The impact of change in teacher-to-pupil ratios on school performance

(1)

Master Thesis Public Administration

The impact of change in teacher-to-pupil

ratios on school performance

Measuring the effects of the size of the change in the teacher-pupil ratio on the average school results

Track: Economics & Governance Name: Matthijs Malkus

Student number: 2245302

Thesis Coördinator: Dr. P.E.A. van den Bekerom Second reader: Dr. A. Poama

(2)

2

Abstract

In this paper, we study whether there is a significant effect of both the changes in teacher-to-pupil ratio as the size of the standard deviation of the teacher-to-pupil ratio on the average primary school results of the final examination. In addition, the moderating effect of the supporting staff-to-pupil ratio on this effect are studied. Using multilevel analysis methods, we examined the effects on both the average CITO score and the school advice of the schools. We find that the teacher-to-pupil ratio is significantly associated with an increase in the average final examination scores of schools, but only through a fine balance with supporting staff. On contrary, the size or changes of teacher-to-pupil ratio are not significantly associated with changes in average school advices of schools. Our results suggest that putting effort in balancing the teacher-to-pupil ratio over time is challenging and generates few, if any, benefits for the performance for the school and its pupils.

(3)

3

Chapter 1: Introduction

The Dutch education system ranks among the top systems in the world, with the Netherlands being the third most educated nation worldwide (Wittenburg, 2016). Part of this success is due to the quality of primary schools, which prepare children between four and twelve years old for their future education. Primary schools are continuously striving to improve their performance as first step of the educational process in the Netherlands. An important aspect of the goal to improve education in the Netherlands is knowing the impact of smaller classes in primary education on the learning outcomes of the children (Visser, 2016). To provide the pupils with the best basis for the future school careers, it is relevant to know the exact impact of ‘shocks’ in the ratio between teachers and pupils on the performance of the schools. It can prove beneficial for a school’s performance to mitigate these shocks. With a better ability to handle these dynamic external effects, the school will become more resilient and will most likely have a more stable performance. With the ability to have a stable performance under dynamic circumstances, the schools can improve their educational process, which in turn improves their prospects on continuity.

Impact of dynamism

The environment of an organization is comprised of all factors outside the organization that can affect the organization or parts of it. Examples of these factors are stakeholders, funding, levels of competition and legislation. These external factors are inherently dynamic and lead to the environment of an organization being unstable (Van den Bekerom, 2016: 45). The stability of an organization is challenged by this dynamism, which in turn will negatively affect the outcomes or performance of the organizations (Van den Bekerom et al, 2016). The external influences and changes (dynamism) that result from this unstable environment can affect the performance of primary schools and other public organizations. A study performed in school districts in Texas showed that student performance and dropout rates are negatively influenced by budget shocks or higher student enrollment (Van den Bekerom, 2016:46). In addition, the claim that dynamism negatively impacts performance is supported by various other empirical studies (Boyne, 2009; Meyer & O’Toole, 2009; Van den Bekerom et al., 2016).

(5)

5

School performance

In this study we will mainly focus on the impact of number of pupils and the staff size on school performance, while including the possible moderating effects of supporting staff. The shifting of the number of new pupils also changes the ratio of teachers to pupils in a school, which increases or decreases the time of personal attention a teacher can give to each pupil. In turn, this decrease in personal attention could affect the school results of the pupils. This would suggest that bigger changes in the teacher to pupil ratio would negatively affect the school’s performance. In Van den Bekerom’s 2016 dissertation, she studies how the percentual change of elementary school pupils affect the organizational performance and finds that this form of turbulence indeed negatively affects the school’s performance. This study will instead focus on the impact of changes in the teacher-to-pupil ratio as a form of dynamism and if and how they negatively affect the school’s performance. Using a teacher-to-pupil ratio is expected to yield a better estimate for the effects, since the change in total pupils in a school does not incorporate the possible mitigating effects of the teachers. If the number of teachers rises accordingly to the increase of number of pupils, the effect could be limited since the personal attention a pupil receives, remains the same. In addition, the possible moderating effects of the presence of supporting staff on the school results will be studied, since they can take over some of the duties performed by teachers and assist in supervising the pupils if the availability of the teaching staff does not suffice. This would give the teachers more time to focus of their primary task, namely teaching. The magnitude of the possible effects of the supporting staff are unclear, since there seems to be no previous scientific literature on this subject.

The focus of the study is whether keeping a balance in teacher-to-pupil ratio could affect the performance of the pupils, as a proxy for school performance. It could show the importance of being resilient to this dynamism and avert any negative effects to maintain the stability of the organization and keep performing optimally. The outcomes could be instrumental in forming strategies and policies to keep schools resilient from this dynamism and guarantee their continuity for the years to come.

(6)

6

Study set-up

In this research I will examine the impact of changes of the average teacher-to-pupil ratio of schools in 2014 - 2018 on the school results of primary school in the Netherlands. To capture the effects of shifting pupil ratio, we will look at the size of the changes in the teacher-to-pupil ratio over these four school years, since those are the years where DUO made their school performance data publicly available. The research question of this paper is “What is the impact of

the size of the changes in the teacher-to-pupil ratio on the average primary school results?”. This

research will focus on approximately three thousand elementary schools within the Netherlands. It will be a prospective, quantitative study focused on the estimation of the effects of variation in teacher-to-pupil ratio on the final examination results. The outcomes would contribute to the knowledge of the impacts of keeping a stable teacher-to-pupil ratio and mediating impact of supporting staff on school performance.

Internationally, various studies were done concerning the direct impact of teacher-to-pupil ratio on school performance (Dabo, 2015; Duflo, 2015). However, there is limited research done in this field using the change of the teacher-to-pupil ratios over time in the Netherlands. Applying this concept in a longitudinal quantitative research within the Netherlands is new. From a societal perspective, knowing if the change in teacher-to-pupil ratios can impact the school results and what the impact of supporting staff can be, could be useful. It could stimulate primary school managers/boards to act accordingly in the case of sudden shocks of new applications of pupils or unexpected departure of teacher staff. Therefore, the results could contribute in reshaping the optimal pedagogical and economic conditions for the education programs in the Netherlands.

The Dutch education system

The education system of the Netherlands starts at the elementary level, with children being allowed to attend at 4 years and compulsory attendance at 5 years. This compulsory learning trajectory (in Dutch: ‘leerplicht’) lasts until their 16 years old. After this the qualification-requirement holds, pupils must attend school between the age of 16 and 18 or until they graduate with a diploma (Nuffic, 2018). A schoolyear lasts from August in until July in the next year, with slight differences among regions. An academic year lasts from September first until August 31st in the next year. The Dutch ministry of Education, Culture and wet op primair onderwijs Science is responsible for

(7)

7

the educational system of the Netherlands, and must abide the educational legislation, such as the primary education law (Nuffic, 2018).

Primary and Secondary Education

In the Netherlands the primary school consists of eight consecutive schoolyears. At 4/5 years, children enter at class one ‘groep 1’ and continue this until class eight ‘groep 8’. Normally children attend the primary schools until they are 12 years of age. After this a combination of school advice and the results of a final examination will determine their following school career (Nuffic, 2018). The school advice is key in this process. The secondary schools can be subdivided into two categories. General education (vmbo-tl, havo and vwo) and preparatory secondary vocational education (vmbo-bb/vmbo-kb/vmbo-gl). Vwo is the pre-university education any grands those who graduate entry to a bachelor’s degree research university. Havo is the senior general secondary education and grands graduates entry into bachelor’s degrees of universities of applied science (hbo). Finally, a vmbo diploma gives right to a senior secondary vocational education (mbo), which if completed can grand entry into both bachelor’s and associate degrees of a university of applied science (Nuffic, 2018).

Reading Guide

In the next chapter, theoretical framework, we will give a summary of the current scientific literature on the effects of dynamism on public organizations, the role of managers in this process and the effects of teacher-to-pupil ratio on school performance. In the following chapter, research

design, the scientific design and the statistical methods of the study will be explained as well as

the operationalization of the variables. Subsequently, in the results chapter the outcomes will be discussed. Finally, the conclusion will focus on the interpretation of the results and give policy recommendations to incorporate the results into meaningful policies.

(8)

8

Chapter 2: Theoretical framework

Organizational environment

The environment of the schools will be the primary point of interest in this study. The environment of an organization can be defined as all aspects outside of an organization that could possibly affect the whole or parts of the organization (Daft, 2010). Organizations are affected by the environment by the provision and withholding of resources (Dess & Beard, 1984). The environment of an organization can be subdivided in an institutional environment and the task environment. The institutional environment includes all governmental regulations and policies, while the task environment comprises elements that focus on the role of markets, competition and resources (Oliver, 1997; Van den Bekerom, 2016: 45). According to Dess & Beard (1984), the organizational environment can be subdivided into three dimensions; munificence, complexity and dynamism. This has become a leading conceptualization in studies focusing on organizational environments (Van den Bekerom & Meier, 2016). Munificence is the level of availability of resources to (public) organizations. Organizations tend to search for environments that allow stability and organizational growth (Dess & Beard, 1984; Van den Bekerom & Meier, 2016). With a higher munificence, the organizations function better. The second dimension, complexity, is defined as the level of similarity between the elements of the population where the organization interacts with. In a more complex environment, the range of activities will be more diverse, and the organization will perceive more uncertainty (Dess & Beard, 1984). Finally, the dimension dynamism is focused on the first two dimensions. It indicates the level of change in both munificence and complexity (Dess & Beard, 1984; Van den Bekerom & Meier, 2016). With a higher dynamism the unpredictability of the change heightens, and the organizational environment will become more unstable. Scharfman (1991) made use of a similar tri-fold dimension concept, using slightly different terminology. To further define the concept of dynamism on public organizations and the models associated with the concept we will use the “environmental factors” used in the “logic of governance model” of Lynn, Heinrich and Hill (2000). They identify eight different forms of environmental factors: Political structures, level of monitoring, state of the economy, competition level, dependencies, target population, legal institutions and technological dynamism (Lynn, 2000). Van den Bekerom & Meier (2018) use similar subdimensions, according to the PESTEL framework. These subdimensions are Political, Economic, Social, Technological, Environmental and Legal. The political dimension focusses on the role of government, economics

(9)

9

on the changes in growth rates, social on the population and demographical changes, technical on innovations, environmental on the issues around environmental preservation and pollution and finally legal focusses on legislative issues (Van den Bekerom & Meier, 2018). The three dimensions of Dess & Beard munificence, complexity and dynamism perceive different aspects of the environment, making it complementary to the PESTEL framework. In Van den Bekerom & Meier’s article, they combine the PESTEL framework with Dess and Beards first two dimensions, giving each of the political, economic, social, technological, environmental and legal dimensions both a munificence and complexity dimension (2016). Although all environmental dimensions can affect the performance of primary schools, the social complexity and economic munificence dimensions seem to be the most applicable to the case. The social complexity is relevant because the performance of schools is dependent on the human capital and the diversity in social backgrounds and educational capacity of both the pupils and the staff. The economics munificence plays an important role as well, since the levels of funding can critically affect the level of educational performance a school can produce. The level of funding can determine if better teaching methods can be bought, teachers can be hired, and school buildings can be maintained. Human capital, characteristics of pupils and the financial situation of schools is thought to be the most relevant of the environmental factors influencing a school’s performance.

Turbulence, dynamism & environmental change

In this study we will primarily focus on the dynamism/stability dimension. There is no real consensus on the correct terminology for the dynamism/stability dimension. By some scholars these external influences are referred to as dynamism or turbulence (Boyne, 2009; Van den Bekerom et al, 2016). However, (environmental) dynamism and turbulence are often used interchangeably in the scientific literature or are considered partial elements of each other (Dess & Beard, 1984; Mascarenhas, 1984; Scharfman, 1991; Volberda, 1994). In the scientific literature, there is slight discrepancy between the definitions of turbulence and dynamism used by scholars. ‘Turbulence’ is defined by Fisher (2012) as the negative property of an organization’s environment. In contrast, ‘dynamism’ is defined by Dess and Beard in their 1984 article as a difficult to predict change that elevates the uncertainty for important organizational members. Although the definitions show a remarkable resemblance, Dess and Beard’s definition emphasizes the change as part of the uncertainty, while the definition used by Fisher is more widely defined and incorporates more steady negative elements as well, instead of only focusing on changes. To

(10)

10

be consistent, this paper will only use the term dynamism. We will use the definition of Dess and Beard (1984) since this incorporates a smaller scope and only includes the forms of dynamism we are focusing on, namely the sudden changes. Dynamism can take on many shapes, a hurricane or a global treat can be a form of dynamism, but on a smaller scale dynamism can be caused by the closure of a factory or the shifting of certain preferences and behavior within a population (Boyne, 2009). In this study the new and existing pupils in the school can be considered the target population. The competitive level for primary schools is mainly generated by the neighboring schools and their performances that might attract potential pupils and the dependency of the primary schools on the governmental institutions for their financing. If the factors change, this can negatively affect the performance of the organization (Lynn, 2000). New regulation by the government, more competition in the area and shocks in the number of new pupils will require the schools to adapt and change, to be able to keep performing at the same level and acquire the necessary financial means to ensure their continuity.

Performance of public organizations

The effects of dynamism on public organizations, such as schools, has been thoroughly studied by several scholars (Boyne, 2009; Perrott, 2012; Piening, 2013; O’Toole, 2014; Bekerom et al., 2016). The environment of an organization is composed of multiple dynamic elements that can influence, and be influenced by, public organizations. For example, technological, social, economic circumstances can affect the way an organization is managed. The effects of dynamism can negatively influence the performance of an organization (Boyne, 2009; O’Toole, 2014). With a higher dynamism, the organizational environment will be more complex. Subsequently, organizations in more complex environments are associated with lower performance (O’Toole, 2014). In their 2011 book, O’Toole & Meier introduce their model for predicting organizational output. In its most simple form, the model predicts the organizational outcome (performance) based on the performance of the last measure moment, discounted for the rate of stability and the shocks to the system (O’Toole & Meier, 2011:23). Since shocks will affect the stability of an organization, it will affect its performance. For public organizations, keeping the organizational stability and performance up can be perceived to be the best way to measure their resilience to dynamism (Boyne, 2009).

(11)

11

The manager’s role

To heighten the resilience to environmental dynamism, the role of managers in public organizations becomes relevant. Managers are generally required to mitigate the negative effects of the environment to keep the organization running stable and continuously. As leading figures in the organizations, they set the course and the strategies. By correctly anticipating and responding to dynamism they ought to minimize the adverse effects on the performance of their organizations. Different scholars have shown that the negative effects of this dynamism can indeed be (partially) mitigated by the right managerial actions (Meier, 2003; Boyne, 2009; Van den Bekerom et al, 2016). O’Tool & Meier acknowledge the positive effects of managers as well and incorporate managerial efforts in their more advances model (2011: 131).

In her 2016 study, Van den Bekerom used the percentual change of number of pupils of primary schools as a dynamic factor that could influence the school performance, which was operationalized as average school results of the pupils. Her results show that within schools internally oriented networking activities, such as team involvement, neutralize the negative effects of changes in number of pupils on the average school results of the pupils. Meier et al. (2003) found in their longitudinal study of 500 U.S. school districts that effective management of dynamism (here depicted as networking with other organizations and populations) frees educational institutions from their routines and makes them use resources more effectively. This is merely a management tool, so not necessarily the standard managerial tasks that are performed within the educational sector, but it shows how capable management can improve performance outcomes within schools.

Primary schools and the teacher-pupil ratio

In the past years, a lot of research has been done on the effects of both class sizes and teacher-to-pupil ratios on the learning performances of students (Barro & Lee, 1996; Lewit & Baker, 1997; Hattie, 2005). There is a slight discrepancy between the two terms, so I will merely focus on the teacher-to-pupil ratio (Lewit & Baker, 1997). According to the outcomes of most international and national literature on the subject, smaller classes are associated with higher learning outcomes (Blatchfort, 1994; Lewit & Baker, 1997; van Gorsel, 2013, Dabo, 2015; Dulfo, 2015). The difference in school performance between pupils in larger and smaller classes can possibly be attributed to the different teaching methods for larger classes compared to the smaller classes, with

(12)

12

teachers being forced to put less emphasis on smaller group experiences since those are harder to organize (Blatchfort, 1994; Lewit & Baker, 1997; Dabo, 2015). Other explanations can be that smaller classes increase the teacher-pupil interaction and promote better individual evaluations of pupils (Lewit & Baker, 1997). These more personal approaches could affect the performance of pupils and allow teacher to better evaluate performance levels of students and teach accordingly. However, the effects found in international literature cannot automatically be generalized to the Dutch primary schools, for example due to school cultural differences, differences in educational requirements for teachers or additional resources/teaching methods available.

Although there are several studies that focus on the class size in relation to the school performance, there is considerably less research onto the impact of the magnitude of changes in teacher-to-pupil ratios on the school performance of pupils. The size of these changes of teacher-to-pupil ratios within schools could affect the learning performance as well. Teachers that are used to teach a certain number of pupils might have to adapt their learning methods to the bigger or smaller classes, which could affect the quality of their teaching. In turn, the change in quality of teaching could affect school results of the pupils. As Van den Bekerom’s (2016) paper concluded, the shocks in number of pupils can negatively affect school performance. This study will aim to see if this hypothesis still holds if the model corrects for the standard deviation in the teacher-to-pupil ratio. The standard deviation indicates the length of the spectrum in which the teacher-to-pupil ratio changes over a period of 2014 to 2018. In this study the changes in the teacher-to-pupil ratio as well as the overall standard deviation of the teacher-to-pupil ratio will be the main elements of dynamism. The main hypothesis is formulated:

H0: there is no significant effect of both the changes in teacher-to-pupil ratio as the size of the

standard deviation of the teacher-to-pupil ratio on the average primary school results.

If there is no correlation between the change as well as the standard deviation in teachers relative to the number of pupils in a primary school and the average examination results this hypothesis would be accepted. However, if the H0 hypothesis is rejected, the alternative hypothesis (HA) can be accepted. The alternative hypothesis being:

HA: The changes in to-pupil ratio and/or the size of the standard deviation of the teacher-to-pupil ratio significantly decreases the average primary school rsults.

(13)

13

This would suggest that balancing the teacher-to-student ratio over time can positively affect the final examination results of primary school pupils. It would indicate that being able to anticipate on the “shocks” in teacher-pupil ratio could improve the school performance. The causal mechanism in this case would being able to the keep the ratio of teacher to pupil as constant by using management strategies to anticipate and act as soon as a teacher leaves the school.

Supporting staff

In addition, the supporting staff of a school could potentially influence this performance as well. Supporting staff are for example: (technical) teaching assistants, janitors and administrative workers (aob, 2019). The presence of a more prominent supporting staff can help reduce the workload of the teachers and create a more effective working environment (Butt, 2005). It would give the teachers more time to focus on their primary task, namely teaching and let the supporting staff take over some more general tasks. Having more time for their teaching, teachers in schools with more supporting staff could be able to perform better, which could result in better school results of the pupils. A study by Goldhaber at al. (2012) shows that letting primary school teachers specialize in particular subjects is associated with better school achievements. To incorporate the possible moderating effect of a bigger supporting staff on primary schools on school performance, we add an additional hypothesis:

H0: there is no significant effect of the changes in the supporting staff-to-pupil ratio and/or the

size of the supporting staff-to-pupil ratio on the effect of changes in teacher-to-pupil ratios on the average primary school results.

It would correct for the possible moderating effects of supporting staff on the possible effects of the first hypothesis, namely the impact of changes in teacher-to-pupil ratio on school results. The alternative hypothesis would be:

HA: The changes in the supporting pupil ratio and/or the size of the supporting staff-to-pupil ratio significantly moderate the effects of the changes of the teacher-to-staff-to-pupil ratios on the average primary school results.

The outcomes of this study could have important implications how primary school managers perceive and act to keep the stability in the balance of teachers to students. If the first hypothesis is rejected, this would send a signal that being able to keep the teacher-pupil ratio stable over the

(14)

14

years benefits the average school results and thus become a more attractive school for future pupils. In addition, the second hypothesis can indicate if the number of supporting staff is able to moderate the possible effects. It would indicate if having extra supporting staff to compensate for the ‘shocks’ in teacher-to-pupil ratio is beneficial on the school performance.

(15)

15

Chapter 3: Research Design

This study uses quantitative methods derive to test the hypotheses. It uses longitudinal data over a period of four school years to test if changes in teacher-to-pupil ratio affect school results. Each school’s results will be measured from the four different cohorts that make the final examination in this period. It uses a large-N design, since there are more than 3000 primary schools included our dataset and analysis. The data is strictly observational, with the cohorts not being affected or influenced by the researcher during the time of observation.

Operationalization

The key concepts of this research are limited to three concepts. The first is the change in the teacher-pupil ratio. This is the projection of the number of fulltime equivalents (fte) worked by teaching staff in a school divided by the number of pupils. Because this research stretches over a period of 2014 - 2018, the number of teachers fte’s and number of students are needed for every school in the sample over the full period. This data is publicly available through the Educational department of the Dutch Government (DUO, 2018). Using the yearly ratios, the standard deviation sx is calculated using cumulative squared individual ratios of a school on a point in time xi minus the mean of the ratio of the school 𝑥̅. This number is divided by the number of ratios nx and finally the square root is extracted to find the standard deviation. This is shown in equation 1. This is done for every school in the sample and generates one standard deviation per school over the period of 2014 - 2018. Subsequently, the outcomes will be incorporated in the regressions.

equation 1.

In addition, to measure performance, we use the results of the final examinations at the end of primary school. For a consistent result and to be able to better compare scores, only the primary schools that use the ‘Centraal Instituut voor Toetsontwikkeling’ (CITO) examination during the four years will be selected in the sample. This test is the most used examination among primary schools. The outcome of the CITO lies between 500 and 550 and gives an indication of the level of secondary school that might be fitting for the pupil (CITO, 2018). Although other examinations are gaining popularity as well, it was only in 2014/2015 that the freedom of examination choice

(16)

16

was instituted in the Netherlands (Verus, 2014). Therefore, other tests are relatively small in the earlier years and the CITO has the highest participation grade overall. Choosing only the sample that uses the CITO test might bias the generalizability of the outcomes to the whole sample. To test if the outcomes of the various examinations are significantly different, we use a proxy variable, namely the school advice given by the teachers. This advice is based on the perception of the teachers on the performance of a child, mostly since the sixth year (‘groep zes’) and therefore have an important discrepancy with the CITO scores, which is a standardized test and merely a momentarily performance (Schumacher, 2018). Since 2014-2015, the school advice is given before the final examination. This makes the impact of the advice weigh more heavily than before (Schumacher, 2018). We measure the share of pupils that receive an advice for ‘havo’ or higher education forms in both groups. In addition, we do not only use the school advice as a proxy, but as the dependent variable instead of CITO examination as well.

We use a two independent samples t-test to test if the means of the outcomes of school advice (in percentage attributed to each school level) significantly differ between the schools that did and did not use the CITO examination. If there is no significant difference between the school advices of the CITO compared to the other tests, we assume that the choice for examination is done randomly and the results can be generalized to the whole population. However, the school advice and the results of the chosen examination are not necessarily perfectly correlated, so this remains an assumption. Furthermore, school advise is less precise than the CITO examination so small deviations might remain unobserved. The average CITO scores per year per school are publicly available as well trough the data from the Educational department (DUO, 2018). Because the number of cases (schools) and observations (ratio per year) within the research design is large, an observational large-N design is the only correct approach for this study. More specific, because the outcome variable is an interval number (CITO score 500-550), an interval regression will be used to measure the effects of average teacher-pupil ratios on the CITO score.

The units of analysis that are being analyzed in this study are the total pupils within each primary school, they are the units of observation as well. The length of the study is only 4 schoolyears (‘14/’15, ‘15/’16, ‘16/’17, ‘17/’18), since that is the available data accessible via DUO (DUO, 2018). However, there are some limitations to the data provided that limit the methods of this study as well. A limitation in the data is the lack of data specific to teachers-to-pupils within a grade.

(17)

17

The only data that is available is the total number of teachers in a school, which is divided by the total number of pupils in a school. This only gives a mean of teacher-to-pupil ratio of that school and has no necessary implications for the class-level developments. There could be a large variance between the class sizes within a school and teacher assigned to it, but due to the data limitation there is no way of telling. However, we assume that the board within a school will strive to equalize the teacher-to-pupil ratio over all classes. Nonetheless this is a big assumption. The cases that will be selected for this study are all available (nonspecial) primary schools within the Netherlands that use the CITO examination. Schools will only be selected if the data concerning number of teachers and number of students over the period of 2014 – 2018 is complete, in addition to the average CITO scores of those years. It is possible that by eliminating schools that did not provide the needed data, the outcomes are biased upwards. However, it is difficult to check and correct for this potential bias. Another possible problem is the variety effects of smaller school with smaller classes, since they have a high teacher-to-pupil ratio but solely due to the smaller classes. This ratio could more heavily vary over the years compared to the bigger schools, due to the small number of new students and the relative impact on the teacher-to-pupil ratio. To limit these extreme effects only schools with classes that consist of at least 10 pupils will be included in the analyses. However, doing this could affect our external validity so a Shapiro-Wilk test will be done to check if the CITO results and school advise are still normally distributed.

To further optimize the explanatory power of the model we include more (available) covariates in the regression, such as: ideological foundation of the school, based upon the five biggest classes: general, general special, protestant-Christian, reformed and catholic schools, aside from a ‘other’ group that contains all other smaller ideological foundations. According to a 2013 study by Jaap Donkers, the ideological foundation of a primary school can impact the school results (rtl, 2013). The socio-economic location of the school is put in the equation as well, with schools in postal code areas with a combination of high unemployment and low average wages getting a score of one and the other schools getting a zero (DUO, 2018). Gordon and Monostiriotis (2006) studied the effects of location on the secondary school results. Their results suggest that there is a significant association between social class composition in the neighborhood of a school and the school results of the pupils, although their results are based on secondary schools. Furthermore, the number of students that must repeat a class are included, with the definition of the repeaters being that a pupil is at October first of the measured year 12 years or older and in the final class

(18)

18

(“groep acht”) whilst spending more than the required eight years in the primary school (DUO, 2018). According to Fertig (2004) there is a negative association between school results and repeating a class within German schools. The percentage of male teachers in the school is added as well, with the variable being constructed by dividing the fte’s worked by male teachers by the total fte’s worked by teaching personnel. The number of male teachers in primary schools has been declining for some years, and some authors acknowledge that missing male role models might affect the personal development of both boys and girls (Mills, 2000; De Vries, 2014). Finally, school ‘weight’ is included which is based on the educational background of the parents of the attending pupils. The pupil is given a weight of 0.3 if both the parents do not have more than a secondary vocational education ‘vmbo kadergerichte leerweg’ as highest completed education. The weight 1.2 is assigned if one of the parents only has elementary school and the other no more than a secondary vocational education ‘vmbo kadergerichte leerweg’ as highest completed education. To calculate the total weight of a school the total weights of all students are accumulated. The total number of pupils on that school times 0,06 is subtracted from the total weights number. The outcome is rounded up to an integer. If the outcome is lower than zero, the final number of that school is zero. There is an upper limit in this approach. If 0,8 times the total number of pupils is lower than the final number, the final weight attributed to the school is 0,8 times the total students, rounded up to an integer. Scientific research has shown that educational background of parents can have a significant effect on the school results or the school advice of their children (Driessen, 2005; Oomen, 2010; Bussemaker; 2016; Broekroelofs, 2018). However, the schools with a weight are financially compensated for the burden of having to educate a student with a weight (DUO, 2018).

Threats to inference

When using the current research design, one needs to be limiting the biases that can affect the outcome of the regression. The endogeneity bias is an important factor to consider in the methodological set-up of the research. Endogeneity bias consists of reverse causality bias and omitted variable bias. However, because of the use of longitudinal data, the possible inference of reverse causality is minimized in this study. The dependent variables (CITO result and School advice) are generated near the end of the period of longitudinal data gathering. It cannot be possible due to the time element for the CITO result to influence the teacher-to-pupil ratio, since those data were from an earlier point in time. In addition, the use of panel data decreases the probability of

(19)

19

omitted variable bias in the analysis, since time fixed omitted variables are automatically eliminated. However, time varying omitted variables are still possible in the analysis and are not eliminated. To eliminate this bias would require us to use an Instrumental Variable (IV) regression. However, due to time limitations of the research this will not be done is this study. This means that the possibility of omitted variable bias, although small, remains and therefore we cannot estimate causality in this study. At best an association between the average teacher-to-pupil ratio of a schoolyear and the average CITO results of that schoolyear can be estimated. To check if an earlier CITO scores affect the subsequent CITO scores in a school, a lagged variable of the previous CITO in included in the model. If this variable is significant this suggests that earlier CITO scores in a school are correlated with the later CITO scores. A similar variable is generated for the dependent variable on school advice. Finally, the threat of inference through randomness is limited by using a random sample. Although through the sample selection some cases are expelled, by not using CITO examination or having missing data in the dataset, we check if those cases are fully random as well. Using the independent samples t-tests we check if the schools that use a CITO have the same share of pupils that get a havo advice or higher as schools that use another examination method. In addition, the data will be cleaned and checked on possible outliers and missing data. Given that this assumption holds, the threat for randomness is minimalized.

(20)

20

Chapter 4: Results

Descriptive statistics

In the table 1 below the data is depicted. Of all primary schools available in the DUO dataset, the schools that provided special education for students with special needs are not included in the regressions. As mentioned in the previous chapter, we only selected the schools that used the CITO examination. In the table below, we show all the variables with their description, number of observations, mean and standard error. The variable ‘public’ is the only dummy variable, for which the percentual distribution of each class is shown. It is remarkable how between 14/15 and 17/18 the average teacher-to-pupil ratio remains relatively steady around 0.055 (1 teacher on 18.2 students) while in the same time the percentage of pupils that receive a school advice of havo or higher steadily increases by 1.72 percent point. The results of the CITO examination remain constant, with only a little dip in the 2015-2016 examination. However, the CITO examination changes every year and its difficulty can vary over the years. Furthermore, the average full-time equivalents worked by supporting staff per pupil decreases with time, with almost a 4% decrease between 2014-2015 and 2017-2018. Other remarkable elements are the decrease of school weight from 9.6 to 7.2 and the percentage of men as teachers from 15.1% to 13.8%. The decrease of school weight seems to be a positive element since it suggests that the total number of children within the sample that have a ‘weight’ assigned to them is steadily decreasing, or the individual weight of the children is lower compared to the earlier years. The decreasing percentage of men as primary school teachers is not a new phenomenon. This trend has been observed before 2014 and seems to continue (De Vries, 2014). The distribution of the denomination of schools are in favor of the non-public schools, with more than 65% having a special ideological or religious denomination and less than 35% being a public or special public school. Mainly due to missing values in the full-time equivalents worked by supporting staff members, the number of observations differ between the variables. However, the total observations per variable are never below the 3032 or higher than 3070.

(21)

21

Table 1 Descriptive statistics

Two-sample t-tests

Before the regressions are ran, the outcomes of the school advice must be compared between schools that use the CITO examination and those that do not. The two-samples t-tests are being used to see if the average share of pupils that get a school advice of havo or higher is significantly different between the two groups. To do so, first a histogram is plotted to see the distribution of the share of pupils in all four schoolyears. The histograms are depicted in graph 1.

Variable name Variable description Observations 2014-2015 2015-2016 2016-2017 2017-2018

Independent variable n-value mean

(standard error) mean (standard error) mean (standard error) mean (standard error) Total_Pupils Average total number of pupils

enrolled in the schools

n = 3070 247.1101 (138.0475) . 246.0147 (138.7199) . 244.584 (139.7867) . 243.3583 (140.7858) . FTE_Teachers Average full time equivalents

worked in the schools by teaching staff n = 3070 13.19961 (7.331868 ) . 13.05399 (7.357366) . 13.05457 (7.407581) . 12.97733 (7.606439) .

Teacherpupilratio Average teacher-to-pupil ratio in

the schools . n = 3070 .055177 (.0161777 ) . .0548344 (.020273 ) . .0553025 (.0172294) . .0550499 (.0143838) .

FTE_Supportingstaff Average full time equivalents worked in the schools by supporting staff n = 3032 1.539303 (1.841491) . 1.490968 (1.820735) . 1.40181 (1.626795) . 1.373556 (1.699077) .

Suppstaffpupilratio Average supporting staff-to-pupil ratio in the schools

n = 3032 .0065685 (.0076915) . .0063832 (.0076243) . .0061265 (.007728) . .0060387 (.0077648 ) . Totalstaffpupilratio Average staff-to-pupil ratio in the

schools n = 3032 13.26418 (7.341622) 13.11996 (7.368578) 13.11875 (7.419885) 13.03828 (7.62288) Public Dummy variable on wheater the

school is public 0 = non public 1 = public n = 3067 0 = 65.77% 1 = 34.23% . 1 = 65.77% 1 = 34.23% . 2 = 65.77% 1 = 34.23% . 3 = 65.77% 1 = 34.23% .

Statusneighborhood Average socio-economic status of

the neighbourhood of the schools

. n = 3070 .2374593 (.4255953) . .2371336 (.4253941) . .2371336 (.4253941) . .2361564 (.4247885 ) .

School_weight Average school weight n = 3070 9.600977 (26.75586 ) . 8.548208 (24.63383) . 7.757003 (22.98851) . 7.249186 (21.15617 ) .

Perc_male Average percentage male teachers n = 3069 15.13218

(9.607215) . 14.68884 (9.528978) . 14.23546 (9.346348) . 13.8064 (9.258834) .

Perc_classrep Average percentage class repeaters 2014 n= 3070 | 2016 n= 3070 2015 n = 3062| 2017 n= 3063 11.42218 (.0811938) 11.77681 (.0860755) 11.75886 (.0879754) 11.06851 (.0862429) Dependent variable

Cito_average Average cito score in schools n = 3070 535.4166 (3.979075) . 534.977 (3.912929) . 535.5707 (3.878007) . 535.5109 (3.675423) . Perc_havohigher Average percentage of pupils

getting a school advise of havo or

higher . n = 3070 (15/16 = 3069) 47.45023 (16.82242) . 48.03489 (16.7949) . 49.60048 (16.99851) . 49.17479 (16.98299) .

(22)

22

Graph 1: Histograms of the havo advise between CITO and non-CITO schools

A remarkable observation that can be made from the histograms is that the schools with other examinations is that the schools with other examinations have a higher peak in the zero percent. This peak is absent with the CITO schools and causes the distribution to be visibly higher around the middle for those schools. A possible explanation for this peak on zero can be that schools with on average lower performing pupils choose a different examination than the CITO. The two-sample t-tests confirm that there is a significant difference between the means of pupils that get a havo or higher advice between the CITO and non-CITO pupils. Havo or higher education was advised to subsequently 47.5% (2014-2015), 48.0% (2015-2016), 49.6% (2016-2017) and 49.2% (2017-2018) of the pupils that did the CITO examination. For the other pupils this percentage was subsequently: 41.4%, 41.9%, 42.6% and 42.3%. The precise outcomes are shown in appendix 1. This difference is significant in t-tests for all four schoolyears. This has important implications for the generalizability of the results, that cannot be generalized to the pupils with other examination tests. The results generated in this study can only be used for the population of pupils that uses the CITO examination. 0 .0 1 .0 2 .0 3 0 50 100 0 50 100

CITO OTHER EXAMINATION

D e n si ty HAVO+ 1415 Graphs by noncito 0 .0 1 .0 2 .0 3 0 50 100 0 50 100

D e n si ty HAVO+ 1718 Graphs by noncito

(23)

23

Preliminary tests for multilevel analyses

The statistical tests that are done are measuring the impact of the teacher-to-pupil ratio over time and the standard deviation of this teacher-to-pupil ratio on the average CITO scores of a class. In addition, the same test will be done using school advice as the dependent outcome. For each of the regressions the effect with and without the possible moderating effects of supporting staff. The appropriate test for a panel dataset with multiple separate observations within schools is a multilevel analysis. However, a random intercept model must be done to check if the nested data of the schools are clustered. First a fixed-effects multilevel regression is done, with only the dependent variable (cito_average) and time variable (time) with the school number (newid) as parameter. We then check if the variance of the residuals is significantly different from zero. The estimates and standard errors are depicted in table 2. The full outcome is added in appendix 2.

Table 2: variance of intercepts and residuals

Since the confidence intervals of the estimates does not include zero, we can safely assume that the variances are significantly different from zero. Subsequently, we run an intraclass correlation. The outcome, as shown in table 3, gives an index of the proportion of variation between the different schools. In this case, 48.8% of the variation in CITO scores is occurring between schools (level 2 units). This suggests that the bigger share of the variation is occurring within schools (level 1) which makes an argument for using multilevel modelling.

Table 3: Intraclass correlation coefficient

Random Effects Parameters Estimate Std. Err. 95% Conf. Interval Newid: Identity

var (_cons) 7.306243 .2371606 6.855895 7.786174 var (Residual) 7.653893 .112863 7.435852 7.878328

Level ICC Std. Err [95% Conf. Interval]

(24)

24

Assumptions

To properly run a multilevel analysis, there are four assumptions that must be satisfied (UvA, 2017):

• Linearity

• Homoskedasticity of the residuals

• Residuals should be normally distributed in all groups • Independent observations

The linearity assumptions states that there should be a linear correlation between the dependent variable and each independent variable. Plotting the scatterplots with the continuous variables in the regressions shows that this assumption is for filled (UvA, 2017). The scatterplots are shown in appendix 3.

The next assumption requires the variance of the residuals to be equal for level 1 groups (classes). This assumption is tested by running an IM-test. The outcome is shown in table 4. The p-value of heteroskedasticity is 0.0000 indicating that the hypothesis that there is heteroskedasticity must be rejected and homoskedasticity of the residuals can be assumed (UvA, 2017).

The normal distribution of residuals in all groups is the next assumption. To test this, a Kernel density plot is produced to compare the normal density with the density of the residuals in the model. It is shown in graph 2. The Kernel density aligns properly with the normal distribution, so we consider this assumption for filled as well.

Cameron & Trivedi’s decomposition of IM-test

Source chi2 df p

Heteroskedasticity 620.05 26 0.0000

Skewness 157.83 6 0.0000

Kurtosis 57.75 1 0.0000

Total 835.63 33 0.0000

(25)

25

Graph 2: Kernel density estimate

The final assumption is about the independence of the observations. Using a multilevel analysis, by definition, violates the assumption of independence of all observations (Hox, 2010:14). It requires that the observations on level 2 (between schools) cannot correlate. However, this is controlled for level 1 (within schools) by the choice for statistical tests. For the other independent variables, a high correlation between two variables can affect the results. To control for this, a correlation matrix is generated for each schoolyear. The results are depicted in table 6 below. The high correlation between the two dependent variables is expected and the only correlation above 0.5. There is a high correlation (0.484 – 0.457) between school weight and status of the neighborhood. Both the variables are linked to the socio-economic domain of the schools. Since most pupils are expected to live in the close neighborhoods around the school location, this explains the close link between both. The outcomes give no indication for a multicollinearity. With this final assumption checked, the results generated by a multilevel analysis can be considered trustworthy. 0 .0 5 .1 .1 5 D e n si ty -20 -10 0 10 20 Residuals

Kernel density estimate Normal density

kernel = epanechnikov, bandwidth = 0.4426

(26)

26

2014-2015 CITO_av havohigh Teach-pupil Suppst-pupil Status neighb. School weight Percent. male Total pupils Perc. Class rep Public school CITO_average 1.0000 Perc_havohigher 0.6645 1.0000 teacherpupilratio -0.1834 -0.1697 1.0000 Suppstaffpupilr. -0.2010 -0.1770 0.3540 1.0000 Statusneighborh. -0.2923 -0.2631 0.2432 0.2646 1.0000 School weight -0.3475 -0.3339 0.2992 0.2975 0.4838 1.0000 Perc_male 0.0780 -0.0158 0.0150 0.0470 0.0403 0.0103 1.0000 Total_pupils 0.1225 0.1920 -0.1959 -0.0842 0.0075 0.0899 0.0342 1.0000 Perc_classrep -0.2098 -0.2494 0.0811 0.0378 0.1175 0.1752 -0.0222 -0.0130 1.0000 Public -0.0988 0.0080 0.0847 0.0110 0.0995 0.0886 0.0289 0.0231 0.0206 1.0000

2015-2016 CITO_av havohigh Teach-pupil Suppst-pupil Status neighb. School weight Percent. male Total pupils Perc. Class rep Public school CITO_average 1.0000 Perc_havohigher 0.6671 1.0000 teacherpupilratio -0.1215 -0.1303 1.0000 Suppstaffpupilr. -0.2008 -0.1640 0.3179 1.0000 Statusneighborh. -0.2900 -0.2717 0.1807 0.2521 1.0000 School weight -0.3502 -0.3294 0.2075 0.2774 0.4693 1.0000 Perc_male 0.0426 -0.0083 0.0304 0.0402 0.0441 0.0146 1.0000 Total_pupils 0.1024 0.1966 -0.1569 -0.0804 0.0188 0.0996 0.0366 1.0000 Perc_classrep -0.2317 -0.2771 0.0399 0.0611 0.1567 0.1998 -0.0333 -0.0276 1.0000 Public -0.0637 0.0002 0.0776 0.0172 0.0998 0.0888 0.0350 0.0258 0.0285 1.0000

2016-2017 CITO_av Havohigh Teach-pupil Suppst-pupil Status neighb. School weight Percent. male Total pupils Perc. Class rep Public school CITO_average 1.0000 Perc_havohigher 0.6647 1.0000 teacherpupilratio -0.1262 -0.1443 1.0000 Suppstaffpupilr. -0.1760 -0.1602 0.4052 1.0000 Statusneighborh. -0.2824 -0.2564 0.1656 0.2094 1.0000 School weight -0.3539 -0.3337 0.2078 0.2333 0.4567 1.0000 Perc_male 0.0449 0.0038 0.0338 0.0377 0.0478 0.0142 1.0000 Total_pupils 0.1179 0.1904 -0.1958 -0.0945 0.0317 0.1127 0.0449 1.0000 Perc_classrep -0.2536 -0.2797 0.0704 0.0255 0.1548 0.1960 -0.0089 -0.0607 1.0000 Public -0.0848 0.0001 0.0854 -0.0056 0.1001 0.0918 0.0475 0.0279 0.0280 1.0000

(27)

27 2017-2018 CITO_ average Perc_ havohigh Teach-pupil Suppst-pupil Status neighb. School weight Percent. male Total pupils Perc. Class rep Public school CITO_average 1.0000 Perc_havohigher 0.6753 1.0000 teacherpupilratio -0.1131 -0.1542 1.0000 Suppstaffpupilr. -0.1385 -0.1494 0.3936 1.0000 Statusneighborh. -0.2717 -0.2451 0.1771 0.2005 1.0000 School weight -0.3359 -0.3284 0.2215 0.2268 0.4576 1.0000 Perc_male 0.0697 0.0339 0.0310 0.0242 0.0626 0.0140 1.0000 Total_pupils 0.1083 0.2014 -0.2068 -0.0942 0.0348 0.1123 0.0602 1.0000 Perc_classrep -0.2326 -0.2640 0.0648 0.0549 0.1251 0.1789 -0.0218 -0.0514 1.0000 Public -0.1027 -0.0134 0.0277 -0.0227 0.0996 0.0919 0.0577 0.0285 0.0370 1.0000

Tables 6: correlation matrices of independent variables

Multilevel analyses

The first three multilevel analyses use the average CITO score (cito_average) as the dependent variable. The independent variables that are included in the regression are teacher-pupil ratio (teacherpupilratio), standard deviation of teacher-pupil ratio (standdev), supporting staff-pupil ratio (suppstaffpupilratio), socio-economic area of the school (statusneighborhood), school weight (school_weight), share of male teachers in a school (perc_male), total pupils in the school (total_pupils), number of class repeaters in the eight class (perc_classrep), a dummy variable for the school being public or not (public) and the lagged variable of the CITO score of the previous year. The variables are listed by their nested level, the first variables (teacherpupilratio, suppstaffpupilratio, statusneighborhood, school_weight, perc_male, total_pupils and perc_classrep) are on level 1, within the schools. The other variabels (standdev and public) are on level 2, between schools. Three different multilevel analyses are conducted, the first (A1) with all standard variables except for the lagged CITO score. The second analysis (A2) with the lagged variable included. Finally, an analysis is done without the lagged variable and with an interaction term of teacher-to-pupil ratio and the supportive staff-to-pupil ratio (suppxteacher) (A3). The relevant results are depicted in table 6. The extended results are shown in appendix 4.1.

(28)

28

Table 6: Results multilevel analyses: CITO scores | * p < 0.1 **, p < 0.05 ***, p < 0.01

Nearly all outcomes of the analyses are significant at 1%, with exceptions for the teacher-pupil ratio and the standard deviation of the teacher-to-pupil ratio. However, in the third analysis, the interaction effects of teacher-to-pupil with supporting staff-to-pupil are significant at 1% with a coefficient of 262.624. This means that the H0 of the first hypothesis, that there is no significant effect of both the changes in teacher-to-pupil ratio as the size of the standard deviation of the teacher-to-pupil ratio on the average primary school results, must be rejected for this dependent variable. The third analysis shows that changes in the size of teacher-pupil ratio are associated with changes in average CITO score, but only trough the interaction term. This means that there is a correlation between the number pupils per teacher and the final examination scores of those pupils. The H0 of our second hypothesis, on the moderating effects of supporting staff on the effects of teacher-pupil ratio on school performance, is rejected as well. Since the interaction term indicates that the effects of teacher-pupil ratio are dependent on the change of the supporting staff-to-pupil ratio, for this dependent variable. This means that the balance between the supporting staff members and the pupils in a school can moderate the effect of teacher-to-pupil ratio on the average CITO scores. cito_average A1 A2 A3 teacherpupilratio 3.041 1.675 -2.495 suppstaffpupilratio -30.963*** -19.589*** -53.184*** statusneighborhood -1.248*** -.620*** -1.214*** school_weight -.038*** -.024*** -.038*** perc_male .023*** .012*** .023*** total_pupils .003*** .002*** .003*** perc_classrep -7.342*** -6.534*** -7.323*** lagcito_average .390*** suppxteacher 262.624*** standdev 20.877 32.720 2.333 public -.440*** -.241*** -.435*** _cons 535.844*** 327.051*** 536.194***

(29)

29

In addition, the direct effect of the supporting staff ratio on the average CITO scores is significant at 1%. The interpreting of the effects of the supporting staff ratio cannot be done in isolation, since the interaction variable is significant as well (Stattrek, 2019). Thus, if both the teacher-to-pupil ratio and the supporting staff-to-pupil ratio are heightened, the average CITO scores can increase or decrease depending on the balance between the increase of both ratios. This suggests that increasing the number of teachers in a school only has an impact if the number of supporting staff is increased as well and a fine balance is kept. If this balance is off, it can even be negatively correlated with the average CITO scores.

Furthermore, the first analysis shows that if the neighborhood in which a school is located is assigned a lower socio-economic status the school results are on average between 0.620 and 1.248 point lower compared with schools without a lower socio-economic status, ceteris paribus. If the schools weight increases by one, the average school results are on average between 0.024 and 0.038 lower, ceteris paribus. Both effects are significant at 1%. These effects are in line with the expectations from scientific literature (Driessen, 2005; Gordon & Monostiriotis, 2006; Oomen, 2010; Bussemaker; 2016; Broekroelofs, 2018). Having a higher share of men in the teaching staff and a higher number of pupils in the school increases the average CITO score, with an increase of one in the percentage male teachers increasing the average CITO scores with between 0.012 and 0.023 point and an additional increase of one extra pupil in the school increasing the average CITO by a score between 0.002 and 0.003, ceteris paribus. These effects are significant at 1% as well and in line with the conclusions of Mills (2000) and De Vries (2014). Having 1% more class repeater among the pupils that take the CITO examination lowers the average CITO score of that school by between 6.515 and 7.342 points. This effect is significant at 1%, ceteris paribus. Finally, whether the school is a public school or not matters. If the school is a public school or a special public school the average school results are on average between 0.241 and 0.440 lower compared to schools that have no ‘public’ denomination, ceteris paribus. This effect is significant at 1%. Finally, the lagged score ratio that is included in the A2 and A3 is significant at 1%, ceteris paribus. If the schools previous CITO score is a point higher the average CITO scores of that year are on average 0.390 point higher.

The next analyses include similar independent variables, but percentage of students that earn a school advice of havo or higher (perc_havohigher) as the dependent variable. The lagged variable

(30)

30

uses the school advice instead of CITO scores. The main results are depicted below in table 7, the full results are shown in appendix 4.2.

Table 7: Results multilevel analyses 2: school advice | * p < 0.1, ** p < 0.05, *** p < 0.01

The results of the multilevel analyses with school advice (perc_havohigher) are in part similar to the results generated in the first analyses. In the analyses without the interaction term the teacher-pupil ratio and the standard deviation do not have a significant effect on the school advice of the primary schools at 10%, ceteris paribus. For the dependent variable school advice, the H0 of the first hypothesis is therefore not rejected. There is no significant effect of both the changes in teacher-to-pupil ratio as the size of the standard deviation of the teacher-to-pupil ratio on the average primary school results. Remarkably, in the final analysis with the interaction term the effect for the teacher-pupil ratio is significant not significant at 10% either, ceteris paribus. This means that the H0 of the second hypothesis, that there are no significant moderating effects of supporting staff on the effects of teacher-pupil ratio on school performance, is not rejected as well. The number of supporting staff members per pupil in a school do not moderate the effects of teacher-to-pupil ratio on the school advices.

perc_havohigher A1 A2 A3 teacherpupilratio 9.775 -8.309 -1.247 suppstaffpupilratio -87.795*** -40.722** -132.393*** statusneighborhood -4.899*** -1.813*** -4.831*** school_weight -.169*** -.088*** -.168*** perc_male -.010 .005 -.010 total_pupils .024*** .012*** .024*** perc_classrep -32.682*** -31.386*** -32.649*** laghavohigher .501*** suppxteacher 528.560 standdev 1.659 120.878 -36.697 public 1.127** .415 1.138** _cons 48.779*** 26.896*** 49.480***

(31)

31

However, the supporting staff-pupil ratio has a negative coefficient between 40.722 and 132.393 which is significant at 1%. This means that if the staff-pupil ratio increases by 0.1 the average percentage of school advices being havo or higher decreases by between 4.072 and 13.239 percent point, ceteris paribus. The effects of having a low socio- economic status of the neighborhood are associated with a decrease of percentage school advices of havo and higher with 1.813 to 4.899 percent point, ceteris paribus. Having the school weight increase by one decreases this percentage by 0.088 to 0.169 percentage points, ceteris paribus. Both effects are significant at 1%. Furthermore, in all three analyses the effects of larger share of male teacher have no significant effect on the share of the school advices that is havo or higher, ceteris paribus. This is remarkable given that the earlier analyses showed that a higher percentage of male teachers does significantly positively influence the average CITO score. It suggests that the on average higher CITO scores generated by schools with more male teachers do not lead to an on average higher percentage school advices being havo or higher. The effects of an additional percent of class repeaters seems to have a devastating effect on the percentage of school advices being havo or higher. This lowers between 31.386 and 32.682 percent points at 1% significance, ceteris paribus. In turn, the lagged variable indicates that having on average a higher percentage of school advices being havo or higher in the previous year positively affect the percentage of the following year. This effect is significant at 1%. Finally, the impact of being a public school or not is a bit ambiguous, with the significance of the effect being eliminated if the lagged variable is included. In the other two analyses it indicates that being a public school increases the school advices being havo or higher with 0.415 to 1.138 percentage points, ceteris paribus. These effects are significant at 5%. This is remarkable, with the results from the first analyses suggesting that public schools having on average lower CITO scores. It could suggest that on average public schools do not prepare their pupils optimally for the CITO examination but estimate their pupils to go to more advanced secondary schools nonetheless.

(32)

32

Chapter 5: Conclusions and policy recommendations

In this study, we sought answers to the question whether there is a significant effect of both the changes in teacher-to-pupil ratio as the size of the standard deviation of the teacher-to-pupil ratio on the average primary school results of the final examination. In addition, the moderation effect of the supporting staff-to-pupil ratio on this effect was studied. Using multilevel analysis methods with different sets of variables, we examined the effects on both the average CITO score and the school advice of the schools. We find that the teacher-to-pupil ratio is significantly correlated with the average CITO scores of schools, but only through the interaction variable with supporting staff-to-pupil ratio. The first hypothesis: there is no significant effect of both the changes in

teacher-to-pupil ratio as the size of the standard deviation of the teacher-to-teacher-to-pupil ratio on the average primary school results, must therefore be rejected for this dependent variable. The results suggest that a

fine balance must be kept between increasing the ratio of teachers-to-pupils and the number of supporting staff-to-pupils to be associated with an increase in the average CITO scores of a school. If the balance is off, this is even negatively correlated with the average CITO scores. It also shows that supporting staff-to-pupil ratio has a moderating effect on the correlation between teacher-to-pupil ratio and the average CITO score, rejecting the second hypothesis that there is no significant

effect of the changes in the supporting to-pupil ratio and/or the size of the supporting staff-to-pupil ratio on the effect of changes in teacher-staff-to-pupil ratios on the average primary school results. The size of the teacher-to-pupil ratio, the standard deviation of teacher-to-pupil ratio and

the interaction variable of teacher-to-pupil ratio and supporting staff-to-pupil ratio do not seem to be correlated to the average school advices of the schools. Both the hypotheses are therefore not rejected for the dependent variable of school advice. This suggests that a better balance in teacher-to-pupil ratio and supporting staff-teacher-to-pupil ratio can prepare the pupils better for their final school examination, but the school advice will remain stable regardless of the teacher-to-pupil ratio and the supporting staff-to-pupil ratio. In addition, increasing the supporting staff-to-pupil ratio while keeping the teacher-to-pupil ratio the same is associated with a decrease in the height of school advices. A possible explanation could be that schools with more supporting staff could use this additional staff for educational tasks for which teachers are better suited and thus decrease the performance of their pupils. Another reason could be that bad performing schools attract more supporting staff to help increase the school’s performance by giving the teaching staff more supporting hands. The overall results suggest that putting effort in balancing the teacher-to-pupil

The impact of change in teacher-to-pupil ratios on school performance