20) includes the overall intercept, the effect
of the dummy variable public/private school (γ01), the group mean (γ02) and the between-group error
term u0j. γ02 represents the increase in the predicted reading score, that is associated with a one-unit
increase in the school mean SES. For this hypothesis, it is expected that γ02 has a significant positive
value.
As the results in table 6 show, the effect of mean SES of the school is significant in all three educational systems. In the Netherlands, a one-unit increase of the mean SES of the school, results for
19 an average student in a literacy score that is 68 points higher. For the Flemish and French part of Belgium, the increase that accompanies a one-unit increase in mean SES of the school, is respectively 104 and 101 points. The influence of a students’ SES decreases only a little bit, compared with the model that only included SES as a predictor (namely from 8.5 in the Netherlands, 14.2 in the Flemish part and 18.9 in the French part of Belgium). Thus, the compositional effect has a unique contribution to the explanation of variance. It is also interesting to see that the non-significant effect of attending a private school (see table 4) decreases especially in Flanders. This indicates that the differences in mean SES between private and public schools are relatively large in Flanders.
Table 6
Relationship family SES, mean SES and academic achievement in the three educational systems The Netherlands Flemish part of Belgium French part of Belgium
Intercept (γ00) 522** (SD=7.8) 518** (7.7) 4785** (8.4)
SES (γ10) 8.4** (1.0) 14.0** (1.1) 18.0** (1.5)
Private school (γ01) -8.7 (9.6, p=.37) -.33 (9.1, p=.97) 13 (10.6, p=.23)
Mean SES (γ02) 68** (11.9) 104** (10.4) 100** (10.1)
** indicates a p-value lower than .001.
The improvement of the prediction is significant for all educational systems (the Netherlands: χ2 =104, df=2, p<.0001, the Flemish part of Belgium: χ2
=232, df=2, p<.0001, the French part of Belgium: χ2
=217, df=2, p<.0001). It explains a considerable part of the variance on reading literacy. For the Netherlands, it explains 15% of the overall variance and 23% of the variance of group means. For the Flemish part of Belgium, it explains 31% of the overall variance and 53% of the variance of group means. For the French part of Belgium, 38% of the total variance is explained, and 62% of the variance of the group means.
21 Conclusion
The relationship between socioeconomic status (SES) and academic achievement is often reflected in a single measurement. This study tried to offer a more comprehensive understanding of this relationship in the Netherlands and Belgium (Flemish and French part), as well as pointing out the similarities and differences. As could be expected, based on earlier studies, there is a relationship between SES and academic achievement in all three educational systems. This relationship is the smallest in the Netherlands. Separating the French and Flemish part of Belgium showed that the SES-academic achievement relationship is stronger in the French part than in Flanders. This is interesting, since the French part of Belgium selects students, at least officially, at a later age for different educational tracks. However, with regard to the compositional effect, the French and Flemish educational systems have comparable results. The final model, including individual SES and school mean SES as predictors, explained quite a large part of the variance, especially in Belgium. The explained variance in Belgium is more than twice as large as in the Netherlands.
This study did not aim at finding out why the SES-academic achievement relationship is weaker in the Netherlands than in Belgium, especially the French part of Belgium. Since this is a relevant question, future research might focus on this. In certain respects, the educational systems of the Netherlands and Belgium are very comparable. This already excludes some possible causes of the differences. However, other policy differences might have an impact. For example, in the Netherlands financing of the schools depends, amongst others, on the background of students, and this is since 2002 also the case in in Flanders (Ledoux & Veen, 2009, Ooghe, 2011). The French part of Belgium does not have such a system (Demeuse, Derobertmasure, & Fraint, 2010). However, these differences are not reflected in the compositional effect, which is bigger in Flanders than in the French part of Belgium. Literature shows that the effect of the school mean can be interpreted in different ways, one of which is the contextual factors. However, these contextual factor do not only have to do with money, but also, for example, with the ability of schools to attract good teachers. Peer effects are also named as a possible interpretation. In the light of this comparison, this would mean that peer effects
22 somehow have a different influence in the different educational systems. Therefore, it is not likely that peer effects on its own explain the compositional effect.
Next to a SES-academic achievement relationship and a compositional effect, which are already studied thoroughly in existing literature, the predictors that were tested in this study did not significantly attribute to the prediction of reading literacy. There were no diminishing returns in the Netherlands and in Belgium. The results of both parts of Belgium even showed non-significant “increasing returns”, a stronger relationship between SES and predicted reading performance at a higher level of SES. As for the hypotheses formulated on the freedom of education, the results showed it does not matter if a student attends a public or private schools for the prediction of reading performance. This is not consistent with the study of Driessen and Van der Slik (2001), who have found small differences between the academic achievement at public and private schools, taking into account SES (Driessen & Van der Slik, 2001). There are several possible explanations for these differences. For example, Driessen and Van der Slik (2001) studied primary school children and used different measurements of SES and academic achievement. Also, in this study, no other background variables than SES were used. Based on the results found in this study, Ammermüllers’ (2005) claim that a large share of private schools increase inequality in a school system, doesn’t seem applicable for the Netherlands and Belgium. An explanation for this could be that the distinction between private schools, driven by market theory, and private schools in an educational system that is built around freedom of education, is valid (Frankenberg, 2013). Also, the SES-academic achievement relationship doesn’t look different in public and private schools, as suggested by Montt (2011).
An important limitation of this study, is the validity of the SES-construct. This was composed based on the questionnaires that were filled out by the students. However, students’ reports only partially correspond to parents’ reports (Schulz, 2005). Next to that, there are some limitations to the handling of the data. There were quite a lot of missing values on variables that were included in SES. These were imputed, based on linear regression techniques. Although this is in correspondence with the literature, this doesn’t add to the construct-validity of the SES measurement. With regard to the analysis, a plausible reading score was used as dependent variable. This plausible score is drawn from
23 a distribution of scores that are plausible for an individual by PISA. It therefore includes random error variance. In the ideal situation, the analysis would have been undertaken with five different plausible scores and the results would have been combined. Another limitation, that directly refers to suggestions for future research, is the usage of the legal status of the schools. The results suggested that there are big differences between the private schools. This indicates that a distinction between public and private might be too rough. There were some interesting differences in the patterns in the different educational systems with regard to the legal status of the schools. Future research could investigate these differences of the effect of freedom of education more detailed, by taking into account different types of private education. It might, for example, differ whether a private school is based upon religious or pedagogical principles. There might even be differences between schools based on different Christian denominations. Taking a closer look at the relationship between SES and academic achievement in the Netherlands and Belgium, shows the complexity of this research field and also brings up a lot of questions for further research. However, it is beyond question that it studying inequality in different educational systems is relevant and can be very valuable for policymakers.
24 Literature
Alexander, K.L., D.R. Entwisle, & L.S. Olson. 2007. Lasting consequences of the summer learning gap. American Sociological Review 72(2), 167–80.
Ammermüller, A. 2005. Educational opportunities and the role of institutions. ZEW Discussion Paper no. 7, Mannheim.
Bourdieu, P., 1986. ‘The forms of capital’ in G. Richardson (ed.), Handbook of Theory and Research for the Sociology of Education.
Caro, D., McDonald, J, & Willms, D. 2009. Socio-economic Status and Academic Achievement Trajectories from Childhood to Adolescence. Canadian Journal of Education, 32(2), 558-590.
Caro, D., and Lenkeit, J. 2012. An analytical approach to study educational inequalities: 10 hypothesis tests in PIRLS 2006, International Journal of Research and Method in Education, 35(1), 3-30.
Demeuse, M., Derobertmasure, A., & Friant, N. 2010. Differentiated financing of schools in French-speaking Belgium: prspectives for regulating a school quasi-market. Educational Research and Evaluation 16 (2), 197-216.
De Rynck, S. 2005. Regional autonomy and education policy in Belgium, Regional & Federal Studies, 15(4), 485-500.
Dumay, X., & Dupriez, V. 2008. Does the school composition effect matter? Evidence from Belgium data, British Journal of Educational Studies, 56(4), 440-477.
Driessen, G., & Van der Slik. F. (2001). Denomination, and education in the Netherlands: Cognitive and noncognitive outcomes after an era of secularization. Journal for the scientific study of religion 40(4), 561-572.
Eurydice, 2013. The structure of the European education systems 2013/2014, retrieved at 16 February 2014 from http://eacea.ec.europa.eu/education/eurydice/facts_and_figures_en.php#diagrams Ganzach, Y. 2000. Parents’ education, cognitive ability, educational expectations and educational
25 Gottfried, A.W., Gottfried, A.E., Bathurst, K., Guerin, D.W., & Parramore, M.M. 2003.
‘Socioeconomic Status in Children’s Development and Family Environment: Infancy Through Adolescence’, in M.C. Bornstein, and R.H. Bradley (Eds), Socioeconomic status, parenting, and child development. Mahwah, NJ: Lawrence Erlbaum, 189-209.
Hanushek, E., & Wößmann, L. 2006. Does Educational Tracking Affect Performance and Inequality? Differences-in-Differences Evidence across Countries. The Economic Journal, 116(510), 63-76.
Harker, R., & Tymms, P. 2004. The effects of student composition on school outcomes, school effectiveness and school improvement, An international journal of research, policy and practice, 15(2), 177-199.
Hirtt, N. 2014. Waarom zijn België en Frankrijk de kampioenen van sociale ongelijkheid in het onderwijs? PISA 2012 ontsluierd: De naakte cijfers. Studiedienst Oproep voor een democratische school, verkregen op 18 mei 2014 van www.democratischeschool.org.
Horn, D. 2009. Age of selection counts: a cross-national analysis of educational institutions. Educational Research and Evaluation: An International Journal on Theory and Practice, 15(40), 343-366.
Lee, V., Bryk. A., & Smith, J. 1993. The organization of effective secondary schools, Review of research in education, 19, 171-267.
Ledoux, G., & Veen, A.M. 2009. Beleidsdoorlichting onderwijsachterstandenbeleid. Periode 2002-2008. Amsterdam: SCO-Kohnstamm Instituut (rapport nr. 826, projectnummer 40352). Marks, G., Cresswell, J, & Ainley, J., 2006. Explaining Socioeconomic Inequalities in Student
Achievement: The role of home and school factors, Educational Research and Evaluation 12(2), 105-128.
Mentink, D., & Vermeulen, B.P. 2011. Artikel 23 grondwet. De basis van het Nederlandse onderwijsrecht. Den Haag: Sdu Uitgevers.
Montt, G., 2011. Cross-national differences in educational achievement inequality. Sociology of Education, 84(1), 49-68.
26 Mueller, C., & Parcel., T. 1981. Measures of socioeconomic status: Alternatives and
recommendations. Child Development, 52(1), 13-30.
Ooghe, E. 2011. De ‘Gelijke onderwijskansen (GOK)’-financiering: Werkt ze? Voor wie? En wat kunnen we er uit leren? Retrieved at 28 June 2014 at www.econ.kuleuven.be/onderzoek.htm. Organisation for Economic Co-operation and Development 2010a. PISA 2009 Results: Overcoming
social background – equity in learning opportunities and outcomes (Volume II), PISA, OECD Publishing.
Organisation for Economic Co-operation and Development. 2010b. StatExtracts: Income Distribution and Poverty. Verkregen op 8-6-2014 van http://stats.oecd.org/index.aspx?queryid=46189. Organisation for Economic Co-operation and Development. 2012. PISA 2009 Technical Report,
PISA, OECD Publishing.
Organisation for Economic Co-operation and Development. 2013a. Education at a Glance 2013: OECD Indicators, OECD Publishing.
Organisation for Economic Co-operation and Development. 2013b. PISA 2012 Results in Focus: What 15-year-olds know and what they can do with what they know, PISA, OECD Publishing.
Organisation for Economic Co-operation and Development. 2013c. PISA 2012 Results: Excellence Through Equity: Giving every student the chance to succeed (Volume II), PISA, OECD Publishing.
Organisation for Economic Co-operation and Development. 2013d. Synergies for better learning, retrieved at 16 February 2014 from http://www.oecd.org/edu/school/synergies-for-better-learning.htm#1.
Schulz, W. 2005. Measuring the socio-economic background of students and its effect on achievement in PISA 2000 and PISA 2003. Paper prepared for the Annual Meeting of the American Educational Research Association in San Fransisco, 7-11 April 2005.
Schütz, G., West, M., & Wöβmann, L. 2007. School Accountability, choice, and the equity of student achievement: International evidence from PISA 2003. OECD Education working papers series 9.
27 Sirin, S.R. 2005. Socioeconomic status and academic achievement: A meta-analytic review of
research. Review of Educational Research, 75(3), 417–53.
Tabachnick, B., & Fidell, L. 2001. Using multivariate statistics. Fourth edition. Allyn and Bacon. Thomas, V., Wang, Y., & Fan, X. 2001. Measuring education inequality: Gini coefficients of
education. Policy Research Working Paper 2525 (World Bank Institute).
United Nations Development Programme. 2011. Mean years of schooling (of adults) (years). Retrieved from at 21-06-2014 https://data.undp.org/dataset/Mean-years-of-schooling-of-adults-years-/m67k-vi5c at 8-6-2014.
United Nations Development Programme. 2012. Table 2: Human Development Index trends. Retrieved at 21-06-2014 from https://data.undp.org/dataset/Table-2-Human-Development-Index-trends/efc4-gjvq.
Van de Werfhorst, H, & Mijs, J., 2010. Achievement Inequality and the Institutional Structure of Educational Systems: A Comparative Perspective. Annual Review of Sociology, 407-428. Willms, J.D., & Somer, M.A. 2010. Family, Classroom, and School Effects on Childrens Educational
Outcomes. Latin America, School Effectiveness and School Improvement, An International Journal of Research, Policy and Practice, 12(4), 409-445.
Willms, J.D. 2003. Literacy Proficiency of youth: evidence of converging socioeconomic gradients. International Journal of Educational Research, 39, 247-252.
Willms, J.D. 2004. School composition and Contextual Effects on Student Outcomes, Teacher College Record, 112(4), 1008-1037.
Willms, J.D. 2006. Learning Divides: Ten Policy Questions about the Performance and Equity of Schools and Schooling Systems. UIS Working Paper No. 5
Yeung, W., Linver, M. & Brooks-Gunn, J. 2002. How Money Matters for Young Children’s Development: Parental Investment and Family Processes. Child Development, 73(6), 1861-1879.
Young, M. 1994. Meritocracy revisited. Society, 31(6), 87-89.
Zimmer, R., & Toma, E. 2000. Peer effects in private and public schools across countries. Journal of Policy Analysis and Management, 19(1), 75-92.
29 Appendix 1: Factor loadings
Composite variable: Home possessions
For the selection of variables for this composite variable, principal component analysis was used. Fifteen variables were included, namely the possession of a desk to study at, a room of your own, a quiet place to study, educational software, a link to the internet, books to help with you school work, technical reference books, a dictionary, a dishwasher, a DVD player, the number of cellular phones, televisions, computers, cars and rooms with a bath or shower. The Kaiser-Meyer-Olkin Measure of Sampling Adequacy was .70, which is the proportion of variance that might be caused by underlying factors. This value indicates that factor analysis is appropriate. Bartlett’s test of sphericity was not significant (χ ²=14358, DF=105, p<.0001). All variables that had a factor loading above .40 were selected. In table one, the selected variables and their factor loadings are shown. Cronbach’s alpha for this variable is quite low, namely .50.
Table 1
Factor loadings for the composite variable home posesssions
Component Factor loading Answering categories
Possession internet .43 Yes, no
Possession technical reference books .44 Yes, no
Possession dishwasher .47 Yes, no
How many computers .61 None, one, two, three or more
30 Composite variable: Cultural possessions
Four items were included in this principal component analysis. They are all included in the composite for cultural possessions. The factor loadings are shown in table 2. The Kaiser-Meyer-Olkin Measure of Sampling Adequacy was .72, which is the proportion of variance that might be caused by underlying factors. This value indicates that factor analysis is appropriate. Bartlett’s test of sphericity was not significant (χ ²=8320, DF=6, p<.0001). All variables that had a factor loading above .40 were selected. In table one, the selected variables and their factor loadings are shown. Cronbach’s alpha for this composite variable is satisfactory, namely .69.
Table 2
Factor loadings for the composite variable cultural possessions
Component Factor loading Answering categories
Possession classical literature .79 Yes, no
Possession poetry .76 Yes, no
Possession art .57 Yes, no
How many books in the house .75 0-10, 11-25, 26-100, 101-200, 201-500,
more than 500
Composite variable: SES
The SES composite includes home possessions, cultural possessions, and the highest schooling and occupational status of both parents. The factor loadings are shown in table 2. The Kaiser-Meyer-Olkin Measure of Sampling Adequacy is .76, which is the proportion of variance that might be caused by underlying factors. This value indicates that factor analysis is appropriate. Bartlett’s test of sphericity was not significant (χ ²=17105, DF=15, p<.0001). The eigenvalue of this variable is 2.7. Cronbachs’ alpha for this composite variable is satisfactory, varying from .70 to .77 in the different educational systems. The overall Cronbachs’ alpha is .72.
31 Table 3
Factor loadings of the composite SES
Component Factor loading
Home possessions .41
Cultural possessions .65
Mothers highest schooling .72
Fathers’ highest schooling .73
Mother occupational status .71
32 Faculty of Social and Behavioural Sciences
Graduate School of Childhood Development and Education
T
HE RELATIONSHIP BETWEEN REFERRAL TO SPECIAL NEEDS
EDUCATION AND GRADE RETENTION
The influence of individual characteristics, school composition and structural
school characteristics on referral to special needs education in the Netherlands
Research Master Child Development and Education
Thesis 2
Ruth van de Pol
Supervisor (UvA): Mw. Dr. T.T.D. Peetsma
Supervisor (Inspectorate of Education): Dhr. Dr. B.A.N.M. Vreeburg
Date: 03-07-2015
33 Abstract
The purpose of this study was to determine if there is a relationship between referral to special needs education and the amount of grade repeaters in a school. Between 2010 and 2012, 0.77% of the total population of primary school pupils were referred to special primary education, which is meant for pupils with mild learning or behavioral difficulties, and 0.22% of the pupils were referred to special education for pupils with a handicap, chronic disease or disorder. The data that are used in this study consist of information about 2.5 million school year transitions between 2010 and 2012. Several control variables on individual and school level are included in a multinomial logistic regression analysis that predicts the chance of referral to special (primary) education. Pupils with a low SES are 3.3 times more likely to be referred to special primary education and pupils who have already repeated a grade 3.7 times more likely. The sex of a pupil has a particular strong effect on referral to special education: Boys are 2.8 times more likely to be referred. Characteristics on school level only have a modest influence on the chance of referral. Pupils in schools with few grade repeaters (0-5%) are less likely to be referred than pupils in schools with 5-10% grade repeaters. Pupils in schools with more than 30% grade repeaters (about 1% of the schools), are also less likely to be referred (compared to schools with 5-10% grade repeaters). Grade repetition (on individual and school level) has a larger influence on the chance of referral to special primary education than on the chance of referral to special education. The fact that the effect of grade repetition on individual level is quite large, might indicate that schools do not prefer either grade repetition or referral, but often apply a combination of the two measures.
34
I
NTRODUCTIONClassroom teaching is the most important form of instruction in Dutch primary education. To deliver classical instruction effectively, pupils are supposed to develop in a similar pace. Therefore, a school should take action when a pupil is lagging behind. Examples of possible actions are applying forms of differentiation within the classroom, performing an intervention or extending the learning time (see Driessen, Leest, Mulder, Paas & Verrijt, 2014). In practice, schools also quite regularly choose to take pupils out of the class, in different ways. Grade retention is a common measure in the Netherlands: About 18% of the pupils has repeated at least one grade by the time they leave primary school (Inspectie van het Onderwijs, 2012). Schools can also choose for a more severe measure, namely to refer a pupil to special needs education. Every year, approximately 1% of the total student population in primary education is referred to a form of special needs education (Smeets, 2007).
Referral to special needs education and grade retention both indicate that a school is unable to provide adequate educational support to a pupil within his or her current classroom. Also, both measures are most often taken in the third and fourth grade of primary school, when pupils are approximately between the age of 6 and 8 (Smeets, 2007, Driessen et al., 2014). Next to that, it is striking that boys, pupils with a non-western background and pupils with a low socioeconomic status have to repeat a grade more often and are more often referred to special needs education (Smeets, 2007, Driessen et al., 2014). Schools make different decisions about comparable pupils in terms of background characteristics and cognitive and behavioral problems: Pupils are referred at one school, while comparable pupils are retained at another school (Jepma, 2003, Peetsma, Vergeer, Roeleveld & Karsten, 2001). There are massive and unexplained differences between schools when looking at the share of pupils that have to repeat a grade (Driessen et al., 2014). Therefore, the question rises if referral to special needs education and grade retention are, to some extent, interchangeable on school level. In other words, are some schools more inclined to refer pupils, while other schools choose for other options, like grade retention? This has been suggested by Driessen et al. (2014): Differences in the use of referral of pupils might explain differences between schools in grade retention.
35 The relation between referral at a school and grade retention has only been discussed in studies incidentally. Schools that have a lot of kindergarten extension refer their pupils more often (Jepma, 2003). Overmaat, Ledoux and Koopman (1997) also found that schools without grade repeaters are less likely to refer their pupils. However, this only concerns a small percentage of the schools that take an extreme position concerning grade retention. Thus, not much is known about the relation between referral to special needs education and grade retention. This study attempts to fill this knowledge gap for several reasons. Studying this relation might provide insight in differences in the way primary schools deal with students who, for some reason, cannot keep up in the classroom. Next to that, grade retention and referral to special needs education receive a lot of attention from educational policy makers, because both measures are considered undesirable by the Dutch government. Scientific studies have cast strong doubts on the effectiveness of grade retention and the financial costs of grade retention are considerable (Goos, Belfi, De Fraine, Van Damme, Onghena & Petry, 2013, Inspectie van het Onderwijs, 2013, CPB, 2015). Also, the government has tried to limit the referral to special needs education with the law on inclusive education. If extra support is needed, this is offered as much as possible in the pupils' environment, at their own school and in their own class. Better coordination between care from different agencies and reducing the costs of care are two important aims for this policy (Messing & Bouma, 2011). A comparison of matched pupils shows that pupils who stay in regular education make more cognitive progress than their counterparts who were sent to special needs education (Peetsma et al., 2001). Thus, knowledge about referral to special needs education and referral could also have practical implications for policy makers.
S
PECIAL NEEDS EDUCATIONThe Dutch system distinguishes two types of special needs education. First of all, special primary education (speciaal basisonderwijs or SBO) is a form of special needs education in which pupils are taught the same curriculum as in regular primary education. Children with learning difficulties, behavioral difficulties and pupils who need extra attention and/or support can go to a special primary
36 education school. These schools teach in smaller groups and there are more staff members who are experts on special educational needs. After special primary education, pupils go to vocational school (vmbo), practical education (praktijkonderwijs) or secondary special education. The second form of special needs education is named special education (speciaal onderwijs or SO) and is meant for pupils with a handicap, chronic disease or disorder. Special education is divided in four clusters. Children with visual disability go to cluster one; Cluster two is meant for children with hearing disabilities; Children with a physical disability, mental disability or a chronic illness go to cluster three; Finally, cluster four is meant for children with psychiatric disorders or severe behavioral problems. Most pupils from special education continue their schooling at special secondary education (Inspecite van het Onderwijs, 2012). Every year, about 0.6% of the total population in primary education is referred to special primary education and 0.2% to special education (Smeets, 2007).
In the last decades, several policy changes have been made with regard to special needs education in the Netherlands. After a rapid growth of the number of pupils in special education in the 1970s and 1980s, especially in the southern, catholic part of the Netherlands, the policy in the 1990s was aimed at reducing the amount of pupils in special education. The ‘Weer Samen Naar School’ policy (Back to school together) introduced ‘Samenwerkingsverbanden’, partnership between regular and special schools. In 2002, extra financing for special educational needs in regular education (LGF) was introduced, in an attempt to further decrease the referral to special (primary) education. This rapidly changing situation implicates that less recent research has to be interpreted carefully (Smeets & Rispen, 2008).
G
RADE RETENTIONIn the Netherlands, there are no national guidelines about grade retention (Eurydice, 2011). Schools are free to determine their own policy on grade retention, which they have to describe in their school guide. Consequently, the differences between schools in terms of how many pupils have to repeat a grade is quite large (Driessen et al., 2014). Some schools have extremely low or high percentages of
37 repeaters: In 6% of the schools, none of the pupils had to repeat a grade in 2011. However, in other schools, 25% to 50% had to repeat a grade (Inspectie van het Onderwijs, 2012).
P
REDICTORS ON INDIVIDUAL LEVELBoth grade retention and referral to special education are often exerted in the third and fourth grade (Smeets, 2007, Driessen et al., 2014). It is interesting that the differences between the cognitive problems of pupils who repeat a grade versus those who are referred to special (primary) education are very limited (Paas, Mulder & Roeleveld, 2014). Van der Veen, Smeets and Derriks (2010) find that referral of pupils is more strongly associated with pupils’ cognitive attainment levels than with behavioral problems. The latter hardly contributes to the prediction of referral (except when it concerns autism). However, other authors find that next to pupils with an intellectual disability and autism, pupils with externalizing problems are also referred more often (Ledoux, Van der Veen, Smeets, & Roeleveld, 2008, Paas et al., 2014). In addition to learning difficulties and behavioral problems, there are some pupil characteristics that influence the chance of being referred and of having to repeat a grade.
Socioeconomic status (SES) refers to an individual’s or family’s ranking in a social hierarchy, based on the access to or control over a combination of valued commodities, like wealth, power, and social status (Mueller & Parcel, 1981). Students with a low SES have a higher chance of referral to special (primary) education and grade retention (Maas & Meijnen, 1999, Driessen et al., 2014). SES is related to several aspects of child development, which can be important for both referral and grade retention. Children with a low SES experience relatively more health problems with more severe consequences; they relatively often lag behind in terms of cognitive and academic attainment and also show more symptoms of psychiatric disturbances and maladaptive social functioning (Bradley & Corwyn, 2002). There are indications that while taking into account the diagnosis of a pupil, lower SES still increases the chance of referral to special needs education. For example, pupils with intellectual disability have a higher chance of going to special needs education if they have a low SES (Szumski & Karwowski, 2012).
38 Boys have a bigger chance of referral and grade retention than girls (Driessen et al., 2014, Smeets, 2007). This can be seen in the broader context of differences between boys and girls in primary education. In general, primary school teachers evaluate the social emotional development of boys lower than the social emotional development of girls. This can be explained by several biological factors: The brains of boys develop more slowly and cause more aggressive and moveable behavior. There are also possible explanations from a cultural and social perspective: Parents are more involved with the education of daughters than of sons, a masculine ideal leads to a mismatch between boys and school culture and the school culture is too feminine (Driessen & Van Langen, 2011).
Students with a non-Dutch background also have a higher chance of referral to special (primary) education and grade retention (Driessen et al., 2014). They also relatively often lag behind in terms of learning performances. Most of his gap can be explained by a low socioeconomic status and educational attainment of parents. Next to that, these pupils are often behind in the Dutch language. In special (primary) education, pupils with a non-Western background are especially overrepresented in cluster 3 (physical disability, mental disability or a chronic illness) and cluster 4 (psychiatric disorders or severe behavioral problems). There are differences between different ethnicities. For example, some parents with a Turkish or Moroccan background do not want their children to be referred to special education (cluster 3 and 4). This leads to relatively many Turkish and Moroccan pupils with behavioral problems in special primary education (Smeets, Elfering & Hovius, 2009).
S
CHOOL COMPOSITIONThere are big differences between schools in referral to special needs education, and in grade retention. However, it is difficult to explain these differences. Some school characteristics correlate with a high percentage of grade repetition and/or referral, but the literature is mixed and only a small part of the variation between schools can be explained by the models used in literature so far.
The composition of the school or class explains part of the differences between schools. Driessen et al. (2014) found that schools with a high percentage of pupils with lower educated or
non-39 western parents have a higher rate of grade repetition. Also, relatively more pupils are referred to special needs education at these schools (Smeets, 2007, Ledoux, Smeets & Van der Veen, 2005). However, Reezigt, Swanborn and Vreeburg (2013) did not find an effect of school composition on grade repetition using multilevel analysis. Studies that include class composition show that more boys or pupils with disruptive behavior increase the chance of being referred (Maas & Meijnen, 1999). Also, when there are relatively more pupils with special needs in the class, the chance of referral decreases (Ledoux et al., 2008).
S
TRUCTURAL CHARACTERISTICS OF THE SCHOOLThere are also structural characteristics that correlate with the share of referral or grade retention at a school. First of all, school size might explain differences between schools. Grade repetition is less common at bigger schools (Driessen et al., 2014), although Reezigt et al. (2013) did not find this effect. At bigger schools, referral to special needs education occurs less often according to Smeets (2004, 2007, multilevel analysis), but more often according to Jepma (2003, matching technique).
Secondly, the denomination of a school might have an impact on the amount of grade repetition and referral. At public schools, pupils are referred and have to repeat a grade quite often (Smeets, 2004, 2007, Reezigt et al., 2013). Protestant primary schools have a relatively low percentage of grade repeaters, but a high percentage of referred pupils (Reezigt et al., 2013, Smeets, 2007). Pupils at Catholic schools are referred relatively often (Jepma, 2003). Driessen et al. (2014) distinguish between reformist schools, which have few grade repeaters, and Islamic schools, which have a lot of grade repeaters.
Jepma (2003) hypothesized that traditional schools refer pupils more often, because they put more emphasis on the simultaneous development of pupils. However, he did not find this effect. Driessen et al. (2014) also indicate that grade repetition could be influenced by school pedagogy: Non-traditional schools have less grade repeaters (e.g. Montessori schools), although anthroposophist schools have a lot of grade repeaters.
