UvA-DARE is a service provided by the library of the University of Amsterdam (http://dare.uva.nl)

UvA-DARE (Digital Academic Repository)

Empirical studies in the economics of education

Ruijs, N.M.

Link to publication

Citation for published version (APA):Ruijs, N. M. (2015). Empirical studies in the economics of education. Amsterdam: TIER.

General rightsIt is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s),other than for strictly personal, individual use, unless the work is under an open content license (like Creative Commons).

Disclaimer/Complaints regulationsIf you believe that digital publication of certain material infringes any of your rights or (privacy) interests, please let the Library know, statingyour reasons. In case of a legitimate complaint, the Library will make the material inaccessible and/or remove it from the website. Please Askthe Library: https://uba.uva.nl/en/contact, or a letter to: Library of the University of Amsterdam, Secretariat, Singel 425, 1012 WP Amsterdam,The Netherlands. You will be contacted as soon as possible.

Download date: 02 Oct 2020

TIER Research SeriesUniversity of Amsterdam

This thesis consists of four studies in the economics of education. All chapters use

applied microeconometric techniques to answer questions on education.

Chapter two studies determinants of school choice in Amsterdam. Contrasting to

a popular argument on school choice, quality indicators are not consistent predictors

of school choice. Instead, students prefer schools close to their home, and appear to

prefer schools that many of their former classmates choose.

Chapter three investigates the impact of losing a school admission lottery on

achievement. It turns out that lottery-losing students are educated in schools with

less advantaged peers. Yet, losing an admission lottery does not harm students’

academic achievement after four years of secondary education.

Chapter four studies the causal effects of Montessori education by using school

admission lotteries. On both academic and socio-emotional outcomes, the results

show that Montessori education provides an alternative route towards similar student

outcomes.

Chapter five investigates whether the presence of students with special needs in

regular education classrooms affects the academic achievement of their classmates.

Three empirical strategies yield consistent results: in a context with substantial

additional funding, inclusive education does not help or harm the academic

achievement of regular students.

Nienke Ruijs (1985) holds a Research Master degree in Educational Sciences (2009)

and a Master degree in Clinical Developmental Psychology (2009) from the University

of Amsterdam. Before that, she obtained a Bachelor in Psychology (2006) and in

Educational Sciences (2007) from the same university. She wrote her PhD thesis at

TIER-UvA, thereby switching fields to the economics of education at the Faculty of

Economics and Business. She is currently employed at the UvA as a postdoc for Yield,

which is a research priority area binding researchers in education, child development,

psychology and economics.

TRS n

o. V

I – Emp

irica

l Studie

s in the Ec

ono

mic

s of Ed

uca

tion

– Nie

nke

Ru

ijs

Top Institute for Evidence

Based Education Research

Empirical Studies in the Economics of EducationNienke Ruijs

EMPIRICAL STUDIES IN THEECONOMICS OF EDUCATION

© Nienke Ruijs, Amsterdam, 2014

All rights reserved. No part of this publication may be reproduced, stored in aretrieval system or transmitted in any form, or by any means, electronic, mechanical,photocopying, recording, or otherwise, without the prior permission in writing, fromthe author.

ISBN 978-94-003-0090-3

Cover design: Raadhuis voor creatieve communicatie, Alkmaar

This book is no. VI of the TIER Research Series, a PhD thesis series published byTIER.

EMPIRICAL STUDIES IN THEECONOMICS OF EDUCATION

ACADEMISCH PROEFSCHRIFT

ter verkrijging van de graad van doctor

aan de Universiteit van Amsterdam

op gezag van de Rector Magnificus

prof. dr. D.C. van den Boom

ten overstaan van een door het college voor promoties

ingestelde commissie,

in het openbaar te verdedigen in de Agnietenkapel

op vrijdag 6 februari 2015, te 10.00 uur

door

Nienke Maria Ruijs

geboren te Naarden

Promotiecommissie

Promotoren: Prof. dr. H. Maassen van den Brink

Prof. dr. H. Oosterbeek

Overige leden: Prof. dr. R.J. Bosker

Prof. dr. B. van der Klaauw

Prof. dr. F.J. Oort

Prof. dr. E.J.S. Plug

Prof. dr. I.F. de Wolf

Faculteit Economie en Bedrijfskunde

Dankwoord

Het schrijven van dit dankwoord is één van de meest bijzondere taken uit mijn pro-motietraject. Promoveren wordt vaak als eenzame bezigheid gezien, maar dat is nietmijn ervaring. Er zijn veel mensen die op één of andere wijze hebben bijgedragen aande totstandkoming van dit proefschrift. Ik ben blij dat ik hier de ruimte heb om hente bedanken.

Ik wil beginnen bij mijn promotoren. Hessel en Henriëtte, ik wil jullie bedankenvoor het vertrouwen dat ik als psycholoog en onderwijskundige in staat zou zijn om mesnel te bekwamen in de analysetechnieken die gebruikt worden in de onderwijseconomie.Ik ben de afgelopen jaren doordrongen geraakt van het belang van causaal onderzoek.De technieken die ik heb geleerd neem ik met heel veel plezier mee naar toekomstigebanen.

Hessel, als promotor was je altijd bereikbaar voor vragen of om iets te overleggen,en je had altijd nuttig advies. Ik heb veel van je geleerd over academisch schrijven. Ikheb bij een paar stukken tekst geworsteld met hoe ik dat het beste op kon schrijven.De manier waarop jij het herformuleerde was soms zo logisch dat ik me afvroeg waaromik dat niet zelf had bedacht.

Henriëtte, jouw concrete vraag om output heeft me steeds gemotiveerd om resul-taten vlot op papier te zetten. Ook wil ik je bedanken voor het initiëren van TIER. Ikvond het de afgelopen jaren interessant om te zien waar collega’s van andere instellin-gen binnen TIER aan werken en ik ben blij met de contacten die ik dankzij TIER hebopgedaan.

Ik heb de afgelopen jaren erg veel gehad aan mijn collega’s bij TIER-UvA/ HumanCapital. Ik moest even wennen aan de directheid van de work-in-progress sessies; maarjuist dankzij de kritische opmerkingen heb ik er veel van geleerd. De gezamenlijkelunches waren een fijne onderbreking van de dag. José, Thomas, Lydia, Ferry, Stephen,Diana, Jona, Lenny, Monique, Adam, Erik, Noemi en Nadine; Thanks for the goodtimes we shared! Adam en Monique, ik wil jullie bedanken dat ik altijd bij julliebinnen kon lopen als ik iets te vragen had. Erik, dank je wel voor de talloze malenvan eerlijk en nuttig commentaar. Ik heb je geregeld gezien als mijn derde promotor.Nadine en Noemi, I am thankful that I could always use you as guinee-pigs for ideas.

Over the past years, you have been more than collegues to me. Obviously, we wereoffice roommates/neighbors, but I am most happy to have shared important personalmoments with you. I am grateful for your friendship and am very curious to see whereour futures will bring us!

De vier projecten in mijn proefschrift waren niet mogelijk geweest zonder de be-trokkenheid en behulpzaamheid van mensen bij de Onderwijsinspectie, het Ministerievan Onderwijs, de Dienst Uitvoering Onderwijs en de Dienst Maatschappelijke On-twikkeling van de Gemeente Amsterdam. Een aantal mensen wil ik in het bijzonderbedanken. Inge de Wolf, dank voor alle hulp bij het verkrijgen van de data voorde projecten over Schoolkeuze en Loten. Het was een lang proces, maar het is onsgelukt! Dank je wel ook voor alle input bij het onderzoek naar Loten in Amsterdam,jouw opmerkingen waren enorm nuttig in het aanscherpen van de interpretatie van deresultaten en ik kijk uit naar het vervolg van dit project.

Mark de Boer, Hans Plomp, André Hooijveld, Elma Wesselink en Ton Veugen:heel veel dank voor het verstrekken van de gegevens voor de onderzoeken naar Loten,Montessori en/of Zorgleerlingen. Jullie hebben bij DUO een goudmijn aan gegevens.Ik hoop dat ik ook de komende jaren vaak de kans mag krijgen om gegevens van DUOte gebruiken voor onderwijsonderzoek. Herman Ozinga en Kees Waijenberg, dankdat ik de gegevens uit ELKK mocht gebruiken voor de onderzoeken naar Loten enSchoolkeuze.

Ik wil alle scholen bedanken die de afgelopen jaren tijd hebben gemaakt voor hetproject over Loten in Amsterdam, jullie aanvullingen en correcties waren onmisbaarvoor het onderzoek. Ook wil ik Ivo Richaers en de scholen die mee wilden doen met hetonderzoek naar Huiswerkbegeleiding bedanken. Ik ben vereerd door de grote interessevoor dit project, en hoop dat we het in de toekomst alsnog van de grond krijgen.Verder wil ik de twee Montessorischolen bedanken voor hun bereidheid om mee tedoen met het onderzoek naar de effecten van Montessorionderwijs. Wiebe Brouwer enIngrid van der Neut, ik realiseer me dat dit voor jullie scholen een behoorlijk spannendproject moet zijn geweest, en bewonder de moed die nodig was om hieraan mee te doen.Rolf Schoevaart en Brigitte van Marwijk, dank voor de praktische aanlevering van degegevens. Michael Rubinstein, dank je wel voor het enthousiasme waarmee je mij enhet onderzoek naar Montessorionderwijs hebt geintroduceerd in de Montessoriwereld.Sjoerd Karsten, dank je wel voor de hulp bij het opstarten van het onderzoek en voornuttig commentaar op eerdere versies.

In de jaren dat ik het grootste deel van mijn dagen al proefschrift-schrijvend hebdoorgebracht, was het extra fijn om tijd door te brengen met familie en vrienden. Jelte,Sietske en Hiske, ik ben heel blij met de band die wij als broer en zussen met elkaarhebben. Ik vind het altijd gezellig om met jullie te zijn. Nienke en Bart, Mijntje

en Bart, Linda en Ivo: Ik geniet altijd erg van jullie gezelschap en uitjes met onzegroeiende hoeveelheid kinderen.

Pap en Mam, dank dat jullie mij hebben opgevoed tot wie ik ben. Ik ben erg blij metde vaardigheden die ik vanuit huis mee heb gekregen. Tijdens mijn promotieonderzoekis doorzettingsvermogen (’afmaken waar je aan begint’) een bijzonder nuttige waardegebleken. Mam, ik wil je ook bedanken voor alle praktische hulp. Zonder jou was hetniet gelukt om mijn proefschrift in vijf jaar af te ronden. Sybren en Jeldou, wat benik blij dat jullie er zijn. Wat is het leuk om jullie op te zien groeien, te zien lachen,kruipen, praten en om jullie blije gezichtjes te zien als ik thuis kom.

Melle, de laatste woorden zijn voor jou. Het klinkt zo obligaat om te zeggen: zonderjou was het niet gelukt, maar ik kan het niet anders zeggen. Jij hebt een erg belangrijkerol gespeeld in mijn promotieonderzoek. Het eerste jaar was het enorm fijn om naarjou toe te kunnen gaan met wiskunde of econometrievragen wanneer ik dacht dat ietseen ’domme’ vraag was. Daarna was het fijn dat ik altijd mijn verhaal kwijt kon alshet even niet zo voorspoedig liep, en dat ik samen met jou kon bedenken hoe ik ietshet handigste geregeld kon krijgen. Ik hou van je, en hoop dat we nog heel veel mooiejaren mogen delen.

Contents

1 Introduction 1

2 School choice in Amsterdam: Which schools are chosen when schoolchoice is free? 52.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.2 Context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.2.1 School choice in the Netherlands . . . . . . . . . . . . . . . . . . 82.2.2 Secondary education in the Netherlands . . . . . . . . . . . . . 82.2.3 Secondary school choice in Amsterdam . . . . . . . . . . . . . . 9

2.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102.3.1 Student information . . . . . . . . . . . . . . . . . . . . . . . . 102.3.2 School information . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.4 Empirical strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152.4.1 Main analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152.4.2 Peer effects in school choices . . . . . . . . . . . . . . . . . . . . 19

2.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202.5.1 Main findings . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202.5.2 Peer effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

2.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322.A Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

2.A.1 School*year fixed effects . . . . . . . . . . . . . . . . . . . . . . 332.A.2 Additional table . . . . . . . . . . . . . . . . . . . . . . . . . . 35

3 The impact of losing a school admission lottery on school outcomes 373.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373.2 The Dutch context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

3.2.1 Secondary education in the Netherlands . . . . . . . . . . . . . 393.2.2 Secondary school choice and admission lotteries in Amsterdam . 41

3.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423.4 Empirical strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 463.5 Treatment characteristics . . . . . . . . . . . . . . . . . . . . . . . . . 48

3.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 533.A Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

3.A.1 Balancing tests with original lottery data . . . . . . . . . . . . 563.A.2 Additional tables . . . . . . . . . . . . . . . . . . . . . . . . . . 60

4 The effects of Montessori education: Evidence from admission lotter-ies 634.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 634.2 Montessori education . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

4.2.1 Montessori primary education . . . . . . . . . . . . . . . . . . . 654.2.2 Montessori secondary education . . . . . . . . . . . . . . . . . . 66

4.3 Secondary education in the Netherlands . . . . . . . . . . . . . . . . . 664.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

4.4.1 Admission lotteries in Montessori schools . . . . . . . . . . . . 684.4.2 Data on academic achievement . . . . . . . . . . . . . . . . . . 704.4.3 Questionnaire on socio-emotional functioning . . . . . . . . . . 73

4.5 Empirical strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 754.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

4.6.1 The effects of Montessori education on academic outcomes . . . 774.6.2 The effects of Montessori education on socio-emotional outcomes 814.6.3 Heterogeneous effects . . . . . . . . . . . . . . . . . . . . . . . . 834.6.4 Difference between lottery participants and students with prior-

ity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 844.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 864.A Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

4.A.1 Details on questionnaire sampling and response . . . . . . . . . 894.A.2 First stage coefficients . . . . . . . . . . . . . . . . . . . . . . . 924.A.3 Heterogeneous effects . . . . . . . . . . . . . . . . . . . . . . . 94

5 Special needs students in regular education: Do they affect their class-mates? 995.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 995.2 Related literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1005.3 The Dutch context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

5.3.1 Primary and secondary education in the Netherlands . . . . . . 1025.3.2 Inclusive education . . . . . . . . . . . . . . . . . . . . . . . . . 103

5.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1055.4.1 Data primary education . . . . . . . . . . . . . . . . . . . . . . 1055.4.2 Data secondary education . . . . . . . . . . . . . . . . . . . . . 107

5.5 Empirical strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1095.5.1 Empirical strategy 1: Student fixed effects . . . . . . . . . . . . 1095.5.2 Empirical strategy 2: School fixed effects . . . . . . . . . . . . 1125.5.3 Empirical strategy 3: Exploiting neighborhood variation . . . . 116

5.5.3.1 Neighborhood IV . . . . . . . . . . . . . . . . . . . . . 1195.5.3.2 Neighborhood fixed effects . . . . . . . . . . . . . . . . 120

5.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1205.6.1 Results for empirical strategy 1: Student fixed effects . . . . . . 1205.6.2 Results for empirical strategy 2: School fixed effects . . . . . . . 1225.6.3 Results for empirical strategy 3: Exploiting neighborhood variation126

5.6.3.1 Neighborhood IV . . . . . . . . . . . . . . . . . . . . . 1265.6.3.2 Neighborhood fixed effects . . . . . . . . . . . . . . . . 126

5.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1295.A Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

6 Summary 151

Chapter 1

Introduction

This thesis consists of four studies in the field of the economics of education. Eachchapter can be read as an independent research paper. In this first chapter, I brieflyintroduce the topics of the four studies and discuss how they are related.

Chapter 2 studies the determinants of secondary school choice in the city of Amster-dam, with a focus on the role of information about school quality. In many countries,school choice is a prominent issue in the education debate. In countries such as the USand the UK, this debate led to policies increasing school choice, such as charter schools,vouchers and inter- or intra-district school choice. One of the main arguments in favorof school choice is that it creates competition between schools, which gives schools anincentive to improve education quality. For this channel to work, it is important thatstudents and parents indeed base their choices on school quality.

The city of Amsterdam provides an ideal environment to investigate whether par-ents are affected by publicly available information on school quality: the Netherlandshas a long tradition of free school choice, tuition fees are low, and schools in Amster-dam are easy to reach since distances are relatively short. Using conditional logit andmixed logit discrete choice models, Chapter 2 studies whether information about schoolquality, distance and conducting a lottery in the previous year predict students’ schoolchoices. Contrasting to the often made ’quality-argument’, it turns out that informa-tion about school quality and changes in school quality are inconsistent predictors ofschool choice. Instead, students prefer schools closer to their home and the results areconsistent with the notion that students are influenced by the secondary school choicesof their primary school peers. Further, there is some evidence that a school admissionlottery in the previous year deters students from choosing that school.

Next to free school choice, these school admission lotteries are an important aspectof secondary school choice in Amsterdam. The secondary schools differ in popularityand the number of students choosing a school sometimes exceeds the number of studentsthat can be enrolled. Oversubscribed schools are not allowed to ’cherry-pick’ students:

1

they conduct lotteries to allocate the available places. Every year, the school lotteriesare prominently covered in the local news, and parents and students are afraid tolose a lottery. Chapter 3 investigates whether losing a school admission lottery for asecondary school in Amsterdam actually affects students’ school performance.

Comparing students who win and lose a school admission lottery, it turns out thatlosing students are educated at schools with ’worse’ peers in terms of prior academicachievement and social background than lottery winners. Students’ academic achieve-ment, however, is not harmed by losing a school admission lottery. After four yearsof secondary education, lottery losing students even seem to achieve somewhat betterthan lottery winning students. The results are in line with the school assessment bythe Dutch education inspectorate: lottery losers attend schools that perform betteron grade progression in the lower grades, although their schools score worse on gradeprogression in the higher grades and on exam scores. As most students in this studyare too young to have finished secondary school, it is yet unclear whether the resultscarry over to the final exams. The results can be read as provisional reassurance for(parents of) students losing school admission lotteries. Contrasting to the strong nega-tive emotions voiced after losing a school admission lottery, the academic achievementof lottery losing students measure up to those of lottery winning students after fouryears of secondary education.

Since Chapter 3 studies the majority of Amsterdam schools that conduct lotteries,the results focus on a variety of rather different schools. School admission lotteriesalso yield an interesting opportunity to study specific educational methods. Chap-ter 4 exploits admission lotteries to study the effects of being exposed to Montessorieducation. Proponents of Montessori education often claim that it has positive ef-fects on students’ independence, while critics argue that this may come at the cost oflower academic achievement. Even though many students in many different countriesare educated using the Montessori pedagogy, little is known about the effectivenessof Montessori methods to support these claims. The reason is that self-selection intoMontessori schools makes it difficult to compare students educated in these schools tostudents educated in regular schools.

By using school admission lotteries, Chapter 4 circumvents the selection issues; theadmission lotteries basically create a series of randomized experiments. The resultsbased on this approach indicate that neither the claims of the proponents nor of thecritics are supported by the evidence. Students exposed to Montessori education insecondary school do not score better than other students on various measures of inde-pendence. Nor does the academic achievement of Montessori students differ from theacademic achievement of students following regular secondary education. Montessoristudents obtain their secondary school degree without delay at the same rate and withsimilar grades as non-Montessori students. Overall, the results of this study indicate

2

that Montessori education provides an alternative route to similar outcomes.

The final empirical study of this thesis focusses on the classmates of students withspecial educational needs. Chapter 2 found school choice patterns that are consistentwith students’ choices of secondary school being influenced by their peers in primaryschool. This supports the notion that peers are important for students. Chapter 5continues to study this topic by investigating whether the inclusion of students withspecial educational needs (such as visual or behavioral problems) in regular educationclasses affects their regular education peers. Since the 1990’s, there has been an in-ternational trend towards educating children with special educational needs in regularschools instead of special schools. This trend has lead to fierce debates, where theconsequences for the academic achievement of regular education students are a mainconcern. On the one hand, teachers and parents are afraid that students get distractedby the behavior of students with special educational needs and state that studentswith special needs demand more teacher attention at the expense of ’regular’ students.On the other hand, advocates of inclusive education claim that there is more teachersupport in inclusive classrooms, which could also positively affect the regular educationstudents.

Using three different empirical strategies, Chapter 5 studies the impact of inclu-sive education on the academic achievement of students without special educationalneeds. Specifically, it utilizes student fixed effects, school fixed effects and neighbor-hood variation in the number of special needs students. The results consistently showthat the presence of students with special educational needs has no impact on the aca-demic achievement of regular education students. The results imply that in a situationwith substantial additional funding, inclusive education does not adversely or favorablyaffect the academic achievement of regular education students.

The final chapter, Chapter 6, summarizes the main findings and conclusions of thefour studies presented in this thesis.

3

Chapter 2

School choice in Amsterdam: Whichschools are chosen when school choiceis free?1

2.1 Introduction

This chapter analyses the determinants of secondary school choice in the city of Am-sterdam. Students in this city can choose any school that offers education at theiracademic level, tuition fees are low, and schools are easy to reach due to relativelyshort distances. These features create an ideal environment to examine whether schoolchoices are affected by published information about school quality. We do so by es-timating discrete choice models using data of individual school choices for the years2007-2010. Analyzing individual school choices (instead of aggregate school enrollmentdata) allows us to compare the impact of published information on school quality onschool choices with the impact of other factors including traveling distance, whetherthe school was oversubscribed in the previous year, and the number of classmates fromprimary school that attends the secondary school.

Analyzing the impact of published school quality information on school choice isimportant from the perspective of the debate on school choice and competition. Themain arguments in favor of these policies are that by having more choice students canfind a school that better matches their specific needs and that exposure to competi-tion gives schools an incentive to improve education quality.2 The direct evidence ofthe impact of school competition on school quality and student achievement is mixed(Böhlmark & Lindahl, 2013; Cullen et al., 2006; Hoxby, 2000a, 2007; Hsieh & Urquiola,

1This chapter is based on Ruijs and Oosterbeek (2014).2Throughout this chapter we will phrase as if students are the decision-makers. We acknowledge

that in reality their parents have an important say in this.

5

2006; Rothstein, 2007).3 A prerequisite for such an impact to emerge, is that schoolchoices are at least partially driven by school quality. If this is not the case, improve-ment of school quality does not raise student inflow and is therefore not a dimensionon which schools should compete.

In a recent study examining whether published information on school quality af-fects secondary school enrollment in the Netherlands, Koning & Van der Wiel (2013)found that higher quality scores indeed increase enrollment. There are three reasons toreexamine the findings of that study. First, Koning & Van der Wiel analyze data fromthe whole country thereby also including more remote areas where most schools are defacto local monopolists. By analyzing data from a city where the number of schoolsat any academic level is abundant, we focus on a situation where students have morepossibilities to respond to published quality information.

Second, due to data limitations Koning & Van der Wiel use enrollment in the thirdyear of secondary school as a proxy for applications in the first year. Since on average3.9% of Dutch secondary school students repeat a grade in a given year in the first threeyears of secondary school (CBS, 2012) , this implies that in the third year, 7.8% of theenrollees have been replaced by 11.7% repeaters from previous cohorts. These repeatersfrom previous cohorts have based their school choices on published information onschool quality from earlier years. In addition, a non-negligible fraction of 23% ofsecondary school students are transferred to a higher or lower academic track by thetime they get into the third grade or have moved to another school (Onderwijsinspectie,2007). Moreover, since the most popular schools are oversubscribed and cannot admitall applicants, school choice in response to published information on school quality isbest measured by applications instead of enrollment. Our data on school choice in thecity of Amsterdam contains information on first year applications.

Third, the analysis of Koning & Van der Wiel does not include students in thelowest levels of pre-vocational education, and thereby omits 25% of Dutch secondaryschool students. We extend the analysis to all academic levels.

Related to this chapter is also the work of Hastings et al. (2009) who use data fromprimary and middle school choices from a school district in the US. In that district, aschool choice program was introduced where parents could list their top three schools.Using exploded mixed logit models, they find that the weight parents place on school

3In a famous study Hoxby (2000a) finds that more school competition (measured as the number ofschool districts in a metropolitan area) boosts student achievement in the US. Identification is basedon variation in the number of school districts caused by the numbers of streams (rivers). Rothstein(2007) has shown that Hoxby’s results are sensitive to the way in which streams are counted (see alsoHoxby, 2007). Hsieh & Urquiola (2006) find no evidence that choice, triggered by the provision ofvouchers, improved student achievement in Chile. Using a comparable reform in Sweden, Böhlmark& Lindahl (2013) find, however, a positive effect of choice on student achievement. Using data fromadmission lotteries from the Chicago public school system, Cullen et al. (2006) find that students wholost the lottery – and therefore have a restricted choice set – are not harmed in their achievement.

6

characteristics is heterogeneous. Parents, especially those of high SES, tend to preferschools with high test scores. Also, parents tend to prefer schools in which the majorityof the students has the same race as they have. These results imply that minorityparents face a trade-off between high performance and ethnic composition. Further,the distance to schools and the availability of transportation are relevant determinantsof school choice. While Hastings et al. focus on test scores as indicator of schoolquality, we look at the impact of published school quality information. Test scores arearguably a measure of gross school quality including the quality of the student inflow,the published school quality information that we look at is intended to be a measureof a school’s value added (Subsection 2.3.2 provides details).

To match students to schools, the city of Amsterdam uses an algorithm that is verysimilar to the so-called Boston mechanism. If the school to which a student applies,is oversubscribed and the student is not accepted at that school (loses the lottery),he can subsequently only choose from the schools that then still have vacant seats. Itis well-known that this matching algorithm can give rise to strategic behavior wherestudents do not report their truly preferred school (Abdulkadiroğlu & Sönmez, 2003;Calsamiglia & Guell, 2014). We examine strategic choices by including a variablecapturing whether the school was oversubscribed in the previous year.

In addition to measures of school quality and oversubscription in the previous year,we also inquire the impact on school choice of the number of classmates from primaryschool that attends a secondary school. Peer effects are intrinsically difficult to identify.De Giorgi et al. (2010) exploit that peer groups do not fully overlap to identify peereffects in the choice of college majors. Lacking such a source of variation and inthe absence of random assignment of classmates as in Sacerdote (2001), we restrictourselves to testing the null-hypothesis of no peer effects in school choice.4

The main results of this chapter are threefold. Unlike Koning & Van der Wiel(2013) we find that school quality and changes in school quality are inconsistent pre-dictors of school choice in Amsterdam. Second, we find some evidence that a lottery inthe previous year deters students from choosing that school, suggesting that strategicbehavior occurs. Third, our results strongly reject the null-hypothesis of no peer effectsin school choice. When we correct for the general popularity of a secondary school ina primary school, we find that when a larger share of the primary school peers choosesfor a certain secondary school, students are more likely to pick that school as well.Although this result does not prove the importance of peer effects in school choices, itis consistent with it.

This chapter proceeds as follows. The next section describes the context of sec-ondary school choice in Amsterdam. Section 2.3 describes the data. Section 2.4provides details of the empirical strategy that we employ. Section 2.5 presents and

4Angrist (2014) gives a critical assessment of the peer effects literature.

7

discusses the empirical findings. Section 2.6 summarizes and concludes.

2.2 Context

2.2.1 School choice in the Netherlands

The Netherlands has a long history of free school choice. The constitution of 1848guarantees the freedom to provide education. In 1917, this freedom was extended withthe amendment that all schools receive state funding (Eurydice, 2009). In currentpractice, this means that privately-run schools (either with a religious background orsubscribing to specific pedagogical approaches such as Montessori and Dalton) arepublicly funded at the same level as publicly-run schools. In return, schools have toadhere to certain rules. In particular, they are subject to quality inspections by theDutch Education Inspectorate.

For students, these regulations imply that they are free to choose the school theywant; they are not restricted by measures such as catchment areas.5 Dutch studentsmake extensive use of this option: 70% of the students in primary education and75% of the students in secondary education is enrolled in a publicly-funded privately-run school (CBS, 2009). Privately funded schools are virtually non-existent: in 2009,only 0.3% of the students in secondary education attended a privately funded school(Onderwijsinspectie, 2010).

The government funding of schools is to a large extent dependent on student num-bers, in which the money follows the student. For disadvantaged students there areadditional funding schemes. In primary education there is the system of weighted stu-dent funding. For disadvantaged students in terms of low parental education, schoolscan get additional funding up to 1.2 times the regular per student funding. In thischapter, we will use the the weighted student funding to identify disadvantaged stu-dents.

2.2.2 Secondary education in the Netherlands

Dutch secondary education starts around age 12 and lasts four to six years. The lengthof secondary education depends on the school track: the Netherlands has a tracked sec-ondary school system. The lowest tracks (pre-vocational secondary education, vmbo)last four years, and give access to vocational education programs. Within the pre-vocational track, there are four different levels, each giving access to different levels of

5 A small number of municipalities have put restrictions on primary school choice to foster deseg-regation (Ladd et al., 2011). In those projects, preferences are important in placement decisions aswell.

8

vocational education programs. In this chapter, they will be indicated with the num-bers I to IV, with IV being the highest level.6 The intermediate track (senior generalsecondary education, havo) takes five years, and gives access to higher professionaleducation. The highest track (pre-university education, vwo) takes six years, and givesaccess to university education (Eurydice, 2009).

Which school track a student should take is mainly decided at the end of primaryeducation. It is partly determined by standardized tests (in most cases the nationwideexit test called the “citotoets"), and partly determined by the assessment of the primaryschool teacher. Not all secondary schools offer all school tracks, so the track advice isan important factor in secondary school choice. When offering more than one schooltrack, schools are allowed to educate children of different tracks together. Depending onstudent achievement and school policies, students can change track during secondaryeducation. Also, they can decide to take a higher track after finishing a lower track.7

Subject to some conditions, students can choose which courses they want to takein the second half of secondary school. Secondary schools have to follow nationalcurriculum guidelines. Students take centrally determined national exams at the endof secondary school. The national exams count for 50% of students’ final grades, theother 50% is determined by school specific exams taken in the last two or three yearsof secondary education.

2.2.3 Secondary school choice in Amsterdam

Amsterdam is the capital of the Netherlands and is with 750,000 inhabitants its largestcity. Each year, 5,500 to 6,000 primary school students enter secondary education.In the city of Amsterdam, there are about 54 secondary schools, excluding schoolsfor students with special educational needs. Not surprisingly, some schools are morepopular than others. Each year some schools are oversubscribed and conduct lotteriesto allocate the available places.

Starting in 2005, the secondary schools in Amsterdam run a centralized applicationand admission system. In the first round students can only apply to one school thatoffers the track of their academic level. Students that make the standardized test in linewith the primary school teachers’ advice cannot be rejected by the secondary school,which effectively means that it is hard for schools in Amsterdam to select students.8

6Their Dutch names are bbl, kbl, gl and tl. Pre-vocational III (gl) will not be taken into accountin the analyses, since only 2% of the students receive a pre-vocational III advice and only 10 schoolsoffer this school track.

7In that case, students enter the higher track in the year before the final exams. When doingpre-university education after senior general secondary education, for example, students enter in year5 of pre-university education. When graduating at once, they have their pre-university degree after 7years of secondary education instead of 6.

8The definitions for corresponding testscores are strictly prescribed. When a student has a lowercitoscore than can be expected from the teacher’s advice, the secondary school should discuss the

9

When a school is oversubscribed, a lottery is conducted at the school itself. Someschools have over-subscription for some school tracks, but not for others. In that case,the lottery is conducted for each school track separately. Schools are allowed to use alimited number of priority rules: they can grant priority to siblings of current students,children of staff members and to students from a primary school with similar specialprograms (for example, Montessori secondary schools can grant priority to childrenfrom a Montessori primary school). These priority rules have to be announced beforethe application date, so they are known to parents. When losing a lottery, studentshave to apply to one of the schools that after the first round still has places available.In the years that we study, around 5% of the students could not be placed on the schoolof their first choice because they lost the lottery. Note that the school of first choice isnot necessarily the actual first preference: there may be strategic behavior in choosingschools.

2.3 Data

2.3.1 Student information

Data come from the centralized application and placement system of the city of Am-sterdam. This database has information on 21,117 Amsterdam students choosing asecondary school in Amsterdam in the four years from 2007 to 2010. The databaseprovides information on student background characteristics, such as sex and ethnicity(but not income), and on primary school achievement, such as school track advice,citoscore and grade repetition.

For each student, we know which school was listed as the school of first choice andwhether the student was enrolled at that school. For students who are not placed at theschool of their first choice, subsequent choices are also registered. This information is,however, not used here because the choice sets after the first round are not sufficientlyclear.

Using information on students’ school track advice and information on the schooltracks offered by each secondary school, we can create the choice set for each student.Although schools outside of Amsterdam can also be chosen, we limit the choice set toschools in the city of Amsterdam. Schools outside of Amsterdam do not follow the sameenrollment rules and their students are not registered in the Amsterdam enrollmentsystem. Similarly, children outside of Amsterdam can choose schools in Amsterdam.

student with the primary school and/or conduct an extra standardized test. In these cases, thesecondary school has some discretion in rejecting the student, which happens in about 5% of thecases. For most school tracks, the majority of the students (52%) has a test-score in line with theprimary school advice. An exception to this are the vocational levels with additional support. Here,all students are placed after discussing them with the primary school.

10

Because their data is not consistently registered, these children are omitted from theanalysis as well. Because the enrollment procedures are different for special educationalneeds schools, we drop those schools from the sample. Moreover, we drop some studentswho have missing values on key variables.9

Because we know each student’s primary school, we can construct variables thatcapture choices of each student’s peer group. To study whether students take intoaccount the decision of their classmates in primary school, we calculated for eachstudent the shares of classmates in primary school that apply to each of the secondaryschools.

Some secondary schools may traditionally be more popular amongst students fromcertain primary schools. For example, students from a Montessori primary schoolwill more often choose for a Montessori secondary school. Therefore, we computed asimilar variable with the predicted share of classmates applying to each of the secondaryschools. For these predictions we used the actual shares of peers in the different yearsto estimate a linear trend over the four years of our study, and used the predictedvalues of those regressions as a measure of predicted popularity. When including bothvariables, the share of peers basically captures the popularity of a particular secondaryschool among a student’s classmates in deviation of its trend. This is akin to thepopulation variation used by for example Hoxby (2000b).

Table 2.1 reports descriptive statistics on student characteristics. The first columnshows that the pre-university track attracts around 21% of the students. Another 28%of the students enter secondary school at the combined senior general secondary/pre-university track or the senior general secondary track. The remaining 51% of thestudents start secondary education at the pre-vocational tracks. The second columnshows that the share of boys is fairly constant across the different secondary schooltracks. They are only somewhat underrepresented at the two lowest tracks. Column(3) demonstrates the segregation of secondary school tracks in Amsterdam along thelines of immigrant status. While the share of students with both parents born in theNetherlands in the population is only around one third, their share in the highest trackis 0.6 and this decreases monotonically to 0.14 in the lowest track. This carries overto column (5) which shows the share of students without a disadvantaged backgroundby school track. Column (7) shows the monotonic relation between students’ score onthe final test in primary school (citoscore) and their track in secondary school. (The

9Overall, 7432 students are dropped out of the initial sample of 28,549. 3,913 of them are droppedbecause they are living outside of Amsterdam or going to primary schools outside of Amsterdam.1,488 students are dropped because they are going to special education needs schools. 103 studentsare dropped because their primary school is unknown, and 236 students are dropped because theiraddress is not registered. Finally, 1,539 students are dropped because they choose schools that shouldnot be in their choice set given their primary school advice and 153 students are dropped becausethey go to a few very small secondary schools, which do not have enough observations to take intoaccount in this study. These are mainly religious schools, such as Jewish or Islamic schools.

11

scale for this variable runs from 500 to 550.)

2.3.2 School information

Table 2.2 reports descriptive information on the choice set and the chosen school forthe different school tracks. The first column reports the numbers of schools at eachtrack from which students can choose. Many schools appear in this column multipletimes because they offer more than one secondary school track.

Using information on students’ home addresses, we calculated the distances fromtheir house to Amsterdam schools in their choice set. These distances are calculatedin a straight line based on coordinates.10 Columns (2) to (4) provide means andstandard deviations of the distances that students have to travel to the school theychose (column (2)), to the school offering their advised track nearest to where they live(column (3)) and the average over all schools at their advised track (column (4)). Themean distance to the nearest school is around 1 kilometer and this is very similar forthe different school tracks. The mean distance to the school that was actually chosenis around 3 kilometers at all school tracks, whereas the mean of the average distanceto all schools at a given track is 5 to 6 kilometers, again with little variation acrossschool tracks.

We constructed an indicator that equals one for tracks in schools that had a lotteryin the year prior to the year of application.11 Column (5) of Table 2.2 shows the shareof schools that conducted a lottery, averaged over all four years.

Finally columns (6) to (9) provide information about the share of classmates inprimary school that go to the same secondary school (actual and predicted) and theshares of classmates that would go to the same school if they choose their secondaryschools randomly. These shares show that classmates from the same primary schooltend to choose the same secondary school (the actual share of peers in chosen schoolsis larger than the actual share of peers in all schools) even when corrected for the usualpopularity of secondary schools among students from a given primary school (the actualshare of peers in chosen schools is larger than the predicted share of peers in chosenschools).

As described in Section 2.2, Dutch schools are subject to quality inspections by theDutch Education Inspectorate. Their quality information is based on several aspects,

10For a random sample of 100 students, we calculated the road distance as well. The road distanceturned out to be very closely related to the distance in a straight line.

11The information we use for the lottery indicator is from a booklet published annually by themunicipality of Amsterdam. This booklet includes one page of information on each secondary school,together with information on the general enrollment procedure. It is handed out to all students inthe last year of Amsterdam primary schools. From our contacts with schools, we noticed that thebooklets have a few errors on the lotteries. Since students will visit the school before subscribing, andthey will get the correct information there, we decided to adjust these cases to match the informationfrom the schools.

12

Tab

le2.1:

StudentCharacteristics

Bothpa

rents

Belon

ging

to

NSh

are

born

inthe

disadv

antaged

Citoscore

stud

ents

ofbo

ysNetherlan

dsgrou

p

Yes

Missing

Yes

Missing

Mean

SDMissing

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Pre-uni.

4,374

0.50

0.60

0.04

0.14

0.12

544.1

7.4

0.01

Senior

gen./p

re-uni.

2,819

0.49

0.45

0.05

0.27

0.11

540.5

6.1

0.01

Senior

gen.

3,133

0.50

0.39

0.05

0.34

0.12

537.7

6.4

0.02

Pre-voc.IV

/seniorgen.

2,028

0.49

0.28

0.06

0.47

0.10

534.4

6.2

0.02

Pre-voc.IV

2,881

0.50

0.22

0.05

0.54

0.10

531.5

6.7

0.03

Pre-voc.II

2,130

0.49

0.16

0.06

0.67

0.09

526.7

8.1

0.19

Pre-voc.I/II

1,039

0.48

0.14

0.05

0.65

0.15

524.7

9.0

0.33

Pre-voc.I

2,713

0.47

0.14

0.07

0.64

0.19

520.4

10.6

0.58

Total

21,117

0.49

0.34

0.05

0.42

0.12

535.7

10.1

0.12

Note:

Colum

ns(3)to

(6)repo

rtshares.The

’No’

category

isom

itted.

13

Tab

le2.2:

Descriptive

inform

ationon

students’choice

sets

Distance

Actua

lsha

reof

peers

Predicted

shareof

peers

Nchosen

nearest

all

Shareof

chosen

all

chosen

all

Scho

ols

scho

olscho

olscho

ols

scho

ols

scho

olscho

ols

scho

olscho

ols

M(SD)

M(SD)

M(SD)

withlotteries

M(SD)

M(SD)

M(SD)

M(SD)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Pre-uni.

273.09

1.04

5.00

0.08

0.12

0.03

0.11

0.03

(2.13)

(0.71)

(2.90)

(0.12)

(0.06)

(0.11)

(0.06)

Senior

gen./p

re-uni.

213.17

1.20

5.60

0.10

0.14

0.03

0.13

0.03

(2.15)

(0.78)

(3.08)

(0.14)

(0.07)

(0.12)

(0.06)

Senior

gen.

233.08

1.14

5.57

0.10

0.14

0.03

0.14

0.03

(2.16)

(0.73)

(3.04)

(0.14)

(0.07)

(0.13)

(0.06)

Pre-voc.IV

/seniorgen.

183.01

1.24

6.03

0.11

0.17

0.03

0.16

0.03

(2.10)

(0.71)

(3.28)

(0.16)

(0.07)

(0.14)

(0.07)

Pre-voc.IV

232.85

1.14

6.03

0.16

0.18

0.03

0.17

0.03

(2.08)

(0.72)

(3.20)

(0.17)

(0.07)

(0.15)

(0.07)

Pre-voc.II

222.98

0.96

6.28

0.05

0.13

0.02

0.12

0.02

(2.29)

(0.59)

(3.40)

(0.15)

(0.06)

(0.12)

(0.06)

Pre-voc.I/II

212.85

1.00

5.82

0.06

0.12

0.02

0.11

0.02

(2.23)

(0.66)

(3.16)

(0.15)

(0.06)

(0.12)

(0.05)

Pre-voc.I

222.82

1.05

6.00

0.06

0.12

0.02

0.11

0.02

(2.12)

(0.69)

(3.15)

(0.13)

(0.07)

(0.11)

(0.06)

Note:

Colum

ns(3)an

d(4)repo

rtdistan

cesto

thene

arestscho

olwithinastud

ents’c

hoicesetan

dto

alls

choo

lswithinastud

ents’c

hoiceset.

14

such as exam results and school visits. Since the 1990’s, the inspectorate’s qualityassessment of schools is public information, published on a yearly basis. A nationalnewspaper called Trouwmakes the quantitative part of the information accessible to thebroader public, by publishing exam grades and other characteristics, and by computingan overall quality score measured on a 5-point scale ("- -", "-", "0", "+", "++").12

The inputs for the Trouw quality scores are: final exam grades in the school track,the percentage of students getting a degree without grade repetition in the higherclasses and the performance of the school in the lower years. In this last number,changes in school track play a role: students getting a lower track than their initialadvice reduce the score, students getting a higher track increase it. The overall qualityscore is calculated by Trouw for each school track separately, so within a school, theremay be different quality scores. Very small schools or schools that miss values on theunderlying scores do not get an overall quality score.

The quality information we use for each student is the quality information publishedin December before the student’s school choice in April. In 2009 - the last publicationrelevant in our data - the 5-point scale in Trouw was abolished and only the underlyingnumbers were published. Since the 5-point scale was computed in the same way everyyear, we are able to reconstruct the scores. In the analyses we will study the 2009quality scores separately.

Since only Amsterdam schools are included in this study, there are not enoughschools with the extreme quality scores (“- -” and “++”) in the sample. Therefore, wetransformed the five-point scale into a three-point scale. As can be seen in Figure 2.1,there is substantial variation in the quality scores of the Amsterdam schools13. Thequality scores are not very stable over time. Table 2.3 shows the variation of the qualityscores within schools in the four-year period of the study. For example, only 10 out ofthe 27 schools that offer the pre-university track keep the same quality score for fourconsecutive years.

2.4 Empirical strategy

2.4.1 Main analysis

To study the determinants of secondary school choice we use discrete choice models.Specifically we will estimate conditional and mixed logit models. These models followfrom a random utility framework in which a student is assumed to choose the school

12In 2001, a news magazine called "Elsevier" also started publishing school quality informationbased on the inspectorate information. We choose to use the Trouw data since they were the first topublish this information and since it has a wider circulation than Elsevier.

13Four schools (with seven overlapping school tracks) are registered formally as two separate schools.These two pairs of schools cooperate very closely, and only get one quality score for the pair. Therefore,we used this quality score for both schools.

15

Table 2.3: Changes in school quality scores over timeSchool N N stable Quality Quality scores in previous year

track Schools schools scores in Mis. Neg. Neut. Pos.

current

year

Pre-university

Pre-uni. 27 10 Missing 9 0 0 0

Senior gen./pre-uni. 21 6 Negative 4 11 2 0

Neutral 2 8 8 2

Positive 0 1 9 24

Total 15 20 19 26

Senior general secondary

Senior gen./pre-uni. 21 7 Missing 13 0 0 0

Senior gen. 23 9 Negative 1 0 3 1

Pre-voc. IV/senior gen. 18 7 Neutral 0 6 21 4

Positive 1 4 5 22

Total 15 10 29 27

Pre-vocational IV

Pre-voc. IV/senior gen. 18 4 Missing 14 0 1 1

Pre-voc. IV 23 5 Negative 0 3 6 2

Neutral 4 8 16 2

Positive 1 1 6 2

Total 19 12 29 7

Pre-vocational II

Pre-voc. II 22 5 Missing 3 0 3 1

Pre-voc. I/II 21 5 Negative 0 4 2 0

Neutral 2 5 17 8

Positive 2 1 5 13

Total 7 10 27 22

Pre-vocational I

Pre-voc. I/II 21 11 Missing 3 0 0 3

Pre-voc I 22 11 Negative 0 3 0 0

Neutral 1 1 6 4

Positive 1 0 7 36

Total 5 4 13 43Note: In this table, we include all relevant schools that have a quality score for a certain school track.We do not take into account the fact that some schools are not in the choice set for combined tracks(a student with combined senior general secondary/pre-university advice cannot go to a school thatonly offers pre-university education). The number of stable schools indicates the number of schoolsthat have the same quality score over the four included years.

16

Figure 2.1: Variation in Trouw Quality Scores

Note: This figure shows the occurrence of different Trouw quality scores published in December 2006to December 2009 for Amsterdam schools.

that maximizes her utility. The utility that student i choosing in year t derives fromsecondary school s is defined as Uits = Vits + εits, where Vits is the observed portion ofutility and εits is the unobserved portion of utility. εits is assumed to be independently,identically distributed extreme value, in which the independence means that the un-observed utility for one alternative is unrelated to the unobserved utility for anotheralternative (Train, 2009). We will use different specifications of Vits, thereby imposingdifferent assumptions on the model. In all cases, we investigate school choice separatelyfor each group of students with the same school track advice. The reason is that thestudents’ choice sets differ by advised school track.

First, we analyze whether school choice can be predicted by student and schoolcharacteristics, in which Vits = x′itsβ + w′

tsδ + z′itγs. Here xits are characteristics thatare specific for a student-school combination. In the main analysis this will only bethe home to school distance. In additional analyses it will also include the share andthe predicted share of a students’ primary school peers going to that secondary school.wts are secondary school specific factors: the school quality and whether the schoolhad a lottery for that school track in the previous year. zit are student specific factors,specifically gender, ethnicity, test score and student weight. We include indicators forstudents missing values of these characteristics. The probability (pits) that student iin year t chooses for secondary school s, is given by:

17

pits =exp(x′itsβ + w′

tsδ + z′itγs)∑ml exp(x

′itlβ + w′

tlδ + z′itγl)(2.1)

The parameters β, δ and γ are estimated using maximum likelihood estimation, inwhich the log likelihood function is given by LL(β, γs, δ) =

∑Ss=1

∑i yits ln pits, where

yits is a binary indicator equal to one if student i in year t chooses school s, and zerootherwise.

In specification 2.1, we do not take into account differences between secondaryschools, apart from school quality and having a lottery in the previous year. It islikely, however, that unobserved characteristics (in our database) of the secondaryschool, such as the quality of the building and offering extracurricular activities areimportant in secondary school choice. In specification 2.2, we therefore add secondaryschool fixed effects. dsj is a secondary school dummy, which has dsj = 1 if s = j anddsj = 0 if s 6= j. Adding secondary school dummies changes the interpretation of thequality scores: we no longer use the absolute scores of a secondary school, but studyhow school choices change when school quality scores change.

pits =exp(x′itsβ + w′

tsδ + z′itγs +∑J

j=2 d′sjαj)∑m

l exp(x′itlβ + w′

tlδ + z′itγl +∑J

j=2 d′ljαj)

(2.2)

Models 2.1 and 2.2 are conditional logit models, which assume that the errors εits areindependent of each other such that the unobserved portion of utility for one alternativeis unrelated to the unobserved portion of utility for another alternative. This lack ofcorrelation gives rise to the property of independence of irrelevant alternatives (IIA).IIA means that for any two alternatives j and k, the relative odds of choosing j overk are the same, such that the ratio is not dependent of the presence or attributesof other alternatives (Train, 2009). In practice, this means that two schools (of thesame track) will be equally affected by the opening of a new competing school. This isnot very likely: the school that is closest related in terms of distance or school policywill probably be affected more. Mixed logit models do not have the IIA property.Therefore, we have also estimated mixed logit models for secondary school choice. Ascan be seen in equation 2.3, we estimated random coefficients for the student-schoolspecific characteristics and the school specific characteristics (quality, lottery in theprevious year and distance). φ(β, δ|b, d,W ) indicates the mixing distribution, in thiscase the normal density with means b and d and covariance W .14

14Instead of a normal density, Hastings et al. (2009) assume a negative lognormal distributionfor distance, which imposes that all people dislike commuting. Without explicitly imposing thisrestriction, we consistently find that students have a negative preference for distance.

18

pits =

ˆ (exp(x′itsβ + w′

tsδ + z′itγs +∑J

j=2 d′sjαj)∑m

l exp(x′itlβ + w′

tlδ + z′itγl +∑J

j=2 d′ljαj)

)φ(β, δ|b, d,W ) dβ (2.3)

Another advantage of the mixed logit model is that it allows for variation in pref-erences for school characteristics. It might be that some students put a high weighton the published quality scores, while other do not. Or some very risk-averse studentsmay avoid a school that had a lottery in the previous year, while others do not care somuch. In the mixed logit model, we do not only estimate the means of the preferencesfor school characteristics, but also their standard deviations, which allows us to see towhat extent the preferences for secondary school characteristics vary across students.

2.4.2 Peer effects in school choices

The school choice literature typically assumes that students make their choices indi-vidually and ignores possible influences of peers. Casual observation suggests that thisis unrealistic. Students who go from a primary school to a secondary school seem to beinfluenced by the choices of their classmates in primary school. They may coordinatetheir school choices, or some students may follow the choices of others. There may, inother words, be peer effects in school choices.

Peer effects are intrinsically difficult to analyze (Manski, 1993; Angrist, 2014). Thefollowing identification problems are present. The first is the reflection or simultaneityproblem. In the presence of peer effects, it is not always clear how the causality runs -do peers affect the respondent, or does the respondent affect peers? The second is theself-selection problem. Group membership is endogenous - peers select themselves onsimilar characteristics. In the presence of peer effects, it is not always clear whethereffects are driven by behavior, or by unobserved characteristics that are correlated withit. And related, there is the errors-in-variables problem. It is not always clear how todefine a peer group. What constitutes a peer?

The economics of education literature has so far been unsuccessful in addressingall these problems in a satisfactory way. We follow Sacerdote (2001) and include thefractions of primary school classmates choosing for each secondary school in the logitmodel of students’ secondary school choice. The results are subject to the reflectionproblem and the correlated errors problem and can therefore not be given a causalinterpretation. The results are, however, informative about the degree of correlationin the choices of classmates in primary school.

To address the correlated errors problem, we split the the fractions of primary schoolclassmates choosing for each secondary school into a component that is common for allcohorts from the same primary school (where we allow for a trend) and a idiosyncratic

19

component where the choices of students’ primary school classmates deviate from thechoices of students from other cohorts in that primary school. If inclusion of thefirst component solves the correlated errors problem, the estimate of the effect of theidiosyncratic component in classmates’ choices is only biased due to the reflectionproblem. According to Sacerdote (2001), this allows us to test the null hypothesis ofno peer effects which predicts no relationship between a student’s own school choiceand the idiosyncratic school choices of her classmates in primary school.

2.5 Results

2.5.1 Main findings

Table 2.4 presents estimates from the conditional logit model without school fixedeffects by secondary school track. The top part of the table reports the estimates ofthe impact of school quality scores on students’ choices. For students with a mixedadvice (e.g. senior general secondary/pre-university) the quality scores of both tracksmatter and estimates of the impact of two quality scores are reported, where thefirst estimate pertains to the highest of the two tracks.15 Neutral quality scores arethe reference category. As described before, the quality scores of 2009 are reportedseparately, because only the underlying scores were published by Trouw in this year.The estimated coefficients for the quality scores provide little support for the hypothesisthat school choice depends on published quality scores. Two coefficients for a positivequality score are significantly positive and three coefficients for a negative quality scoreare significantly negative. At the same time two coefficients for a positive qualityscore are significantly negative and two coefficients for a negative quality score aresignificantly positive. Just focusing on the sign of the coefficients, it turns out that 14out of 22 coefficients have a sign that is consistent with the hypothesis that a positive(negative) quality score makes it more (less) likely that a school is chosen. We cannot reject that this number of correct signs is generated by chance (p=0.14). By andlarge we can therefore not reject the hypothesis that published quality score have nosystematic impact on students’ school choices.

Distance turns out to be a consistent and strong predictor of school choice. Thisis in accordance with results from other studies (e.g. Hastings et al., 2009; Koning &Van der Wiel, 2013). One way to interpret the size of the coefficients is in terms ofodds ratios. The logit coefficient for distance to pre-university schools in Table 2.4can be expressed as an odds ratio by taking the exponent: exp(−0.535) = 0.59. This

15Consider for example the estimates for a negative quality score for students with a mixed seniorgeneral secondary/pre-university advice. The estimate of -0.380 is the effect of a negative quality scoreof the pre-university track in a school and the estimate of 0.304 is the effect of a negative quality scoreof the senior general secondary track in a school.

20

Tab

le2.4:

Estim

ates

from

conditional

logitmod

elswithou

tschoo

lfixedeff

ects

Pre-uni.

Senior

gen./

Senior

gen.

Pre-voc.IV

/Pre-voc.IV

Pre-voc.II

Pre-voc.I/II

Pre-voc.I

pre-un

i.senior

gen.

Negativequ

ality

0.371***

-0.380***

0.106

0.001

-0.222***

0.080

-0.115

-0.355

(0.111)

(0.127)

(0.084)

(0.101)

(0.085)

(0.123)

(0.212)

(0.555)

0.304***

-0.307***

-1.037

(0.108)

(0.109)

(0.735)

Positivequ

ality

0.477***

0.474***

0.122

-0.055

-0.189*

0.034

0.171

0.069

(0.093)

(0.105)

(0.077)

(0.124)

(0.098)

(0.089)

(0.131)

(0.105)

0.046

-0.321**

0.190

(0.114)

(0.143)

(0.169)

Missing

qualityscores

-0.414***

-0.377**

-0.775***

-0.602

-0.260**

0.160

-0.382

0.058

(0.136)

(0.181)

(0.169)

(0.413)

(0.129)

(0.145)

(0.556)

(0.139)

-0.705**

0.363

0.380

(0.298)

(0.393)

(0.587)

Negativequ

ality2009

-0.011

0.696**

-0.058

0.108

0.354**

-0.605

-0.763

0.385

(0.229)

(0.268)

(0.383)

(0.459)

(0.169)

(0.482)

(0.578)

(0.404)

-0.503

0.384

(0.420)

(0.289)

Positivequ

ality2009

-0.293**

0.062

-0.274**

0.182

-0.393**

-0.317*

-0.657**

0.238

(0.119)

(0.188)

(0.107)

(0.159)

(0.159)

(0.170)

(0.259)

(0.218)

-0.100

0.181

0.039

(0.162)

(0.208)

(0.293)

Missing

qualityscores

2009

-0.118

0.549

-0.006

-0.222

-0.567***

-0.568**

-0.538

-0.375

(0.270)

(0.590)

(0.173)

(0.263)

(0.191)

(0.258)

(0.650)

(0.377)

-0.768

-0.257

(0.484)

(0.777)

(Tab

leco

ntin

ues

onne

xtpa

ge)

21

Tab

le2.4:

(con

tinu

ed)

Pre-uni.

Senior

gen./

Senior

gen.

Pre-voc.IV

/Pre-voc.IV

Pre-voc.II

Pre-voc.I/II

Pre-voc.I

pre-un

i.senior

gen.

Distance

-0.535***

-0.552***

-0.520***

-0.553***

-0.546***

-0.482***

-0.531***

-0.582***

(0.020)

(0.020)

(0.021)

(0.024)

(0.022)

(0.017)

(0.023)

(0.017)

Lottery

0.335***

-0.078

0.035

0.462***

0.633***

0.113

0.219

0.049

(0.067)

(0.064)

(0.067)

(0.137)

(0.134)

(0.338)

(0.334)

(0.183)

Nstud

ents

4374

2819

3133

2028

2881

2130

1039

2713

Nscho

ols

2721

2318

2322

2122

LogLikelih

ood

10375.3

-5991.1

-7068.0

-3703.4

-5446.1

-4365.7

-2059.1

-5400.9

Pseud

oR-squ

ared

0.15

0.24

0.24

0.31

0.32

0.29

0.31

0.33

Note:

Stan

dard

errors

areclusteredat

theprim

aryscho

ollevel.Includ

edstud

entlevelc

ontrolsaregend

er,D

utch,m

issing

ethn

icity,

test

score,

missing

test

score,

stud

entweigh

tan

dmissing

stud

entweigh

t.***p

<0.01

**p<

0.05

*p<0.10.

22

coefficient indicates that students with a pre-university advice are 41% less likely tochoose for a school that is 1 kilometer farther away, as compared to an identical school1 kilometer closer. The coefficients for the lottery dummy are positive and significantin three out of eight coefficients, but they have the “wrong” sign in the sense thatthey suggest that students are more likely to choose a school that had a lottery inthe previous year. This result is robust for different specifications of the lottery andfor using different samples: we get similar results when (1) we use the percentage ofstudents losing the lottery in the previous year instead of the binary indicator, (2) weaccount for the school making an extra class in the year of school choice and when(3) we drop students who have priority for admittance to their school of first choice.Results are reported in Table 2.ii in Appendix 2.A.2.

Table 2.5 presents estimates from the conditional logit model including secondaryschool fixed effects. This alters the interpretation of the quality scores: we no longeruse the absolute scores of a secondary school, but study how school choices respond tochanges in school quality scores. The estimation results are very similar to those in theprevious table without school fixed effects. On the basis of the estimated coefficients forthe school quality scores we can again not reject the hypothesis that quality scores haveno systematic impact on students’ school choices. Also the estimated effects for distanceare very similar to those in the previous table. The inclusion of school fixed effectsdoes, however, change the estimates of the coefficients of the lottery indicator. Nowfive out of eight coefficients have the expected negative sign, with two coefficients beingsignificant at the 10% level and one coefficient being significant at the 5% level. Thereis thus some evidence that students avoid schools that changed from not conducting alottery to conducting a lottery in the previous year.

Table 2.6 presents estimation results of mixed logit models. In addition to themean of the preference weights for school characteristics the mixed logit model alsoproduces estimates of their standard deviation. This allows us to see to what extentthe preferences for secondary school characteristics vary across students. Again, theestimates indicate that students have a lower preference for schools that are fartherfrom their home address. At the same time, however, we see that the aversion fordistance varies significantly across students. Regarding school quality, we again seeinconsistent coefficients over the different school tracks; we can still not reject thehypothesis that quality scores have no systematic impact on school choice. Some ofthe standard deviations for the quality scores come out significantly, indicating thatthe preferences for certain quality scores vary across students. Regarding the lotteryindicator, it can be seen that only the mean coefficient for pre-vocational II comes outsignificantly. For the other school tracks, the coefficients are mainly negative but notsignificantly different from zero.

23

Tab

le2.5:

Estim

ates

from

conditional

logitmod

elswithschoo

lfixedeff

ects

Pre-uni.

Senior

gen./

Senior

gen.

Pre-voc.IV

/Pre-voc.IV

Pre-voc.II

Pre-voc.I/II

Pre-voc.I

pre-un

i.senior

gen.

Negativequ

ality

0.288**

-0.097

0.071

0.022

-0.153*

0.195

0.082

1.698***

(0.123)

(0.146)

(0.088)

(0.101)

(0.091)

(0.121)

(0.212)

(0.646)

0.174*

-0.079

-0.651

(0.105)

(0.116)

(0.976)

Positivequ

ality

0.397***

0.284***

0.006

0.014

-0.174*

-0.082

0.014

0.052

(0.102)

(0.108)

(0.076)

(0.129)

(0.102)

(0.094)

(0.141)

(0.111)

0.065

-0.321**

0.167

(0.114)

(0.153)

(0.195)

Missing

qualityscores

-0.252

-0.092

-0.069

-0.151

0.022

-0.028

-0.367

0.027

(0.156)

(0.189)

(0.236)

(0.433)

(0.163)

(0.161)

(0.555)

(0.145)

-0.225

0.168

0.109

(0.332)

(0.409)

(0.612)

Negativequ

ality2009

0.321

0.236

0.634

-0.265

0.578***

0.165

-0.861

0.419

(0.248)

(0.280)

(0.466)

(0.493)

(0.189)

(0.520)

(0.598)

(0.408)

0.439

0.523*

(0.492)

(0.308)

Positivequ

ality2009

-0.350***

-0.032

-0.210*

0.100

-0.528***

-0.296*

-0.360

0.217

(0.125)

(0.191)

(0.113)

(0.171)

(0.172)

(0.163)

(0.283)

(0.220)

-0.069

-0.119

-0.215

(0.166)

(0.248)

(0.338)

Missing

qualityscores

2009

0.442

0.349

0.132

-0.334

-0.563***

-0.329

-0.085

-0.438

(0.378)

(0.630)

(0.182)

(0.274)

(0.196)

(0.264)

(0.676)

(0.387)

-0.419

-0.874

(0.473)

(0.832)

(Tab

leco

ntin

ues

onne

xtpa

ge)

24

Tab

le2.5:

(con

tinu

ed)

Pre-uni.

Senior

gen./

Senior

gen.

Pre-voc.IV

/Pre-voc.IV

Pre-voc.II

Pre-voc.I/II

Pre-voc.I

pre-un

i.senior

gen.

Distance

-0.588***

-0.609***

-0.555***

-0.576***

-0.557***

-0.495***

-0.552***

-0.588***

(0.023)

(0.023)

(0.021)

(0.024)

(0.022)

(0.018)

(0.024)

(0.017)

Lottery

0.066

-0.103

-0.113*

0.003

-0.291*

-0.678**

-0.151

0.026

(0.064)

(0.069)

(0.065)

(0.131)

(0.175)

(0.287)

(0.320)

(0.187)

Nstud

ents

4374

2819

3133

2028

2881

2130

1039

2713

Nscho

ols

2721

2318

2322

2122

LogLikelih

ood

-10112.1

-5891.0

-6916.4

-3557.8

-5272.8

-4240.6

-2003.0

-5344.0

Pseud

oR-squ

ared

0.17

0.25

0.25

0.34

0.34

0.31

0.33

0.33

Note:

Cluster

robu

ststan

dard

errors

arein

parentheses,

stan

dard

errors

areclusteredat

theprim

aryscho

ollevel.

Includ

edstud

entlevelc

ontrolsaregend

er,

Dutch,m

issing

ethn

icity,

test

score,

missing

test

score,

stud

entweigh

tan

dmissing

stud

entweigh

t.***p

<0.01

**p<

0.05

*p<0.10.

25

Table 2.6: Estimates from mixed logit modelsPre-uni. Senior gen./ Senior gen.

pre-uni.

Mean SD Mean SD Mean SD

Negative quality -0.105 1.445*** -0.196 0.603 -0.006 0.660

(0.184) (0.301) (0.193) (0.450) (0.225) (0.753)

-0.003 1.025**

(0.241) (0.517)

Positive quality 0.458*** 0.889** 0.299*** 0.854** 0.026 0.418

(0.118) (0.348) (0.116) (0.352) (0.081) (0.328)

0.090 0.769**

(0.134) (0.343)

Missing quality scores -1.122* 1.851** -1.275** 2.486*** -2.431** 3.102***

(0.617) (0.726) (0.513) (0.523) (0.976) (0.777)

-5.073* 5.077**

(2.635) (1.970)

Negative quality 2009 0.314 0.553 0.023 1.656*** 0.139 1.973***

(0.349) (0.480) (0.395) (0.517) (0.747) (0.575)

0.453 0.898

(0.648) (0.554)

Positive quality 2009 -0.192 1.164** 0.087 0.371 -0.218* 0.394

(0.199) (0.585) (0.249) (0.933) (0.121) (0.365)

-0.076 0.495

(0.195) (0.621)

Missing quality -1.695 2.599** 1.739* 0.500 0.484 0.313

scores 2009 (1.170) (1.053) (0.925) (0.599) (0.331) (0.899)

-4.547** 4.680***

(1.995) (1.679)

Distance -0.660*** 0.163** -0.738*** 0.174*** -0.649*** 0.251***

(0.031) (0.071) (0.035) (0.059) (0.031) (0.042)

Lottery 0.028 0.450 -0.115 0.287 -0.107 0.035

(0.086) (0.399) (0.073) (0.246) (0.068) (0.126)

N students 4374 2819 3133

N schools 27 21 23

Log Likelihood 10197.6 -5936.0 -7007.5

Pseudo R-squared 0.16 0.25 0.24(Table continues on next page)

26

Table 2.6: (continued)Pre-voc. IV/ Pre-voc. IV Pre-voc. II

senior gen.

Mean SD Mean SD Mean SD

Negative quality 0.037 0.027 -0.176 0.387 0.181 0.086

(0.109) (0.176) (0.109) (0.307) (0.126) (0.242)

-0.128 0.020

(0.127) (0.200)

Positive quality -0.095 0.899** -0.184* 0.004 -0.272* 1.002***

(0.178) (0.413) (0.106) (0.113) (0.144) (0.351)

-0.351** 0.156

(0.163) (0.182)

Missing quality scores -0.133 0.550 -0.122 0.786 0.005 0.266

(0.473) (0.443) (0.277) (0.496) (0.180) (0.195)

-0.356 1.520***

(0.565) (0.552)

Negative quality 2009 -0.628 1.253*** 0.214 1.645** -2.138 2.512

(0.585) (0.404) (0.434) (0.694) (4.150) (2.481)

0.643* 0.215

(0.358) (0.352)

Positive quality 2009 0.134 0.464 -0.718* 1.075 -0.706 1.808

(0.247) (1.359) (0.379) (0.928) (0.520) (1.247)

-0.165 0.893

(0.303) (0.684)

Missing quality -0.873 1.364 -1.077 1.259 -0.497 0.349

scores 2009 (0.828) (1.222) (0.738) (1.168) (0.318) (0.539)

Distance -0.671*** 0.232*** -0.634*** 0.192*** -0.558*** 0.152***

(0.036) (0.032) (0.025) (0.033) (0.026) (0.035)

Lottery -0.021 0.282 -0.393 0.968 -1.804*** 1.958***

(0.146) (0.289) (0.285) (0.589) (0.666) (0.522)

N students 2028 2881 2130

N schools 18 23 22

Log Likelihood -3649.4 -5358.8 -4227.0

Pseudo R-squared 0.32 0.33 0.31(Table continues on next page)

27

Table 2.6: (continued)Pre-voc. I/II Pre-voc. I

Mean SD Mean SD

Negative quality 0.011 0.690 -0.272 2.955***

(0.336) (0.945) (1.387) (0.615)

-3.989 3.570

(4.330) (3.042)

Positive quality -0.089 0.777 0.050 0.078

(0.218) (0.690) (0.113) (0.241)

0.160 0.334**

(0.206) (0.142)

Missing quality scores -0.364 0.232 0.012 0.087

(0.612) (0.611) (0.155) (0.506)

-0.506 1.410**

(0.826) (0.580)

Negative quality 2009 -3.000 2.810* 0.244 1.333**

(2.028) (1.674) (0.664) (0.676)

Positive quality 2009 -0.575 1.138* 0.212 2.017***

(0.431) (0.658) (0.234) (0.712)

-0.232 0.240

(0.380) (0.319)

Missing quality -0.593 0.599 -1.373* 0.158

scores 2009 (0.754) (0.414) (0.778) (0.487)

-0.862 0.995

(1.000) (0.737)

Distance -0.653*** 0.178*** -0.615*** 0.054

(0.044) (0.055) (0.018) (0.035)

Lottery -0.852 1.946*** -0.017 0.258

(0.690) (0.668) (0.202) (0.677)

N students 1039 2713

N schools 21 22

Log Likelihood -1984.8 -5324.8

Pseudo R-squared 0.34 0.33Note: Cluster robust standard errors are in parentheses, standard errors are clustered at the primaryschool level. Included student level controls are gender, Dutch, missing ethnicity, test score, missingtest score, student weight and missing student weight for pre-vocational I, I/II and II. For the pre-vocational IV to pre-university tracks, the missing indicators could not be included for computationalreasons. When using Wald tests to compare the coefficients for the variables in this table in conditionallogit models with and without the missing indicators, it turns out that the null hypothesis of equalcoefficients cannot be rejected for the pre-vocational IV to pre-university tracks. ***p<0.01 **p<0.05*p<0.10.

28

2.5.2 Peer effects

Figure 2.2 depicts for one single primary school from year-to-year how the studentsspread out across the available secondary schools. It shows that some secondary schools(such as numbers 7, 12, 32 and 57) are in general more popular than other schools forstudents in this primary school. But we also see some strong fluctuations. For exampleschool 58 is not a popular destination in 2006/07 and 2007/08, is the most populardestination in 2008/09 and is back to its original level again in 2009/10. The same istrue for school 86. Similar graphs for other primary school reveal comparable patterns:students who leave the same primary school in the same year tend to choose the samesecondary schools. To some extent this can be attributed to a general popularity of aspecific secondary school among the students from certain primary schools. But thereis also idiosyncratic clustering at – apparently random – destination schools.

Table 2.7 reports results from conditional and mixed logit models of school choicethat also include the shares of classmates and of predicted classmates that choose eachsecondary school as variables in xits. The table only reports the coefficients of thesetwo variables. Consistent with the pattern we observed in Figure 2.2, all coefficients aresignificantly positive. Not surprisingly, we see a positive and significant coefficient forthe predicted share of peers in the primary school. This coefficient captures the generalpopularity of secondary schools in the primary school. On top of the predicted shareof peers, however, we see that the actual share of peers has a positive and significantcoefficient as well. This indicates that when a higher share of the primary school peerschoose for a certain secondary school, the student himself is more likely to pick thisschool as well. We find this result for all specifications and also when we correct forsecondary school times year fixed effects, which makes it unlikely that this result isdriven by sudden changes in the attractiveness of secondary schools. The school timesyear fixed effects results are reported in Appendix 2.A.1.

Due to the reflection problem and correlated unobservables these coefficients cannotbe interpreted as causal effects of the school choices of classmates. The coefficients donot inform us about the social multiplier. Does the choice of one student influence allothers, or is everyone affected by everyone? The coefficients also do not exclude thatthe correlated school choices are caused by a third unobserved factor, such as activerecruitment by a specific secondary school in certain primary schools in a particularyear. The results in Table 2.7 are therefore no proof of the importance of peer effects,they are, however, consistent with it.

To get an idea on the magnitude of the peer coefficients, we can compute will-ingness to travel coefficients if we interpret the peer coefficients as the actual prefer-ence to attend the same secondary school as primary school classmates do: Wtt =

−βpeers/βdistance.

29

Figure 2.2: Secondary school choice in one primary school

Note: Frequency in the number of students choosing each school is depicted on the vertical axis, thehorizontal axis are secondary schools.

30

Tab

le2.7:

Estim

ates

ofpeercoeffi

cients;variou

slogitmod

els

Pre-uni.

Senior

gen./

Senior

gen.

Pre-voc.IV

/Pre-voc.IV

Pre-voc.II

Pre-voc.I/II

Pre-voc.I

pre-un

i.senior

gen.

Con

dition

alLo

git

witho

utscho

olFE

Predicted

share

3.706***

3.063***

4.643***

3.062***

3.625***

2.025***

5.573***

3.642***

ofpe

ers

(0.586)

(0.684)

(0.595)

(0.653)

(0.625)

(0.723)

(1.192)

(0.642)

Shareof

peers

3.464***

4.259***

3.013***

5.127***

3.669***

4.554***

2.743***

3.131***

(0.485)

(0.539)

(0.465)

(0.566)

(0.509)

(0.663)

(0.988)

(0.535)

Con

dition

alLo

git

withscho

olFE

Predicted

share

4.135***

3.537***

4.797***

2.659***

3.382***

1.618**

5.108***

3.530***

ofpe

ers

(0.564)

(0.681)

(0.593)

(0.640)

(0.621)

(0.707)

(1.233)

(0.635)

Shareof

peers

3.336***

4.282***

2.916***

5.131***

3.635***

4.506***

2.741***

3.135***

(0.477)

(0.530)

(0.460)

(0.568)

(0.500)

(0.650)

(1.000)

(0.535)

Mixed

Logit

Predicted

share

4.513***

4.022***

5.957***

4.417***

4.970***

2.786***

7.191***

3.918***

ofpe

ers

(0.644)

(0.785)

(0.699)

(0.862)

(0.760)

(0.838)

(1.705)

(0.783)

Shareof

peers

3.720***

5.721***

3.586***

6.117***

4.318***

5.654***

4.529***

3.808***

(0.564)

(0.649)

(0.564)

(0.742)

(0.594)

(0.831)

(1.305)

(0.718)

SD(predicted

1.526

-0.012

2.897

-5.846***

2.063

3.545***

1.033

-0.820

shareof

peers

(1.699)

(3.647)

(2.954)

(0.965)

(2.792)

(1.122)

(5.083)

(2.728)

SD(sha

reof

peers)

-1.489

3.965***

-3.399

-1.657

-3.886**

4.977***

7.189***

-3.925***

(1.174)

(0.750)

(2.294)

(2.134)

(1.717)

(0.779)

(1.730)

(0.810)

Nstud

ents

4374

2819

3133

2028

2881

2130

1039

2713

Nscho

ols

2721

2318

2322

2122

Note:

Stan

dard

errors

areclusteredat

theprim

aryscho

ollevel.

Allmod

elsalso

includ

eindicators

forscho

olqu

alityan

dlottery,

anddistan

ce.Includ

edstud

ent

levelcontrols

aregend

er,Dutch,missing

ethn

icity,

test

score,

missing

test

score,

stud

entweigh

tan

dmissing

stud

entweigh

t.Fo

rthepre-vo

cation

alIV

topre-un

iversity

tracks,the

missing

indicators

couldno

tbe

includ

edin

themixed

logitspecification

forcompu

tation

alreason

s.W

henusingWaldteststo

compa

rethecoeffi

cients

forthepe

ercharacteristicsan

dtheindicators

forqu

ality,

lotteryan

ddistan

cein

cond

itiona

llogitmod

elswithan

dwitho

utthemissing

indicators,

itturnsou

tthat

thenu

llhy

pothesis

ofequa

lcoefficients

cann

otbe

rejected

fortheselevels.***p

<0.01

**p<

0.05

*p<0.10.

31

For pre-university schools, we get a willingness to travel of −3.464/−0.371 = 9.334,indicating that students are willing to travel 9.3 kilometers for a 100% increase inthe percentage of peers, or 943 meters for a 10% increase in the percentage of peers.Given that the average class size in primary schools is about 25, a 10% increase in thepercentage of peers means an increase of 2 or 3 classmates. For the other school tracks,the willingness to travel for a 10% increase in the percentage of peers varies from 674to 1429 meters.

2.6 Conclusions

We investigated which secondary schools students choose when school choice is free.It turns out that in the city of Amsterdam, publicly available information on schoolquality does not clearly predict school choice. Students do not avoid schools with anegative quality score, or schools that change from a neutral or positive score to anegative score. Neither are they more likely to apply to schools with a positive qualityscore, or to schools that change from a neutral or negative score to a positive score.Students’ choices are consistent with a preference for schools close to their home, andfor schools to which a larger share of their classmates in primary schools go to. Thereis some indication that the current application and admission procedure which followsthe Boston system, triggers students to avoid schools that had an admission lottery inthe previous year.

The results of our analysis have implications for the debate on school choice andcompetition. If the choices of students who can freely choose from a large set of schoolsare not affected by information about school quality, one important channel throughwhich school choice can improve school quality does not operate.

32

2.A Appendix

2.A.1 School*year fixed effects

The school fixed effects models reported in Table 2.7 only capture time-invariant dif-ferences between secondary schools. However, changes in quality scores and having alottery in the previous year will not be the only changes relevant for secondary schoolchoice. Secondary schools may suddenly get more attractive because of more effort inattracting students or because of good publicity in the press. To see whether these un-observed yearly differences within secondary schools drive the positive peer estimates,Table 2.i presents peer estimates including secondary school*year fixed effects. Sincethe secondary school*year fixed effects capture all time varying and time constant dif-ferences between secondary school, indicators for school quality and having a lotteryin the previous year are dropped, leading to specification 2.4. As can be seen in Table2.i, the peer estimates remain similar to the results without school*year fixed effects.

pits =exp(x′itsβ + z′itγs +

∑Jj=2

∑Tt=1 d

′sjtαjt)∑m

l exp(x′itlβ + z′itγl +

∑Jj=2

∑Tt=1 d

′ljtαjt)

(2.4)

33

Tab

le2.i:

Estim

ates

ofpeercoeffi

cients

withschoo

l*year

fixedeff

ects

Pre-uni.

Senior

gen./

Senior

gen.

Pre-voc.IV

/Pre-voc.IV

Pre-voc.II

Pre-voc.I/II

Pre-voc.I

pre-un

i.senior

gen.

Con

dition

alLo

git

scho

ol*yearFE

Predicted

share

4.208***

3.586***

4.828***

2.847***

3.380***

1.915***

5.720***

3.644***

ofpe

ers

(0.555)

(0.713)

(0.607)

(0.675)

(0.645)

(0.701)

(1.234)

(0.629)

Shareof

peers

3.211***

4.194***

2.865***

4.921***

3.603***

4.261***

2.203**

2.964***

(0.480)

(0.556)

(0.450)

(0.592)

(0.513)

(0.651)

(1.019)

(0.537)

Mixed

Logit

scho

ol*yearFE

Predicted

share

4.157***

3.603***

5.402***

3.835***

3.971***

2.446***

6.707***

3.532***

ofpe

ers

(0.572)

(0.747)

(0.662)

(0.804)

(0.656)

(0.764)

(1.312)

(0.636)

Shareof

peers

3.433***

4.747***

3.242***

5.729***

3.878***

4.992***

2.952***

3.280***

(0.508)

(0.586)

(0.517)

(0.691)

(0.517)

(0.709)

(1.057)

(0.571)

SD(predicted

-2.158**

-2.190**

-2.292

4.150***

2.909*

1.730

4.520***

0.370

shareof

peers

(0.861)

(0.909)

(1.432)

(1.322)

(1.605)

(1.657)

(1.448)

(0.355)

SD(sha

reof

peers)

1.623**

3.346***

3.968***

4.290***

2.956**

5.249***

4.671***

3.047***

(0.724)

(0.557)

(0.703)

(1.270)

(1.449)

(0.762)

(1.447)

(0.734)

Nstud

ents

4374

2819

3133

2028

2881

2130

1039

2713

Nscho

ols

2721

2318

2322

2122

Note:

Stan

dard

errors

areclusteredat

theprim

aryscho

ollevel.

Allmod

elsalso

includ

edistan

ce.Includ

edstud

entlevelcontrols

aregend

er,Dutch,missing

ethn

icity,

test

score,

missing

test

score,

stud

entweigh

tan

dmissing

stud

entweigh

tforthecond

itiona

llogitmod

elsan

dthesamevariab

leswitho

utthemissing

indicators

forthemixed

logitmod

els.

***p

<0.01

**p<

0.05

*p<0.10.

34

2.A.2 Additional table

Table 2.ii: Alternative specifications of secondary school lotteries in theprevious year

Pre-uni. Senior gen./ Senior gen. Pre-voc. IV/

pre-uni. senior gen.

Conditional Logit

without school FE

(0) Lottery 0.335*** -0.078 0.035 0.462***

(0.067) (0.064) (0.067) (0.137)

(1) Percentage of students 0.011*** -0.001 0.004 0.006

losing the lottery (0.003) (0.003) (0.004) (0.004)

(2) Lottery controlling for 0.304*** -0.083 0.041 0.441***

extra classes (0.068) (0.065) (0.067) (0.139)

(3) Lottery without 0.263*** -0.180** -0.025 0.283*

priority students (0.075) (0.077) (0.078) (0.147)

Conditional Logit

with school FE

(0) Lottery 0.066 -0.103 -0.113* 0.003

(0.064) (0.069) (0.065) (0.131)

(1) Percentage of students 0.004 -0.006* -0.004 -0.003

losing the lottery (0.003) (0.003) (0.004) (0.004)

(2) Lottery controlling for 0.059 -0.113* -0.116* -0.075

extra classes (0.065) (0.068) (0.065) (0.135)

(3) Lottery without -0.048 -0.188** -0.146* -0.013

priority students (0.075) (0.080) (0.076) (0.151)

N students (0-2) 4374 2819 3133 2028

N priority students (3) 3462 2282 2577 1785(Table continues on next page)

35

Table 2.ii: (continued)Pre-voc. IV Pre-voc. II Pre-voc. I/II Pre-voc. I

Conditional Logit

without school FE

(0) Lottery 0.633*** 0.113 0.219 0.049

(0.134) (0.338) (0.334) (0.183)

(1) Percentage of students 0.019*** 0.008 0.005 0.001

losing the lottery (0.004) (0.007) (0.006) (0.005)

(2) Lottery controlling for 0.640*** 0.140 0.168 0.039

extra classes (0.136) (0.343) (0.340) (0.183)

(3) Lottery without 0.578*** 0.100 0.217 0.049

priority students (0.131) (0.340) (0.334) (0.183)

Conditional Logit

with school FE

(0) Lottery -0.291* -0.678** -0.151 0.026

(0.175) (0.287) (0.320) (0.187)

(1) Percentage of students -0.007 -0.012* -0.002 -0.001

losing the lottery (0.005) (0.006) (0.006) (0.006)

(2) Lottery controlling for -0.326* -0.671** -0.204 0.015

extra classes (0.177) (0.288) (0.320) (0.186)

(3) Lottery without -0.278 -0.691** -0.164 0.027

priority students (0.199) (0.288) (0.319) (0.187)

N students (0-2) 2881 2130 1039 2713

N priority students (3) 2612 2099 1025 2668Note: Each cell reports the coefficient of a lottery variable from a separate conditional logit model.All models also include indicators for school quality and distance. Included student level controls aregender, Dutch, missing ethnicity, test score, missing test score, student weight and missing studentweight. Specifications (0) repeat the lottery coefficients from Tables 2.4 and 2.5. Specifications (1)include the percentage of students losing the lottery in the previous year instead of the binary indicator.Specifications (2) use the binary indicator and account for the school making an extra class in theyear of school choice. Specifications (3) drop students who have priority for their school of first choice.Standard errors are clustered at the primary school level. ***p<0.01 **p<0.05 *p<0.10.

36

Chapter 3

The impact of losing a schooladmission lottery on school outcomes1

3.1 Introduction

In systems of free school choice, it is common that some schools receive more applica-tions than they can enroll. An often used way to deal with this, is to conduct schooladmission lotteries. Students who lose a lottery cannot attend the school to which theyinitially applied. In this chapter, we use data from admission lotteries for secondaryschools from the years 2006 to 2010 in the Dutch city of Amsterdam, to investigate theeffects of losing a lottery on students’ academic outcomes.

The impact of losing a school admission lottery, instead of winning one, is interestingfor at least four different reasons.2 First, one of the main arguments in favor of schoolchoice is that parents are the best suited to find a school that matches the specificneeds of their child, which would improve student outcomes. Despite the popularity ofthe argument, the claim is hard to study because school choice is endogenous to parentand child characteristics. Since school admission lotteries create random variation inattending the school of first choice, they provide an unique opportunity of studyingthe effects of school choice on students’ academic outcomes. If parents indeed choosethe school that best fits the specific needs of their child, we expect that being withheldfrom the first choice, has a negative effect on outcomes.

The second reason to be interested in the effect of losing a school admission lotteryis related to the first. When the oversubscribed schools are of better quality than theiralternatives, school lotteries can shed light on the relevance of school quality. In thecurrent economic literature, the evidence on the effects of attending a better school, ismixed. On the one hand, Hastings & Weinstein (2008), Hoekstra (2009), Kirabo Jack-

1This chapter is based on joint work with Hessel Oosterbeek and Inge de Wolf.2To avoid confusion, we contrast losing an admission lottery with winning one, not with placement

in a school which was not oversubscribed.

37

son (2010), Pop-Eleches & Urquiola (2013) and Saavedra (2009) find positive impactsof getting into a better school or college on outcomes such as academic achievementand earnings. In contrast, Abdulkadiroğlu et al. (2014), Clark (2010), Cullen et al.(2006), Dobbie & Fryer (2011) and Duflo et al. (2011) find little evidence that gettinginto a higher achievement school or class within a school brings significant benefits.

Third, school admission lotteries can provide a clean measure of school performance.When school populations differ in unobserved characteristics, school performance mea-sures will be distorted by the differences in populations, making it hard to comparethe performance of different schools. Since school lotteries create random variation inschool attendance, they can provide a clean measure of school performance by creatingtwo identical groups that do and do not attend a certain school. This way, schoollotteries can be used to validate school quality measures (Deming, 2014).

Finally, the school lotteries in Amsterdam are particularly interesting in relation tothe debate about school assignment mechanisms. Both in the public debate and in theeconomic literature, there is discussion on how school assignment mechanisms shouldbe organized (e.g. Abdulkadiroğlu et al., 2005, 2009; Ergin & Sönmez, 2006; Kesten,2010). School assignment in Amsterdam is based on a mechanism that is very similar tothe so-called Boston mechanism. This mechanism gets a lot of criticism, since studentswho lose the school admission lottery in the first round can in the second round onlychoose from schools that have seats available after the first round. This may meanthat the school of their second, and perhaps third and fourth, choice are no longeravailable. This implies that losers of admission lotteries in Amsterdam are potentiallyworse off than they would be when another school assignment mechanism would bein use. Estimating the effect of losing an admission lottery in Amsterdam, thereforeassesses how much losers are actually harmed under the Boston mechanism.

Most related to our study is the paper of Cullen et al. (2006), who estimate the effectof the outcome of admission lotteries in the Chicago public school system. While verysimilar in design, there are two important differences between the contexts that Cullenet al. (2006) and the current chapter investigate. The public school system in Chicagoserves the lower end of the education market, since there is also a sizable private schoolsystem. In contrast, many of the admission lotteries in Amsterdam take place at theupper end of the education market, because especially some elite schools are in highdemand. Related to this is the second difference, namely that a vast majority of thestudents that win an admission lottery in the Chicago public school system decide toattend another school than the school for which they won the lottery. In contrast,almost all students winning an admission lottery in Amsterdam attend the school forwhich they won the lottery. This means that where Cullen et al. (2006) estimate a localaverage treatment effect applicable to a group of probably rather special compliers, thecurrent study estimates an average treatment effect.

38

The main results of our analysis are threefold. First, students who lose an admis-sion lottery attend a secondary school with ’worse’ peers, in terms of previous academicachievement and social background, than students who win an admission lottery. Sec-ond, in terms of school quality measures of the Dutch education inspectorate, lotterylosers attend schools that perform better on grade progression in the lower grades butworse on grade progression in the higher grades and on exam scores. Finally, in thefourth year after the admission lotteries, students who lost the lottery are more likelyto be in the fourth year of their advised academic level than lottery winners. Lotterywinners are thus more likely to have repeated a year or to have been directed to alower academic level. This last result may reflect a difference in policies between theschools attended by losers and winners. Alternatively, lottery losers end up with worseperforming peers than lottery winners and they may benefit from their higher rankposition.

A restriction of the current study is that for most students in the sample, we onlyhave outcome data up to four years after the admission lottery in which they partici-pated took place. Most students are too young to have finished secondary school.3

This chapter is structured as follows. Section 3.2 provides background informationon Dutch secondary education and describes the secondary school choice and admissionlottery system in Amsterdam. Section 3.3 describes the data and how we assessedwhether school admission lotteries were conducted in a fair way. Further, this sectioncompares secondary schools which do and do not conduct school admission lotteries.Section 3.4 explains the empirical strategy and shows the results from lottery balancingtests, while Section 3.5 characterizes the treatment. It describes what losing a schooladmission lottery implies in terms of peer characteristics and secondary school quality.Finally, Section 3.6 presents the results and Section 3.7 concludes.

3.2 The Dutch context

3.2.1 Secondary education in the Netherlands

Dutch students and their parents have free school choice, they are not restricted bycatchment areas or school fees. Moreover, privately funded schools are virtually non-existent: nearly all primary and secondary schools are completely publicly funded.This includes religious schools and schools with specific educational philosophies, suchas Montessori or Dalton. The government funding is nationally determined and largely

3Consequentially, the estimates of the impact of losing an admission lottery on exam results arerather imprecise. It also means that we cannot look at the impact of losing an admission lottery fora secondary school on subsequent education choices such as enrollment in higher education and thechoice of an higher education study. This restriction will be lifted when time passes and we can addinformation from new years to our data.

39

Figure 3.1: The Dutch secondary school system

dependent on student numbers. Schools can get additional funding for students fromdisadvantaged backgrounds.4 The quality of education is guarded by the Dutch edu-cational inspectorate. Since the 1990’s, the inspectorate’s quality measures are publicinformation, which can be found on their website and in newspaper rankings.

Dutch secondary education starts at age 12, and lasts four to six years. The sec-ondary school system is highly tracked, Figure 3.1 gives a graphic description. Thelowest tracks (pre-vocational education, vmbo) last four years, and give access to vo-cational education programs. Within the pre-vocational track, there are four differentlevels, each leading to different levels of vocational education. For clarity reasons, theywill be indicated with the numbers I to IV, with IV being the highest level.5 The in-termediate track (senior general secondary education, havo), takes five years and givesaccess to universities of applied sciences, also called higher professional education. Thehighest track (pre-university education, vwo) takes six years, and gives access to uni-versity education.

The initial school track is based on the advice of the primary school teacher andon a standardized high stakes test (in most cases a nationwide exit test called the“citotoets”) at the end of primary school. Not all secondary schools offer all schooltracks and schools can educate children of different tracks together. Dependent onstudent achievement and school policies, students can change track during secondaryeducation. Further, students can follow a higher track after finishing a lower track.6

4Student level eligibilities for such arrangements are used as control variables. Specifically, wecontrol for receiving ’weighted student funding’ in primary education and living in a disadvantagedneighborhood in secondary education.

5From high to low, their actual names are “theoretische leerweg (tl or mavo)”, “gemengde leerweg(gl)”, “kaderberoepsgerichte leerweg (kbl)” and “basisberoepsgerichte leerweg (bbl)”.

6In that case, students enter the higher track in the year before the final exams. When doingpre-university education after senior general secondary education, for example, students enter in year

40

In the second half of secondary education, students can choose their own courses,subject to some conditions. Secondary schools have to follow national curriculumguidelines, and all students take centrally determined national exams at the end ofsecondary school. The national exams count for 50% of the final grade, the other 50%is determined by school specific exams taken in the last two or three years of secondaryeducation.

3.2.2 Secondary school choice and admission lotteries in Ams-

terdam

Amsterdam is the capital and largest city of the Netherlands, it has about 750,000inhabitants. Every year, 5500 to 6000 students transfer from primary to secondaryeducation. Amsterdam has 54 secondary schools, excluding schools for students withspecial educational needs. Some schools are more popular than others: every year, anumber of schools are oversubscribed and conduct lotteries to allocate the availableplaces.

In 2005, Amsterdam introduced a centralized enrollment and placement system.This system is basically a version of the so-called Boston mechanism. In the applicationprocess for a secondary school, students can only apply for one school. When theschool of their choice is oversubscribed, a lottery is conducted at the school itself.It can happen that a school is over-subscribed for some school tracks, but not forothers. Within schools, lotteries are usually conducted for each school track separately.Students can only subscribe at a school that offers the school track of their primaryschool teacher’s advice; the standardized test at the end of primary school serves as acheck on that advice. Students can only be rejected when their test score is too lowfor their primary school advice, which happens in about 5% of the cases. Schools areallowed to use a limited number of priority rules if they announce the rules before theapplication date. Specifically, they can grant priority to siblings of current students,children of staff members and students from a primary school with a similar educationalphilosophy (for example, Montessori schools).

When a student loses the admission lottery for the school of his first choice, he hasto apply to a school that still has places left after the first round. Next to the lotteryconducting schools, some schools fill up exactly in the first round, making it unlikelythat students can subscribe to the school of their second choice in the second round. Inthe years that we study, around 5% of the students could not be placed at the schoolof their first choice because they lost the lottery. When parents and students initiallychoose the best schools, losing the lottery could mean a significant reduction in school

5 of pre-university education. When graduating at once, they have their pre-university degree after 7years of secondary education instead of 6.

41

quality. Actually, some people might not choose their true first choice because theyfear losing the lottery: they might strategically choose for a less preferred but stillacceptable alternative school. Section 3.5 will describe the actual differences betweenschools attended by lottery losers and lottery winners.

3.3 Data

The data used in this study come from the municipality of Amsterdam and the Dutcheducational administration (Dienst Uitvoering Onderwijs, DUO). The municipalityprovided information on school choice, student background characteristics (such asgender and ethnicity) and primary school achievement (including primary school adviceand cito testscore). The sample includes five cohorts of students, starting secondaryeducation in 2006 to 2010. In total, information is available for 23,637 Amsterdamchildren applying for regular secondary education schools in Amsterdam. 1792 lotteryparticipating students are taken into account in this study.7

The information on academic outcomes is provided by DUO. The most recent dataincluded are educational positions in October 2012 and exam scores of the examsadministered in May 2012. We use students’ final exam results to construct an indicatorthat equals unity if the student obtained the degree of the advised school track (orhigher) on time. Otherwise the indicator equals zero. As described in section 3.2.1,students can change school track during secondary school. Since the exams are madefor each school track separately, simple exam grades are not a good outcome measure.When a student changes to a lower track, it is likely that his grades will be higher thanthey would have been at the higher school track. Instead, the school track advice ofthe primary school teacher is used as the school track the student is expected to obtainby the end of secondary school. We compare this advice to the obtained degree. Whena student gets a degree from a lower track or when he is grade retained, he is codedas not obtaining his degree on time. When a student follows the advised or a higherschool track and obtains his degree without delay, he is counted as obtaining his degreeon time.8

Because the last cohort only started secondary education in 2010, final exam results7For 2006, data is only available for students applying to lottery conducting schools. While the

municipality provided data for 24,246 students, we had to drop 2.5% (609 cases) of the observationsbecause of data problems. It turned out that 442 students (1.8%, 27 lottery participants) could notbe matched to education outcomes. These cases are less complete in general, are slightly more oftenat the pre-vocational I track and are more often weighted students. In addition, 167 students (0.7%, 9lottery participants) were dropped because they missed information on key background characteristics.

8It can also be that a student gets a higher degree with some delay. In that case, he is codedas obtaining his degree on time when he gets the higher degree at the same time as he could havegotten that degree if he first followed the school track of his primary school advice. For example, astudent with a senior general secondary education advice is considered to obtain his degree on time ifhe passes his pre-university exams in seven years.

42

are only available for 13.8% of the lottery participating students in the sample. Wetherefore also consider students’ educational positions measured four years after theadmission lottery in which a student participated took place. The fourth year is chosenbecause students are clearly tracked by the time they are in 4th grade.9 In lower grades,students can be educated in combined classes, such as senior general secondary/pre-university education.

For the fourth year, two different outcome measures are used. The first one iswhether a student is in 4th grade at the advised school track or higher after four yearsof secondary education. Basically, this measure captures the students who are on trackto obtain their final exams on time. The second measure is whether a student is in the4th grade of his advised school track or higher after four years, conditional on beingin 4th grade. Here, students who are grade retained are not taken into account. Therelevance of the second measure is that parents might care more about the school tracka student eventually obtains instead of whether he is grade retained or not. The finaloutcome measure in this study is whether a student follows a science program in thehigher grades of the academic school tracks. The science program is considered themore difficult program in the senior general secondary and pre-university school tracks(Buser et al., 2014). Since (not) choosing a science program strongly affects the optionsfor tertiary education, it is interesting to see whether winning students are more or lessinclined to choose the science program.

Returning to the sample, 1792 lottery participating students are taken into account.Table 3.1 shows the number of students at different stages of data processing. Outof the 54 schools, 16 schools conducted at least one lottery in the years considered inthis study. Within those schools, about 77% of the applying students subscribed forschool tracks with a lottery. In the process of obtaining data, we received signals thatthe lottery information from the municipality was neither complete nor fully accurate.Therefore, school visits were conducted to obtain the original admission lottery records.Two schools refused to provide information and four schools were dropped becausetheir records were incomplete and the lotteries were conducted without the presenceof a notary. Furthermore, two schools only have records for the later years and oneschool did not follow the municipality procedures in one year, leaving 10 schools and2729 students in the sample. To illustrate the importance of verifying the lotteryinformation, Appendix 3.A.1 compares the verified lottery information to the originalmunicipality data from the same schools and shows lottery balancing tests from theschools that were dropped from the sample. Section 3.4 describes balancing testsshowing that the verified lotteries were conducted fairly.

As can be seen in Table 3.1, not every student within a school and a school track9In this chapter, 4th grade refers to the 4th grade of Dutch secondary education, in which students

are approximately 16 years old.

43

Table 3.1: Number of observationsN schools N lotteries N students

All schools and students 54 23,637

Schools with lottery 16 7,737

School tracks with lottery 16 5,956

Complete lottery information 10 2,729

Rejected students 6 34

Students with priority 10 903

Lottery winners 10 23 1,206

Lottery losers 10 23 586

Lottery winners with 4th year outcomes 7 13 701

Lottery losers with 4th year outcomes 7 13 347

Lottery winners with exam outcomes 3 5 161

Lottery losers with exam outcomes 3 5 86Note: This table reports descriptive statistics on the number of students at different stages of dataprocessing.

with an admission lottery actually participates in the lottery. 34 students were rejectedbecause their test-score at the end of primary school was too low for their secondaryschool advice and 903 students were placed with priority. It might be that these stu-dents are different from the lottery participating students, which could limit the extentto which the results can be generalized.. Table 3.iv in Appendix 3.A.2 indicates thatstudents with priority are indeed somewhat more advantaged than students who par-ticipate in an admission lottery. They are, for example, less often from disadvantagedneighborhoods or from one parent families. Nevertheless, Table 3.v in Appendix 3.A.2shows that there are no significant differences in 4th grade outcomes between lotterywinners and priority students, although lottery participants obtain their degree in timesomewhat less often.

More important for the extent to which the results can be generalized is which stu-dents actually subscribe to schools conducting lotteries. Table 3.2 displays differencesin student characteristics between students who apply to schools with or without a lot-tery in the year of their secondary school choice. The results in the Table 3.2 indicatethat students who apply to a school with a lottery are quite different from students whoapply to a secondary school without a lottery. Schools with lotteries generally havea more advantaged population of students subscribing than schools without lotteries.For example, students subscribing to lottery schools have higher school track advicesfrom their primary school teachers and are less often from migrant or one parent fam-ilies. This table does not describe the differences between schools attended by lottery

44

Table 3.2: Difference between students choosing schools with and withoutlotteries in the year of school choiceDependent Mean (SD) for Mean (SD) for Difference P-values

variable schools without schools with

lotteries lotteries

(1) (2) (3) (4)

Boy 0.500 0.469 -0.031 0.000

(0.500) (0.499) (0.007)

Age 12.271 12.105 -0.166 0.000

(0.545) (0.459) (0.007)

Disadvantaged neighborhood 0.811 0.665 -0.145 0.000

(0.392) (0.472) (0.006)

Citoscore 534.148 538.834 4.685 0.000

(9.618) (7.996) (0.119)

Pre-vocational IV advice 0.150 0.135 -0.015 0.001

(0.357) (0.341) (0.005)

Pre-vocational IV/senior general 0.085 0.117 0.032 0.000

secondary advice (0.279) (0.322) (0.004)

Senior general secondary advice 0.117 0.208 0.091 0.000

(0.321) (0.406) (0.005)

Senior general secondary/ 0.105 0.180 0.075 0.000

pre-university advice (0.306) (0.384) (0.005)

Pre-university advice 0.137 0.321 0.183 0.000

(0.344) (0.467) (0.006)

Weighted student funding 0.484 0.216 -0.267 0.000

(0.500) (0.412) (0.006)

Grade retained in primary 0.126 0.056 -0.070 0.000

education (0.332) (0.229) (0.004)

Grade skipped in primary 0.005 0.009 0.004 0.001

education (0.071) (0.095) (0.001)

One parent family 0.119 0.059 -0.061 0.000

(0.324) (0.235) (0.004)

First or second generation 0.720 0.513 -0.208 0.000

migrants (0.449) (0.500) (0.007)

Number of students 15900 7737 23637Note: Columns (1) and (2) display the means and standard deviations for students applying to aschool without and with a lottery in the year that they choose a school. Columns (3) and (4) reportseparate regression coefficients and the p-values of the variables indicated in each row on an indicatorvariable equalling 0 if the school had no lottery in that year, and equalling 1 if the school had a lotteryin that year. Robust standard errors are reported in parentheses.

45

losers and lottery winners: after losing a lottery, students can still choose an alternativeschool that has vacant seats. Section 3.5 therefore describes the differences in schoolpopulation and school quality between the schools attended by lottery winners andlottery losers.

3.4 Empirical strategy

The main question addressed in this chapter is the impact of losing a school admissionlottery on student’s academic outcomes. For this, equation 3.1 is estimated:

yil = δLoseil +X ′ilβ + νl + εil (3.1)

Here, yil indicates the outcomes y of student i participating in lottery l . Loseil is anindicator variable for whether student i lost the lottery for the school of his first choice,making δ the parameter of interest. X ′

il is a vector of student characteristics added forprecision. It includes gender, age, living in a disadvantaged neighborhood, ethnicity,cito testscore, missing cito testscore, primary school advice, weighted student funding inprimary education, missing information on weighted student funding, grade retentionin primary education, skipping a grade in primary education, missing informationon grade retention in primary education, living in a one-parent family, being a firstgeneration migrant and being a second generation migrant. νl is a fixed effect forlottery l. When a school has lotteries for multiple school tracks in a single year, orwhen it has lotteries in multiple years, a fixed effect is included for each lottery. Thelottery fixed effects make sure that we compare losing and winning students within thesame lottery, instead of comparing students subscribing for different schools and schooltracks or students subscribing in different years, with probably different characteristics.To be specific, the lottery fixed effects capture potential differences between schoolsand differences between lotteries within the same school. Since the lotteries are year-specific, there are no year fixed effects. εil is the error term.

Table 3.3 shows balancing tests, testing whether there are differences in observablecharacteristics between students who lose and win the school admission lotteries. Whensuch differences exist, students who lose and win the school admission lotteries havedifferent characteristics to begin with, making later comparisons extremely hard tointerpret. Columns 1 and 2 show the means and standard deviations for studentswinning and losing a school admission lottery. Columns 3 and 4 show the actualbalancing tests, regressing an indicator for losing the lottery on the dependent variablesdenoted in each row. To account for differences between lotteries, all regressions includelottery fixed effects.

The balancing tests indicate that for nearly all characteristics, there are no signif-

46

Table 3.3: Balancing test for lottery losers and lottery winnersDependent Mean (SD) for Mean (SD) for Balancing test P-values

variable lottery winners lottery losers lottery

(1) (2) (3) (4)

Boy 0.487 0.485 0.005 0.853

(0.500) (0.500) (0.027)

Age 11.979 12.046 0.045 0.044

(0.418) (0.409) (0.022)

Disadvantaged neighborhood 0.584 0.618 0.035 0.180

(0.493) (0.486) (0.026)

Citoscore 543.281 541.639 -0.110 0.732

(6.837) (7.789) (0.321)

Pre-vocational IV advice 0.026 0.031 -0.005 0.350

(0.158) (0.173) (0.005)

Pre-vocational IV/senior general 0.022 0.039 0.003 0.585

secondary advice (0.145) (0.194) (0.006)

Senior general secondary advice 0.114 0.176 0.007 0.577

(0.318) (0.381) (0.012)

Senior general secondary/ 0.129 0.225 -0.015 0.178

pre-university advice (0.335) (0.418) (0.011)

Pre-university advice 0.710 0.529 0.010 0.245

(0.454) (0.500) (0.009)

Weighted student funding 0.068 0.090 0.013 0.336

(0.252) (0.287) (0.014)

Grade retained in primary 0.032 0.034 -0.010 0.293

education (0.177) (0.182) (0.010)

Grade skipped in primary 0.018 0.012 -0.001 0.872

education (0.134) (0.109) (0.007)

One parent family 0.040 0.068 0.024 0.044

(0.196) (0.252) (0.012)

First or second generation 0.374 0.374 -0.015 0.546

migrants (0.484) (0.484) (0.025)

Number of students 1206 586 1792Note: Columns (1) and (2) display the means and standard deviations for students winning andlosing a school admission lottery. Columns (3) and (4) report separate regression coefficients and thep-values of the variables indicated in each row on an indicator variable equalling 0 if the student wonthe lottery and equalling 1 if the student lost the lottery. All regressions include lottery fixed effects.Robust standard errors are reported in parentheses.

47

icant differences between students who lose and win the admission lotteries. For ageand being from a one-parent family, the coefficients are significant at the 5% level, butsince the actual differences are rather small and many variables were tested, the generalconclusion from the table should be that the lotteries were conducted fairly. This con-clusion is seconded by the results from a joint balancing test. Table 3.3 reports separateregression coefficients, which is to facilitate interpretation. When instead regressing allbackground characteristics on the indicator for losing the lottery, we can test whetherthe coefficients for the background characteristics are jointly significantly different fromzero. Finding F(16,1753)=1.388, p=0.138 indicates that the null hypothesis cannot berejected, supporting the conclusion that the lotteries were fair.

3.5 Treatment characteristics

Before turning to the results, it is important to see how losing a school admissionlottery changes students’ school environment. When there is a large difference inschool and peer characteristics between the schools of winning and losing students, wewould expect to see more differences between lottery losers and lottery winners thanwhen their schools turn out to be rather comparable. On the other hand, when theschools are comparable, potential negative effects of losing the lottery could better beinterpreted as disappointment or a specific misfit between the school and the student.

Table 3.4 describes differences in peer characteristics and school measures fromthe Dutch educational inspectorate between the schools attended in the first year ofsecondary education by students who lose and win a school admission lottery.10 Thefirst three rows of Table 3.4 report indicators for school quality. It turns out that theschools attended by losing students score significantly higher on the index for gradeprogression in the lower grades. This is a measure of the percentage of students gettingto the third grade without grade retention, correcting for students who attend higheror lower school tracks than their primary school advice. This result can either indicatethat schools of losing students are of better quality for the first three years, or couldmean that schools of winning students are more strict with respect to moving gradesand attending higher school tracks in the lower grades. The other two quality measurespertain to the higher grades, and indicate the opposite: schools of losing students havea lower percentage of students without delay in the higher grades and have studentswith lower average grades on the final exams, which both indicate lower school qualityat schools attended by losing students.

The other education inspectorate measures give a more detailed view on the type10The Dutch education inspectorate publishes secondary school information on a yearly basis. For

each student, the measures published in the year of school choice are used. The indicators pertainingto higher grades and final exams are published for each school track separately. For these variables,an average weighted by the number of students in each school track was computed.

48

of schools attended by lottery losers: the schools have a lower percentage of academic(pre-university and senior general secondary education) students in the higher gradesand they have a smaller number of students in general.11 The percentage of studentsfollowing a science program is higher.

With respect to the characteristics of students’ first year peers, it turns out thatlosing a school admission lottery brings students a less advantaged peer group: los-ing students have more first-year peers from disadvantaged neighborhoods, from aone-parent or migrant family and more peers receiving weighted student funding inprimary education. Further, the peers of losing students achieve worse academically:the average citoscore is lower and they were more often grade retained in primaryeducation. Somewhat surprisingly, losing a school admission lottery does not changethe distance from a student’s home to the attended secondary school, nor the shareof primary school peers attending the same school, which makes it unlikely that stu-dents choose their neighborhood school as alternative when they lost the lottery forthe school of their first choice.

To summarize, students who lose a school admission lottery attend schools with’worse’ peers in terms of prior academic achievement and social background than lotterywinners, while it does not change the distance from home to school or the share ofprimary school peers attending the same secondary school. Regarding the quality ofthe secondary schools attended by lottery losers, the picture is mixed. Schools attendedby lottery losers perform better on the measure for the first three years but worse onthe education inspectorate’s quality measures for the higher grades.

3.6 Results

Table 3.5 shows the impact of losing a school admission lottery on secondary schoolperformance. The first row shows that the academic achievement of losing students isnot harmed, even though they attend schools with less advantaged peer groups andsuffer the disappointment of not attending the secondary school of their first choice. Infact, losing students achieve better in the first years of their secondary school career.Compared to students who win the lotteries, losing students are more often in the 4thgrade of the advised school track or better after four years of secondary education.

When a student is not on the advised track in 4th grade after four years of secondaryeducation, this can have two different meanings: it can be that the student is graderetained, or that he has moved to a lower school track. The second line of results

11The latter number contrasts with the first-year cohort size, which is actually higher at schoolsattended by lottery losers. While seemingly contradictory, the explanation for this is that winningstudents more often attend one-track pre-university schools, in which all students follow 6 years ofeducation. At schools offering lower tracks, a significant proportion of the students finish secondaryschool after four or five years, which can make the first grades larger, but the higher grades smaller.

49

Table 3.4: Differences in education inspectorate measures and first yearpeer characteristics because of losing the lotteryDependent variable N Mean (SD) Difference (SE)

Education inspectorate measures for secondary schools

Index grade progression lower grades 1681 101.562 (6.392) 5.684*** (1.911)

% no delay in higher grades 1645 67.483 (12.302) -8.600*** (3.147)

Average grade on final exam 1645 6.415 (0.325) -0.291*** (0.053)

% of academic students in higher grades 1684 93.007 (18.694) -5.613** (2.691)

Number of students at entire school 1686 912.271 (252.931) -95.481*** (33.932)

% of students in one-track classes in first year 1686 58.881 (42.040) -6.504 (9.713)

% of academic students with science program 1634 11.505 (4.174) 2.419** (1.146)

First year peer characteristics

Cohort size 1580 173.064 (54.147) 36.322*** (10.375)

Share of boys 1580 0.502 (0.083) 0.045*** (0.013)

Mean age 1580 12.013 (0.101) 0.062*** (0.016)

Share disadvantaged neighborhood 1580 0.542 (0.149) 0.132*** (0.024)

Mean citoscore 1580 542.081 (4.060) -2.551*** (0.417)

Share weighted student funding 1580 0.109 (0.108) 0.136*** (0.017)

Share grade retained in primary education 1580 0.037 (0.030) 0.010** (0.005)

Share grade skipped in primary education 1580 0.017 (0.013) -0.005** (0.002)

Share one parent family 1580 0.048 (0.032) 0.037*** (0.006)

Share first and second generation migrants 1580 0.384 (0.119) 0.162*** (0.022)

Share of primary school peers attending 1573 0.076 (0.084) 0.006 (0.006)

same school

Distance to school 1792 3.186 (2.137) -0.112 (0.116)Note: Each row reports an OLS regression regressing the dependent variable mentioned in each rowon an indicator for losing the lottery for the school of your first choice. The first column reports thenumber of students in the regression, the second and third mention the mean and standard deviationof the dependent variable. The last two columns mention the regression coefficient and its standarderror. The numbers of students in the regressions differ since not all indicators are available forall years and schools. Standard errors are clustered at the level of school of placement by cohort.All regressions include lottery fixed effects and controls. Controls include gender, age, living in adisadvantaged neighborhood, ethnicity, cito testscore, missing cito testscore, primary school advice,weighted student funding in primary education, missing information on weighted student funding,grade retention in primary education, skipping a grade in primary education, missing information ongrade retention in primary education, living in a one-parent family, first generation immigrant andsecond generation immigrant. ***p<0.01 **p<0.05 *p<0.10.

50

Table 3.5: Impact of losing on school performanceOutcome N Mean (SD) (1) (2)

In 4th grade of advised track or better after 1048 0.800 0.073*** 0.087***

four years (0.400) (0.023) (0.024)

In 4th grade of advised track or better after 889 0.943 0.027* 0.035**

four years (cond. on being in 4th grade) (0.233) (0.016) (0.017)

Obtained degree on time on 247 0.595 -0.024 -0.022

advised track (or higher) (0.492) (0.054) (0.061)

Enrolled in science program 929 0.274 -0.034 -0.041

(0.446) (0.032) (0.032)

Controls !

Note: Each row reports two OLS regressions regressing the dependent variable mentioned in each rowon an indicator for losing the lottery of the school of your first choice. The first column reports thenumber of students in the regression, the second mentions the means and standard deviations for thesestudents on the dependent variable. Standard errors are clustered at the level of school of placementby cohort, and are reported in parentheses. All regressions include lottery fixed effects. Controlsinclude gender, age, living in a disadvantaged neighborhood, ethnicity, cito testscore, missing citotestscore, primary school advice, weighted student funding in primary education, missing informationon weighted student funding, grade retention in primary education, skipping a grade in primaryeducation, missing information on grade retention in primary education, living in a one-parent family,first generation immigrant and second generation immigrant. ***p<0.01 **p<0.05 *p<0.10.

indicates that both may be true: when grade retained students are not taken intoaccount, the coefficients for the impact of losing on school performance are still positiveand significant, although the point estimates are smaller. Since parents might valueretaining a higher school track more than they dislike grade retention, it is interestingto see that winning students in the 4th grade after four years more often attend a lowerschool track than can be expected from their primary school advice. So contrastingto popular beliefs, it seems that the academic achievement of losing students is ratherhelped than harmed in the first years of secondary education.

These results are not replicated for obtaining the degree of the advised schooltrack in time: the coefficients in the third row are negative and insignificant. Thereare two potential explanations for this difference. First, the number of students withinformation on final exam results is much lower: we have 247 students with informationon exams, but 1048 students with information after four years. Therefore, a lot ofprecision is lost, which is illustrated by the larger standard errors of the effect oflosing the lottery on exam outcome. In fact, the standard errors are that large that wecannot exclude the possibility that the coefficients are just as positive as the coefficientsfor the 4th grade outcomes. Reversely, when only taking into account students withexam measures available on the 4th grade outcomes, the 4th grade outcomes becomeinsignificant as well. The standard errors get larger and while the coefficients get less

51

positive, the original estimates still fall within the 95% confidence intervals.12

The second potential explanation is that lottery winners catch up in the highergrades. As described in Section 3.5, schools attended by lottery winners score loweron the education inspectorate’s measure for progression in the lower grades, while theydo better on the quality measures for the higher grades. This can be a reflection ofdifferent school policies: some schools quickly decide that a student should be graderetained or follow a lower school track, while other schools are more lenient in thelower grades. The stricter schools will appear to achieve worse in the lower grades(they retain more students and demote more students to a lower track), while theymight do better in the higher grades, because the low-achieving students have alreadybeen transferred to lower grades or school tracks. On the other hand, the studentsin lenient schools probably achieve worse in the higher grades. If the school trackeventually turns out to be too ambitious, students in lenient schools will more oftenget grade retained, sent to a lower school track or fail exams in the higher grades. Ifthe results for the lower grades indeed reflect school policies, it is well possible that theachievement of lottery losers turns out to be worse than the achievement for lotterywinners in the higher grades, which may change the eventual education outcomes. Forchoosing a science program, no significant differences are found between lottery losersand lottery winners.

It could be that losing a school admission lottery has a different impact on boysand girls. For example, it could be that they respond differently to the initial disap-pointment of losing the school admission lottery. In Table 3.6, no evidence is foundfor such differential effects. None of the interaction effects comes out significantly, andboth lottery losing boys and girls are more often in the 4th grade of the advised trackor better after four years of secondary education.

The lottery winners for whom we have 4th grade outcomes went to seven differentschools. Lottery winners for whom we have exam outcomes went to three differentschools. The results can therefore potentially be driven by one or two schools attendedby winners. To assess whether the findings depend on a few schools only, Table 3.7shows the impact of losing a school admission lottery on school performance for eachlottery conducting school separately. The school names are anonymized. The resultsin the table are consistent with the overall results. For the 4th grade outcomes, 8out of 14 coefficients come out positive and significant and none of the coefficients arenegative and significant, implying that the results are not driven by one or two non-typical schools. For the exam results, the coefficients are insignificant: the standarderrors are very large and the sample sizes are small. As with the overall results, losinga school admission lottery generally has no significant impact on choosing a scienceprogram, although it seems that students admitted to school 5 more often choose the

12Results are summarized in Table 3.vi in Appendix 3.A.2.

52

Table 3.6: Differential effects of losing for boys and girlsTreatment Interaction Treatment

Outcome N effect girls effect effect boys

(1) (2) (1+2)

In 4th grade of advised track or better after 1048 0.081** 0.013 0.094***

four years (0.032) (0.038) (0.030)

In 4th grade of advised track or better after 889 0.028 0.014 0.043**

four years (cond. on being in 4th grade) (0.025) (0.031) (0.021)

Obtained degree on time on 247 -0.068 0.095 0.028

advised track (or higher) (0.080) (0.086) (0.069)

Enrolled in science program 929 -0.005 -0.071 -0.076

(0.036) (0.055) (0.048)Note: Each row reports two OLS regressions regressing the dependent variable mentioned in each rowon an indicator for losing the lottery of the school of your first choice. Columns (1) and (2) displaythe treatment effect for girls and the interaction effect on boys. Column (3) is another OLS regressiondepicting the treatment effect for boys. Standard errors are clustered at the level of school of placementby cohort, and are reported in parentheses. All regressions include lottery fixed effects. Controlsinclude gender, age, living in a disadvantaged neighborhood, ethnicity, cito testscore, missing citotestscore, primary school advice, weighted student funding in primary education, missing informationon weighted student funding, grade retention in primary education, skipping a grade in primaryeducation, missing information on grade retention in primary education, living in a one-parent family,first generation immigrant and second generation immigrant. ***p<0.01 **p<0.05 *p<0.10.

science program than their lottery losing counterparts.

3.7 Conclusions

In this study, we investigated the effects of losing a school admission lottery on thecharacteristics of the school attended and on student outcomes. Students who lostan admission lottery attend schools with ’worse’ peers in terms of prior academicachievement and social background than winners. Lottery losers also attend schoolsthat according to the measures of the Dutch education inspectorate, perform betteron grade progression in the lower grades but worse on grade progression in the highergrades and on exam scores. The results indicate that students’ academic achievementis not harmed by losing a school admission lottery. After four years of secondaryeducation, lottery losing students even seem to achieve somewhat better than lotterywinning students. Students who win a school admission lottery are less often in the 4thgrade of their advised school track or higher after four years of secondary education.Moreover, when they are in 4th grade after four years, they are more often at a lowerschool track than could be expected from their primary school advice.

The results presented in this chapter are preliminary. Because most of the students

53

Table 3.7: Impact of losing on school performance for each school separatelyIn 4th grade of In 4th grade of Obtained Enrolled

advised track or advised track or degree on in science

better after better after four time on program

four years years (cond. on advised track

being in 4th grade) (or higher)

School 1 -0.007 0.065** 0.032 -0.079

(0.016) (0.028) (0.058) (0.064)

307 355 128 350

School 2 0.083*** 0.063 0.001

(0.010) (0.074) (0.038)

113 142 106

School 3 -0.015 0.052 -0.100 -0.117

(0.159) (0.135) (0.125) (0.142)

62 81 69 73

School 4 0.055*** 0.129* 0.069

(0.009) (0.057) (0.059)

110 125 110

School 5 0.006 0.130** -0.205**

(0.023) (0.038) (0.068)

114 132 131

School 6 0.068*** 0.161*** 0.033

(0.013) (0.017) (0.100)

76 83 76

School 7 0.208*** 0.119 -0.269 0.066

(0.063) (0.092) (0.178) (0.070)

107 130 50 83Note: Each row reports four OLS regressions regressing the dependent variable mentioned in thecolumn header on an indicator for losing the lottery for the school in that row. Standard errors areclustered at the level of school of placement by cohort, and are reported in parentheses. The numbersof students in each regression are mentioned below the standard errors. All regressions include lotteryfixed effects and controls. Controls include gender, age, living in a disadvantaged neighborhood,ethnicity, cito testscore, missing cito testscore, primary school advice, weighted student funding inprimary education, missing information on weighted student funding, grade retention in primaryeducation, skipping a grade in primary education, missing information on grade retention in primaryeducation, living in a one-parent family, first generation immigrant and second generation immigrant.***p<0.01 **p<0.05 *p<0.10.

54

in our sample have not yet written their final exams, we do not know whether thestudents that won the admission lotteries will catch up with the students that lostthe lotteries, and perhaps even overtake them. Estimation results based on the smallsubsample that already wrote their final exam, are too imprecise to be informative.

The preliminary finding that lottery losers are not harmed is somewhat unexpected.In a system of free school choice, we would expect parents to choose the best schoolfor their child. If lottery losers turn out to achieve better, there are three potentialexplanations. First, it might be that lottery losers get to be relatively well-achievingstudents in their school, which might boost their self-esteem, and thereby their aca-demic achievement (Murphy & Weinhardt, 2013). Second, it might be that parentschange their behavior when a student loses the lottery: they might decide to rely lesson the school and compensate their child by investing more in his education (Pop-Eleches & Urquiola, 2013). Third, it might be that parents value school characteristicsthat do not necessarily help academic achievement. It could be that parents value anadvantaged peer group because it provides students a ’safe’ group of friends. At least,the results would imply that the disappointment effect of not attending the school ofyour first choice is limited and can be outweighed by other factors.

Similar to the results of Deming (2014), it is interesting to note that the results thatwe have obtained until now are in line with the school assessment by the Dutch educa-tion inspectorate. The schools that are attended by lottery losers perform according tothe inspectorate, better on grade progression in the lower grades than the school thatare attended by lottery winners. Our finding that lottery losers have better outcomesafter four years than lottery winners, indicates that the inspectorate’s assessment alsoholds if we correct in the best possible way (through random assignment) for differencesin student inflow.

Recall that with the school assignment mechanism in Amsterdam, lottery losers aremore likely to end up in a school further down their preference list than with otherschool assignment mechanisms. It is therefore remarkable that, based on outcomes fouryears after the admission lottery, lottery losers are not performing worse than lotterywinners.

We plan to report on the next stage of this project in the future. This will includeestimation of the impact of losing an admission lottery on exam results based oninformation for the entire sample. It will also include estimation of the impact oflosing a lottery on subsequent education choices such as the decision to enroll in highereducation and the choice of the field of study. If the new results confirm the currentfinding that lottery losers are not doing worse than lottery winners, we want to collectdata that could inform us about possible mechanisms behind this finding. Focus groupinterviews with a small groups of parents and students may provide clues about howthey dealt with the setback of losing a lottery.

55

3.A Appendix

3.A.1 Balancing tests with original lottery data

As described in Section 2.3, school visits were conducted to verify the lottery dataprovided by the municipality. In this section, the original data provided by the munici-pality is compared to the verified lottery data. Further, balancing tests are reported forthe schools that are dropped from the sample because the information was incompleteor the schools refused to provide information.

Table 3.i compares the verified information to the municipality data, and showsthat 46 students with priority (5.1%) were not identified as having priority originally.Further, 27 students (1.5%) initially appeared to enroll for a school track that didnot have a lottery, but turned out to be a lottery participant. Reversely, 46 studentsappeared to participate in a lottery, while they turned out to enroll for a school trackthat did not have a lottery (results not shown). A small number of students (7)participated in another lottery than would be expected based on the municipalityinformation. For example, they seemed to participate in a pre-vocational track lottery,while they actually participated in a lottery for academic tracks. Finally, 82 studentswere registered as losing a lottery while the school did not conduct a lottery. Sincethese are obviously administrative mistakes, these students are disregarded.

When checking the lottery balance of the sample schools using the municipalitydata, Table 3.ii indicates that there appears to be a problem with the lotteries. Nextto age and being from a one-parent family, some of the school advices come out sig-nificantly. Further, lottery losers more often seem to be from a disadvantaged neigh-borhood, and more often received weighted student funding in primary education.The F-test for the joint balancing test also comes out significantly (F(16,1818)=1.996,p=0.011), providing further evidence that the original municipality data was not fullyaccurate. When verifying the lottery data, most schools provided the original notarystatements. Since these are legal documents, we are confident that the verified dataare the correct data.

Since six schools were dropped from the sample altogether and three more schoolswere dropped in earlier years, it is also interesting to see the lottery balance for theselotteries. In the balancing test shown in Table 3.iii, some of the variables come outsignificantly; indicating that lottery losers are more often from disadvantaged neighbor-hoods, have lower citoscores, more often received weighted student funding in primaryschool, less often skipped a grade in primary education and are more often first orsecond generation migrants. The F-test for the joint balancing test also comes outsignificantly (F(16,1893)=4.140, p<0.001), so the null hypothesis of no differences be-tween lottery losers and lottery winners should be rejected.

As shown above, the lottery data from the municipality is not fully accurate. Since

56

Table 3.i: Differences between the municipality data and the verified lotterydata

Original lottery data

Students Students Rejected No lottery Total

participated with students for students’

in lottery priority track

Students1,765 0 0 27 1,792participated

Verified in lottery

lottery Students with46 837 0 20 903

data priority

Rejected0 1 31 2 34

students

Total 1,811 838 31 49 2,729Note: This table reports differences in the number of students between the original lottery dataprovided by the municipality and the verified lottery data.

Table 3.iii is completely based on the municipality data, the differences could be dueto inaccuracies in the data. Given that the schools refused or could not provide morereliable information, however, we cannot exclude the possibility that the lotteries inthese schools were actually unbalanced.

57

Table 3.ii: Balancing test for lottery winners and lottery losers using theoriginal dataDependent Mean (SD) for Mean (SD) for Balancing test P-values

variable lottery winners lottery losers lottery

(1) (2) (3) (4)

Boy 0.468 0.483 0.018 0.487

(0.499) (0.500) (0.026)

Age 11.988 12.049 0.049 0.024

(0.427) (0.411) (0.022)

Disadvantaged neighborhood 0.581 0.619 0.045 0.080

(0.494) (0.486) (0.026)

Citoscore 543.111 541.791 -0.269 0.391

(6.890) (7.728) (0.314)

Pre-vocational IV advice 0.033 0.031 0.000 0.964

(0.180) (0.175) (0.006)

Pre-vocational IV/senior general 0.035 0.040 0.000 0.948

secondary advice (0.184) (0.197) (0.006)

Senior general secondary advice 0.111 0.173 0.032 0.005

(0.314) (0.379) (0.011)

Senior general secondary/ 0.145 0.226 -0.035 0.002

pre-university advice (0.352) (0.418) (0.011)

Pre-university advice 0.676 0.530 0.003 0.634

(0.468) (0.500) (0.007)

Weighted student funding 0.065 0.091 0.025 0.066

(0.247) (0.288) (0.014)

Grade retained in primary education 0.034 0.035 -0.004 0.665

(0.182) (0.184) (0.010)

Grade skipped in primary education 0.017 0.012 -0.001 0.915

(0.130) (0.110) (0.006)

One parent family 0.040 0.070 0.030 0.012

(0.195) (0.255) (0.012)

First or second generation migrants 0.366 0.372 0.000 0.994

(0.482) (0.484) (0.025)

Number of students 1285 572 1857Note: Columns (1) and (2) display the means and standard deviations for students winning and losinga lottery. Columns (3) and (4) report separate regression coefficients and the p-values of the dependentvariables indicated in each row on an indicator variable equalling 0 if the student won the lottery andequalling 1 if the student lost the lottery. All regressions include lottery fixed effects. Robust standarderrors are reported in parentheses.

58

Table 3.iii: Balancing test for lottery winners and lottery losers at schoolsthat were dropped from the sampleDependent Mean (SD) for Mean (SD) for Balancing test P-values

variable lottery winners lottery losers lottery

(1) (2) (3) (4)

Boy 0.476 0.467 -0.022 0.447

(0.500) (0.499) (0.029)

Age 12.177 12.266 0.019 0.514

(0.478) (0.539) (0.029)

Disadvantaged neighborhood 0.712 0.821 0.056 0.013

(0.453) (0.384) (0.023)

Citoscore 536.855 532.880 -1.380 0.000

(7.507) (7.631) (0.369)

Pre-vocational IV advice 0.280 0.371 0.000 0.990

(0.449) (0.484) (0.007)

Pre-vocational IV/senior general 0.104 0.112 -0.001 0.825

secondary advice (0.305) (0.316) (0.005)

Senior general secondary advice 0.231 0.165 0.001 0.918

(0.422) (0.371) (0.010)

Senior general secondary/ 0.194 0.065 -0.008 0.383

pre-university advice (0.395) (0.247) (0.009)

Pre-university advice 0.115 0.024 0.000 0.203

(0.319) (0.152) (0.000)

Weighted student funding 0.290 0.395 0.078 0.001

(0.454) (0.489) (0.023)

Grade retained in primary education 0.077 0.087 -0.010 0.506

(0.267) (0.282) (0.015)

Grade skipped in primary education 0.004 0.000 -0.001 0.040

(0.065) (0.000) (0.001)

One parent family 0.077 0.092 0.011 0.459

(0.266) (0.290) (0.015)

First or second generation migrants 0.599 0.783 0.107 0.000

(0.490) (0.413) (0.023)

Number of students 1408 552 1960Note: Columns (1) and (2) display the means and standard deviations for students winning and losinga lottery. Columns (3) and (4) report separate regression coefficients and the p-values of the variablesindicated in each row on an indicator variable equalling 0 if the student won the lottery and equalling1 if the student lost the lottery. All regressions include lottery fixed effects. Robust standard errorsare reported in parentheses.

59

3.A.2 Additional tables

Table 3.iv: Differences in background characteristics between students par-ticipating in the lottery and students being placed with priority

Mean (SD) for Mean (SD) for Difference P-values

Dependent students with students without

variable priority priority

(1) (2) (3) (4)

Boy 0.435 0.486 -0.007 0.755

(0.496) (0.500) (0.022)

Age 12.047 12.001 -0.004 0.826

(0.452) (0.416) (0.019)

Disadvantaged neighborhood 0.557 0.595 0.086 0.000

(0.497) (0.491) (0.022)

Citoscore 541.190 542.744 -0.120 0.650

(7.552) (7.201) (0.265)

Pre-vocational IV advice 0.078 0.027 0.002 0.804

(0.268) (0.163) (0.008)

Pre-vocational IV/senior general 0.094 0.027 -0.010 0.233

secondary advice (0.292) (0.163) (0.009)

Senior general secondary advice 0.155 0.134 0.009 0.436

(0.362) (0.341) (0.012)

Senior general secondary/ 0.190 0.160 0.002 0.891

pre-university advice (0.393) (0.367) (0.012)

Pre-university advice 0.483 0.651 -0.003 0.787

(0.500) (0.477) (0.010)

Weighted student funding 0.042 0.075 0.049 0.000

(0.201) (0.264) (0.010)

Grade retained in primary education 0.055 0.033 -0.004 0.669

(0.229) (0.178) (0.009)

Grade skipped in primary education 0.013 0.016 0.000 0.964

(0.115) (0.126) (0.006)

One parent family 0.039 0.049 0.023 0.010

(0.193) (0.216) (0.009)

First or second generation migrants 0.278 0.374 0.106 0.000

(0.448) (0.484) (0.021)

Number of students 903 1792 2695Note: Columns (1) and (2) display the means and standard deviations for students who got placedwith priority and for students without priority who participated in the lottery. Columns (3) and (4)report separate regression coefficients and the p-values of the dependent variables indicated in eachrow on an indicator variable equalling 0 if the student was placed with priority and equalling 1 ifthe student participated in the lottery. All regressions include lottery fixed effects. Robust standarderrors are reported in parentheses.

60

Table 3.v: Differences in outcomes between students winning the lotteryand students being placed with priorityOutcome N OLS

In 4th grade of advised track or better after 1209 -0.016

four years (0.034)

In 4th grade of advised track or better after 1020 -0.017

four years (cond. on being in 4th grade) (0.026)

Obtained degree on time on 300 -0.090**

advised track (or higher) (0.033)

Enrolled in science program 1014 -0.032

(0.031)

Controls !

Note: Each coefficient reports an OLS regression regressing the outcomes mentioned in each rowon an indicator for lottery participation. The regressions include the students who got placed withpriority and the students who won the lottery for the school they preferred (lottery losers are not takeninto account since they usually do not attend the school of their first choice). Standard errors areclustered at the school of placement by cohort, and are reported in parentheses. All regressions includelottery fixed effects and student controls. Controls include gender, age, living in a disadvantagedneighborhood, ethnicity, cito testscore, missing cito testscore, primary school advice, weighted studentfunding in primary education, missing information on weighted student funding, grade retention inprimary education, skipping a grade in primary education, missing information on grade retention inprimary education, living in a one-parent family, first generation immigrant and second generationimmigrant. ***p<0.01 **p<0.05 *p<0.10.

61

Table 3.vi: Results: Impact of losing on school performance for studentswith exam results availableOutcome N Mean (SD) (1) (2)

In 4th grade of advised track or better after 243 0.745 0.038 0.038

four years (0.437) (0.049) (0.047)

In 4th grade of advised track or better after 199 0.910 -0.013 -0.025

four years (cond. on being in 4th grade) (0.288) (0.035) (0.035)

Obtained degree on time on 247 0.595 -0.024 -0.022

advised track (or higher) (0.492) (0.054) (0.061)

Enrolled in science program 205 0.351 -0.045 -0.062

(0.479) (0.090) (0.095)

Controls !

Note: Each row reports two OLS regressions regressing the dependent variable mentioned in eachrow on an indicator for losing the lottery of the school of your first choice. The first column reportsthe number of students in the regression, the second mentions the means and standard deviations forthese students on the dependent variable. The numbers of students are not equal to 351 because ofvariable definitions. The five students missing in the first row had unclear educational positions in4th grade. The 96 additional students missing in the second row were grade retained and thereforenot taken into account. For the science program, only students in the academic tracks were taken intoaccount. Standard errors are clustered at the level of school of placement by cohort and are reportedin parentheses. All regressions include lottery fixed effects. Controls include gender, age, living in adisadvantaged neighborhood, ethnicity, cito testscore, missing cito testscore, primary school advice,weighted student funding in primary education, missing information on weighted student funding,grade retention in primary education, skipping a grade in primary education, missing information ongrade retention in primary education, living in a one-parent family, first generation immigrant andsecond generation immigrant. ***p<0.01 **p<0.05 *p<0.10.

62

Chapter 4

The effects of Montessori education:Evidence from admission lotteries

4.1 Introduction

This study investigates the causal effects of Montessori secondary education by ex-ploiting school admission lotteries. Currently, over 22,000 schools in 117 countries areeducating their students using the Montessori pedagogy (Association Montessori Inter-nationale, 2013). Notwithstanding the large number of students involved, Montessorimethods have passionate advocates and strong opponents.

The advocates claim that Montessori education has positive effects on students’independence and motivation. By providing students with a large amount of choicein their school tasks, independence would be fostered, while motivation is enhancedby minimizing external rewards (Lillard, 2005). Critics, on the other hand, arguethat Montessori education can harm academic achievement by a lack of structure andacademic standards (Chattin-McNichols, 1992).

For parents, students and policy makers, it is important to know which claimsare valid. When Montessori education has positive effects on students, schools maybe improved by a broader implementation of Montessori methods. Yet, if Montessorieducation negatively affects students, students and parents could better opt for regulareducation schools. When there are both positive and negative effects, this is highlyrelevant information for parents in order to make an informed decision about theirchild’s education. This chapter therefore investigates the effects of being exposed toMontessori education.

Despite its popular character, little is known about the effectiveness of Montes-sori methods. The reason is that students in Montessori schools cannot be directlycompared to regular education students. Students and parents select themselves intoMontessori education. Parents opting for Montessori schools are, for example, gener-

63

ally of higher socioeconomic status and less often from ethnic minorities (Dohrmannet al., 2007; Rathunde & Csikszentmihalyi, 2005a).

While this study is the first to investigate the causal effects of Montessori secondaryeducation, there is a handful of empirical studies for primary education. Lillard & Else-Quest (2006) study students who won and lost an admission lottery for a Montessoriprimary school in Wisconsin. They find that 5 and 12 year old Montessori studentsare doing just as well or better than their non-Montessori counterparts on cognitiveand social measures. In two older studies, Miller et al. (1975) and Karnes et al. (1970)randomly assigned preschool students to Montessori or non-Montessori Head Startprograms, and found no or little initial effects of Montessori education compared toother preschool programs. In the longer run, one study finds higher achievement andIQ for Montessori educated boys, but not for girls (Miller & Bizzell, 1984). The secondstudy finds that Montessori preschoolers more often graduate from high school andexperience less grade retention (Karnes et al., 1983). Both studies (and their follow-ups), however, implemented compromised versions of Montessori programs in terms ofage grouping, teacher training and day routine and suffer from significant attrition andsmall samples.

This chapter circumvents the selection issues described earlier by exploiting schooladmission lotteries. The admission lotteries basically create a series of randomizedexperiments: they determine by chance which students are educated at Montessorischools and thereby create valid treatment and control groups. By using administrativedata on academic achievement and questionnaire data on socio-emotional functioning,this study investigates a wide range of potential outcomes of Montessori education.

The results show that the academic achievement of Montessori students is akin tothe academic achievement of students in regular secondary education. Montessori stu-dents obtain their secondary school degree without delay at the same rate and withsimilar grades as non-Montessori students. The socio-emotional functioning of Montes-sori and non-Montessori students turns out to be comparable as well. Students showsimilar levels of motivation and Montessori students do not score better on measuresof independence. These results imply that neither the claims of the opponents, nor theclaims of the critics are supported by the evidence.

The chapter is structured as follows. Section 4.2 discusses the characteristics ofMontessori education. Section 4.3 provides additional background information on sec-ondary education in the Netherlands. Section 4.4 and 4.5 describe the data and empir-ical strategy. Section 4.6 presents and discusses the results and Section 4.7 concludes.

64

4.2 Montessori education

4.2.1 Montessori primary education

Montessori education was developed by Maria Montessori, starting with the ’Casa deiBambini’ in a poor neighborhood in Rome in 1907, and spreading over the world eversince (Whitescarver & Cossentino, 2008). As Montessori education is most renown forprimary education, this section starts with a description of Montessori primary schools.

When walking into a Montessori classroom, there are several features that dis-tinguish it from traditional primary education classes. While students in traditionaleducation mainly learn from texts, students in Montessori classrooms are often workingwith their hands using Montessori materials (Lillard, 2005). Montessori materials arelargely self-correcting educational objects designed to teach subjects and concepts viarepeated use and in a structured sequence. To teach writing skills, for example, pupilspractice the phonetic sounds of letters while tracing letters made from sandpaper withtheir fingers. In the same period, they develop the motor skills needed for writing withdrawing tasks using metal frames. Later, the two activities come together (Lillard,2005).

Other distinctive features of Montessori classrooms are that classes are organizedin three-year-age groupings, such that younger students can learn from older students(Montessori, 1972). Students are allowed to choose their own activities from a youngage onwards (Mooney, 2000) and generally work and get instructed on their own orin small groups (Whitescarver & Cossentino, 2008). Further, there is an absence offormal grades and tests.

One of the main goals of Montessori education is to help students to become in-dependent. Maria Montessori believed that children need to acquire independence inorder to grow and develop (Montessori, 1989b). In early childhood education, for ex-ample, independence is fostered through practical life exercises such as table washing(Lillard, 2005). Moreover, children in Montessori classrooms can freely choose theirwork, which gives them experience in making choices (Montessori, 1989a).

Another aim of the changes in classroom environment is to create an environmentwhere students are intrinsically motivated to learn. The Montessori materials aredesigned to sparkle children’s interest. Students actively manipulate the objects andlearn through experience, which is claimed to enhance learning and motivation (Lillard,2005). Increasing motivation is also the argument for the absence of formal gradesand tests. Traditional schools offer constant feedback on achievement by providinggrades and rewards, thereby replacing intrinsic motivation towards learning by theneed for external rewards. Maria Montessori believed that such extrinsic rewards arenot necessary and even disrupt students’ learning (Lillard, 2005).

65

4.2.2 Montessori secondary education

While Montessori education is often associated with primary education, it also has alively and international secondary education community. An important part of thehistory of Montessori secondary education lies in the Netherlands. Just 7 years afterthe opening of the ’Casa dei Bambini’, the first Dutch Montessori primary schoolopened. As soon as the first students finished primary school, demand for a Montessorisecondary school developed. In 1928, the first Montessori secondary school in the worldwas opened in the Netherlands (Calff, 1980). Today, the school is still in existence andit is one of the schools considered in this study.1

Montessori secondary schools adhere to the famous Montessori adagio: “help me todo it alone”. Similar to primary education, Montessori methods in secondary educa-tion are organized towards fostering independence (Montessori, 1973) and creating anenvironment for intrinsic motivation (Rathunde & Csikszentmihalyi, 2005b).

Modern2 Montessori secondary schools differ from regular schools in a numberof ways. Most salient is the amount of choice students are allowed in their school-work: part of the hours at school are specifically allocated for free choice of activities(Rathunde & Csikszentmihalyi, 2005a). Students are allowed to choose which teacherthey want to join during these hours. Further, students can choose when they want totake tests in the lower grades of secondary education3 and Montessori secondary schoolsfocus more broadly than academic achievement. Field projects, such as internships,are organized to promote social development and to learn students how to function insociety (Seldin & Epstein, 2003; Rubinstein, 2008).

4.3 Secondary education in the Netherlands

Dutch students and their parents have free school choice, they are not restricted bycatchment areas or school fees. Moreover, virtually all primary and secondary schoolsare completely publicly funded. This includes the religious and special program schools,such as Montessori schools. The government funding is nationally determined andlargely dependent on student numbers. Schools can get additional funding for studentsfrom disadvantaged neighborhoods.

1Nowadays there are 16 Montessori secondary schools in the Netherlands. All schools participatein the Dutch national exams, so the content of the curriculum has large overlap with regular secondaryeducation. The Dutch Montessori secondary schools conduct visitations amongst each other to guardthe Montessori quality of the schools.

2As Maria Montessori passed away while developing Montessori methods for adolescents Lillard(2013), Montessori secondary education is largely developed by the Montessori community. MariaMontessori’s writings on secondary education include a radical boarding school program called ’Erd-kinder’. This program has been implemented very sparsely, and is not the focus of this chapter.

3At some schools, students can also determine the timing of (part of) the tests in the higher gradesof secondary school.

66

Figure 4.1: The Dutch secondary school system

Dutch secondary education starts at age 12, and lasts four to six years. The schoolsystem is highly tracked, Figure 4.1 gives a graphic description. Not all secondaryschools offer all school tracks and schools can educate children of different tracks to-gether. The highest track (pre-university education, vwo) takes six years, and givesaccess to university education. The intermediate track (senior general secondary ed-ucation, havo) takes five years, and gives access to universities of applied sciences,also called higher professional education. The lowest tracks (pre-vocational educa-tion, vmbo) last four years, and give access to vocational education programs. Withinthe pre-vocational track, there are four different levels. Only the highest pre-vocationallevel (tl, from hereon referred to as pre-vocational IV) and the senior general secondaryand pre-university tracks are offered at the Montessori schools in this study.

The initial school track is based on the school track advice of the primary schoolteacher and on a standardized high stakes test (in most cases a nationwide exit testcalled the “citotoets”) at the end of primary school. Dependent on student achievementand school policies, students can change track during secondary education. Further,students can follow a higher track after finishing a lower track.4 Subject to someconditions, students can choose their own courses in the second half of secondaryeducation. Secondary schools have to follow national curriculum guidelines, and allstudents take centrally determined national exams at the end of secondary school. Thenational exams count for 50% of the final grade, the other 50% is determined by schoolspecific exams taken in the last two or three years of secondary education.

4In that case, students enter the higher track in the year before the final exams. When doingpre-university education after senior general secondary education, for example, students enter in year5 of pre-university education. When graduating at once, they have their pre-university degree after 7years of secondary education instead of 6.

67

4.4 Data

4.4.1 Admission lotteries in Montessori schools

In this chapter, data from school admission lotteries in two Dutch Montessori secondaryschools are used. The names of the schools are anonymized to school A and schoolB. The schools generally receive more applications than they can enroll and conductlotteries to allocate the available places. In both schools, the lotteries are organizedfor each school year separately and are executed by a notary.

School A conducts school admission lotteries since 2003. The lotteries are di-vided into a lottery for students of the two highest school tracks (senior generalsecondary/pre-university) and for the highest pre-vocational track (pre-vocational IV).Not all applying students participate in the lottery. Students with a brother or sister atthe school, children from employees and students applying from a Montessori primaryschool get placed with priority.5 Table 4.1 shows the number of lottery participatingstudents and the number of priority students per year and school track. In total, 540students participated in a school admission lottery for school A.6

In school B, students from Montessori primary schools do not get priority (brothersand sisters of current students and children of employees do). Instead, the lotteryis based on residence: the school proportionally takes up students from 3 regions.When there is over-subscription from one region, students from this region participatein a lottery. School B does not offer the pre-vocational IV track. For school B, 260students participated in a school admission lottery, bringing the total number of lotteryparticipating students to 800.7

Since the effects of Montessori education are studied by comparing students winningadmission lotteries to students who lose the lottery and therefore attend another school,it is important to know the characteristics of the alternative schools. The majorityof the lottery losing students (80.1%) are enrolled in schools with regular education

5Since the majority of the lottery participating students subscribed to school A, this implies thatthe results are largely identified on a sample of students who did not attend Montessori primaryschools. Section 4.6.4 describes this issue in more detail. Of all students receiving priority to schoolA, 90.2% gets priority because they were educated at a Montessori primary school.

6The actual number of students applying to school A for years and school tracks that conductedlotteries is slightly higher: 1759 instead of 1709. However, 44 students (2.5%) could not be merged atthe Dutch educational administration. All of these students lost the lottery, mainly in the earlier years,but there is no selection on gender, citoscore and school advice. Other background characteristics arenot available for these students. Further, 6 students (0.3%) are dropped because of missing informationon ethnicity or living in a disadvantaged neighborhood.

7202 students from regions without lotteries in a certain year are dropped from the sample. Com-parable students at school A, in years for which a certain school track did not have a lottery, are notavailable for analysis. 1 student (0.2%) could not be merged at the Dutch educational administration.20 students (4.6%, of whom 13 participated in the lottery) are dropped from the sample because ofmissing information on primary school advice.

68

Table 4.1: Distribution of lottery participating and priority students overyearsSchool year Lottery winner Lottery loser Priority Total

School A: Senior general secondary / pre-university education

2003/2004 50 46 169 265

2004/2005 16 59 203 278

2005/2006 47 9 196 252

2009/2010 37 42 163 242

2010/2011 29 17 178 224

Total 179 173 909 1261

School A: Pre-vocational IV education

2003/2004 17 45 53 115

2004/2005 10 30 60 100

2007/2008 10 29 62 101

2008/2009 29 3 42 74

2009/2010 5 10 43 58

Total 71 117 260 448

School B: Senior general secondary/pre-university education

2008/2009 52 15 60 127

2009/2010 83 4 16 103

2010/2011 38 9 43 90

2011/2012 56 3 34 93

Total 229 31 153 413

Grand Total 479 321 1322 2122Note: This table reports the number of lottery participating students and the number of prioritystudents per school, year and school track.

69

programs. 8.5%8 still attend Montessori schools and 11.4% attend a Dalton school, aschool with a pedagogy comparable to Montessori. With respect to school tracks, mostlottery losing students are enrolled in schools similar to school A and school B. 0.63% ofthe lottery losing students attend schools that only offer the pre-university school track,while 22.8% go to schools that also offer the lower pre-vocational education tracks orattend schools that do not offer the highest track(s).

Table 4.2 describes differences in school characteristics between school A and schoolB and their alternative schools attended by lottery losers in the first year of secondaryeducation. In general, the standard deviations of the average school characteristics ofthe alternative schools are large, especially for school A, indicating that lottery losingstudents attend a wide range of alternative schools. For most characteristics, the al-ternative schools are significantly different from school A and B. The first seven rowsof Table 4.2 pertain to school quality measures of the Dutch education inspectorate.The patterns indicate that the Montessori schools have less delay in the higher gradesand higher average exam scores, although grade progression in the lower grades islower. It is important to note that these differences cannot be interpreted as effects ofMontessori education, as they reflect average school characteristics that are potentiallycreated by differences in school populations. The four bottom rows report more generalschool characteristics. It turns out that school A is significantly larger and school Bis significantly smaller than their alternative schools. With respect to the percentageof students from disadvantaged neighborhoods, school A has more students from dis-advantaged neighborhoods in the pre-vocational III/IV and pre-university tracks, butfewer in the senior general secondary track. There are no significant differences in thepercentage of students from disadvantaged neighborhoods for school B compared tothe alternative schools attended by its lottery losers.9

4.4.2 Data on academic achievement

The information on students’ academic achievement is provided by the Dutch edu-cational administration (Dienst Uitvoering Onderwijs, DUO). The most recent dataavailable are exam scores of the exams administered in May 2012 and educationalpositions in October 2012. Students’ final exam results are used to create an indica-tor that equals one if the student obtained the degree of the advised school track (or

817 losing students (5.4%) are admitted to School A or school B from a waiting list, details are inSection 4.5. The other 10 students (3.2%) attend another Montessori secondary school.

9The high percentage of students from disadvantaged neighborhoods is somewhat misleading forschool A and its alternatives. It is based on a policy taking into account the characteristics ofstudents’ residential neighborhoods. In the city of school A, this caused some peculiarities in which’elite’ schools were suddenly identified as schools with many disadvantaged students. The old policybased on individual student characteristics was still in effect in the first years considered in this study.According to that definition, the percentage of disadvantaged students was nearly zero for school A,and around 4% for its alternatives.

70

Table 4.2: Characteristics of alternative schools attended by lottery losersSchool A School B

Mean Mean (SD) Mean Mean (SD)

school A alternatives school B alternatives

(1) (2) (3) (4)

Index grade progression lower grades 94.07 100.96*** 99.03 101.08**

(13.79) (5.05)

% no delay pre-voc. III/IV 81.19 74.70***

(10.68)

% no delay senior gen. 67.55 45.65*** 75.26 62.19***

(13.47) (4.91)

% no delay pre-uni. 67.57 51.06*** 60.66 68.27***

(15.07) (9.01)

Mean grade pre-voc. III/IV 6.36 6.13***

(0.25)

Mean grade senior gen. 6.44 6.03*** 6.52 6.16***

(0.26) (0.13)

Mean grade pre-uni. 6.57 6.08*** 6.62 6.35***

(0.33) (0.15)

Number of students 1579.97 823.49*** 788.02 1079.96***

(304.67) (301.03)

% disadvantaged pre-voc. III/IV 61.83 33.43***

(39.18)

% disadvantaged senior gen. 53.15 60.28** 7.31 6.97

(24.74) (14.34)

% disadvantaged pre-uni. 56.21 50.49** 3.49 4.73

(21.51) (12.40)Note: Columns (1) and (3) report school characteristics of school A and school B. Columns (2)and (4) report the means and standard deviations of each characteristic for the alternative schoolsattended by lottery losers in the first year of secondary education for school A and B separately.The information in the table is derived from the Quality Cards 2002-2010 of the Dutch educationinspectorate. The education inspectorate publishes secondary school information on a yearly basis.The measures published in the year of school choice are used for each lottery participating student.This creates an average weighted by the number of winning students in each year for the Montessorischools. For the alternative schools, this creates an average weighted by the number of attendinglottery losers in each year. The first seven rows describe school quality measures. The index forgrade progression in the lower grades is a measure of the percentage of students getting to the thirdgrade without grade retention, correcting for students who attend higher or lower school tracks thantheir primary school advice. The other measures pertain to the higher grades, and report on thepercentage of students passing through the higher grades without delay and the average grade on thefinal exams. The stars refer to the p-values of one-sample t-tests testing whether the means of thealternative schools equal the means for school A or school B, where the means of school A and B arethe constants. ***p<0.01 **p<0.05 *p<0.10.

71

higher) on time. Otherwise the indicator equals zero. Simple exam grades are not agood outcome measure, since students can change school track during secondary schooland exams are made for each school track separately. When a student changes to alower track, it is likely that his grades will be higher than they would have been at thehigher school track. Therefore, the school track advice of the primary school teacheris used as the school track the student is expected to obtain by the end of secondaryschool. This advice is compared to the obtained degree. When a student gets a degreefrom a lower track or when he is grade retained, he is coded as not obtaining his degreeon time. When a student follows the advised or a higher school track and obtains hisdegree without delay, he is counted as obtaining his degree on time.10 For students withcombined school track advices (e.g. senior general secondary/pre-university education)delays or lower degrees are counted from the lowest of the two tracks.11

To study whether Montessori education has an effect on gaining high grades, anindicator variable for obtaining a degree on time with at least a grade point averageof 7.5 (on a 10 point scale) is used. Further, indicators for obtaining a degree on timewith at least a passing grade for Dutch, English and mathematics are constructed.12

Since the youngest cohort of students started secondary education in 2011/2012,exam outcomes are only available for 47.9% of the lottery participating students in thesample. To investigate academic outcomes for a larger group of students, I considerstudents’ educational positions measured four years after participating in an admissionlottery. The fourth year is chosen because students are clearly tracked by the timethey are in 4th grade.13 In lower grades, students can be educated in combined classes,such as senior general secondary/pre-university education.

Two different outcome measures are used for the fourth year. The first one iswhether a student is in 4th grade at the advised school track or higher after fouryears of secondary education. Basically, this measure captures the students who areon track to obtain their degree on time: grade retained students are coded as zeros.Since parents and students might care more about the school track a student eventually

10It can also be that a student gets a higher degree with some delay. In that case, he is codedas obtaining his degree on time when he gets the higher degree at the same time as he could havegotten that degree if he first followed the school track of his primary school advice. For example, astudent with a senior general secondary education advice is considered to obtain his degree on time ifhe passes his pre-university exams in seven years.

11Counting from the highest of the two tracks does not change the results for the exam measures.For the 4th year, there is a slight difference, indicating that students at Montessori schools are moreoften at the lower track of the combined advice in 4th grade.

12To ensure that grades are comparable across schools, the national exam grades are used. A gradeof 5.5 or higher (on a 10 point scale) is considered a passing grade. It is possible for Dutch students tofail courses and still obtain their degree. Dutch secondary education has different levels of mathematicswithin the same school track. This distinction is not taken into account in the mathematics variable.Alternative specifications (computing the indicator for each math level separately and correcting mathgrades for difficulty following Leuven et al. (2010)) yield similar results.

13In this chapter, 4th grade refers to the 4th grade of Dutch secondary education, in which studentsare approximately 16 years old.

72

obtains instead of whether he is grade retained or not, grade retainers are not takeninto account in the second measure. This measure indicates whether students who arein 4th grade after four years are educated at their advised school track or higher.

As described in Section 4.3, students can choose a course program in the secondhalf of secondary education. The science program is considered the more difficultprogram in the senior general secondary and pre-university school tracks (Buser et al.,2014). Since (not) choosing a science program strongly affects the options for tertiaryeducation, it is interesting to see whether Montessori students are more or less oftenenrolled in the science program in the higher grades of the academic school tracks.

4.4.3 Questionnaire on socio-emotional functioning

To measure the effects of Montessori education on socio-emotional functioning, a ques-tionnaire was used. The content of the questionnaire is described in Table 4.3. Thetable includes information on the instruments used to measure each outcome, the reli-ability of these instruments and sample questions.

As independence is one of the main characteristics that Montessori education isclaimed to foster, the questionnaire includes two measures of independence. Indepen-dence at school measures the degree to which students are able to effectively work onschool tasks with classmates or on their own. As no existing instrument was available,it was developed for this study. The scale includes questions on taking initiative, work-ing together and evaluating the learning process. Independence out of school focussesmore broadly on functioning as an autonomous individual. It includes sub-scales fromexisting questionnaires on evaluative thinking, need for confirmation (comparative val-idation) and behavioral autonomy.

To study whether Montessori students are indeed more motivated for school, a scaleon motivation towards learning tasks is included. To get a broader view on students’socio-emotional functioning, the questionnaire also includes questions on well-being atschool, interest in society and on problematic behavior like truancy and use of alcoholand drugs. Further, the questionnaire includes items on parental support to schoolmatters and participating in commercial school support to investigate whether selectiveparental compensation diminishes potential differences in academic achievement.

For budgetary reasons, a sample of 910 students was invited to fill out the ques-tionnaire. The details of the sampling procedure are described in Appendix 4.A.1. Thestudents received a letter on university stationery, explaining that they applied for aMontessori secondary school some time ago and that the University of Amsterdam iscurious to see how their secondary school time is going or went. The letter containeda reference to a website, a personal login code and password, and students received areward of €10 when they completed the questionnaire.

73

Tab

le4.3:

Con

tent

andreliab

ilityof

thequ

estion

naire

Outcome

Instrument

Num

berof

Samplequ

estion

Cronb

ach’sα

question

sin

sample

Indepe

ndence

atscho

olDevelop

edforthis

stud

yi22

Ifthereis

something

Ido

n’tun

derstand

,Ifirst

look

for

0.68

anan

swer

myself,be

fore

turningto

theteacher

Indepe

ndence

outof

scho

olCASE

:ET,C

Vii,W

AS:

BA

iii

23Ithinkab

outho

wmyaction

swill

affectothers

0.75

Motivationtowards

learning

tasks

SVL2008:LG

iv8

Iwou

ldlik

eto

learnalotat

scho

ol0.83

Well-b

eing

atscho

olSV

L2008:PS,

RLi

v16

Igetalon

gwellw

ithmostof

myteachers

0.84

Interest

insociety

Develop

edforthis

stud

y7

Iconsider

itvery

impo

rtan

tto

vote

atelection

s0.58

Parentals

uppo

rtDevelop

edforthis

stud

y3

Mypa

rentsoftenhelp

mewithmyho

mew

ork

0.57

Note:

Colum

n5repo

rtsCronb

ach’sαstatistics

which

arecompu

tedusingthestud

ents

who

respon

dedto

thequ

estion

naire.

Cronb

ach’sαis

ameasure

for

relia

bility.

Itis

acoeffi

cientof

internal

consistency,

indicating

theextent

towhich

item

son

thesamescalearemeasuring

onecommon

characteristic.

i Som

eitem

sad

aptedfrom

L&S(V

orst,1

993),ii B

eckert

(2007),iii A

ndersonet

al.(1994),ivSm

its(2008).

74

609 students (67%), including 415 lottery participators, responded to the question-naire. Table 4.ii in Appendix 4.A.1 describes the characteristics of students who didand did not respond to the questionnaire. It turns out that nonrespondents are older,more often lottery losers and have lower school advices. Further, they are more oftenmales and non-western migrants. While it is unfortunate that there is selective non-response, the pattern of nonresponse is not uncommon (e.g. Van Loon et al., 2003;Søgaard et al., 2004). Moreover, the balancing tests reported in Table 4.4 indicate thatnonresponse does not create differences between lottery winners and lottery losers interms of background characteristics.

4.5 Empirical strategy

The effects of Montessori education are hard to identify, since the choice for a Montes-sori school is endogenous. Different parents choose for Montessori and regular educa-tion, and parental characteristics determining Montessori choice are likely to be relatedto student outcomes. The admission lotteries to school A and B, however, basicallycreate a series of randomized experiments on Montessori school attendance. To inves-tigate the effects of Montessori education, lottery participating students attending andnot attending school A and B are compared using:

yil = X′

ilβ + γMONTil + νl + εil (4.1)

where yil indicates outcomes y of student i participating in lottery l. X ′il is a vector

of student characteristics added for precision and includes gender, being a nonwesternor western migrant, student age, living in a disadvantaged neighborhood, cito testscore,making another test than cito, no information on primary school test-score and primaryschool advice. νl is a fixed effect for lottery l, capturing the different lotteries in differentyears and schools. When a school has multiple lotteries in a single year (for differentschool tracks or regions), a lottery fixed effect is included for each lottery. The lotteryfixed effects make sure that students within the same lottery are compared, instead ofcomparing students in lotteries with potentially slightly different characteristics. Sincethe lotteries are year and school specific, there are no year or school fixed effects. εil isthe error term. MONTil is an indicator variable for ever attending Montessori schoolA or B, making γ the parameter of interest.

Even in the sample of lottery participating students, ever attending a Montessorischool may be endogenous. Over time, some students who initially lose the lottery areoffered a place on the Montessori schools. To be specific, 43 students are present atschool A or B in some point of time after they lost the lottery, of whom 17 attend the

75

schools from the first year onwards, the always-takers.14 These students were on thewaiting list, and were placed when a slot came free. Not all students on the waiting listdecide to go to the Montessori schools when a slot comes free, some decide to stay attheir alternative secondary school. This choice may depend on student characteristics,making the indicator for ever attending a Montessori school potentially endogenous,which would bias γ. Therefore, an instrumental variables framework is used, using theoutcome of the lottery as an instrument for ever attending a Montessori school. Thefirst stage is defined as:

MONTil = X ′ilπ + λWIN il + υl + ηil (4.2)

where WIN il is an indicator for whether student i is a lottery winner. In thisframework, the instrument, WIN il, should be both relevant and valid. For the firstcondition, winning the lottery should predict Montessori school attendance. Appendix4.A.2 shows first stage results confirming that the lottery is indeed a relevant instru-ment. For the instrument to be valid, it should be uncorrelated with the error termεil. It is assumed that winning the lottery has no effect on student outcomes exceptfrom the effect via Montessori school attendance. This implies, for example, that dis-appointment about losing the lottery should not lead to a decrease in motivation forschool. Further, the outcome of the lottery should be randomly determined, which isvery likely since the lotteries are executed by a notary.

To show that the lotteries are indeed conducted fairly, Table 4.4 shows balancingtests testing whether there are differences in observable characteristics between lotterywinners and lottery losers. Columns 1 and 2 show the means and standard deviationsfor students losing and winning a school admission lottery. Columns 3 and 4 showthe actual balancing tests, regressing the dependent variables denoted in each row onan indicator for winning the school admission lottery. The balancing tests show thatfor nearly all characteristics, there are no significant differences between winning andlosing students. The exception is age: winning students are somewhat older thanlosing students, although the coefficient is small. Table 4.4 reports separate regressioncoefficients to facilitate interpretation. To test whether the coefficients are jointlysignificantly different from zero, a joint balancing test was conducted. Regressing theindicator for winning the lottery on all background characteristics, it turns out thatthe null hypothesis cannot be rejected (F(10,774)=0.781, p=0.647), supporting theconclusion that the lotteries were indeed conducted fairly.

Since not every lottery participating student participated in the questionnaire,columns 5 and 6 show similar balancing tests for winning and losing students who

14Reversely, no never-takers are found: all students who won the lottery are present at the Montes-sori schools for at least one year.

76

responded to the questionnaire. The results are comparable to the results of the pre-vious balancing tests15, indicating that winning and losing students who responded tothe questionnaire are similar in terms of background characteristics.

4.6 Results

4.6.1 The effects of Montessori education on academic out-

comes

The results in Table 4.5 indicate that there is little impact of Montessori education onstudents’ academic achievement. The first row shows that after four years of secondaryeducation, 75.2% of the students is in 4th grade of the advised track or better. Onthis outcome, there is no significant difference between Montessori and non-Montessoristudents. When a student is not in the advised track in 4th grade after four years ofsecondary education, this can have two different causes: it can be that the student isgrade retained, or that he has moved to a lower school track.

The second line of results indicates that Montessori and non-Montessori studentsdiffer in these causes: when grade retained students are dropped from the sample, itturns out that Montessori students who are in 4th grade after four years more oftenattend a lower school track than can be expected from their primary school advice.This indicates that students at the Montessori schools are more often sent to a lowerschool track, while similar students in traditional schools are grade retained.16

More important than fourth year outcomes are the final exam results. It turns outthat Montessori students obtain the degree of their advised school track on time at thesame rate as non-Montessori students, which matches the results on the first row ofTable 4.5. Since Montessori students are more often educated at lower school tracksin 4th grade, one might expect fewer Montessori students to obtain the degree of theiradvised school track with one year delay: students at a lower track may finalize theirsecondary education at that lower school track, while grade retainers obtain the degreeof their initial track with one year delay. This conjecture is not seconded by the results:Montessori education has no significant effect on obtaining the degree of the advised

15The null hypothesis of the joint balancing test cannot be rejected either (F(10,389)=1.001,p=0.441).

16Both Montessori schools indeed have an avoiding policy towards grade retention. Another po-tential explanation is that students who lose the lottery go to schools that only offer higher schooltracks. To avoid changing schools, students at these schools might prefer grade retention over movingdown a track. Indeed, 41.8% of the lottery losing students at school A attend a school that does notoffer the pre-vocational IV school track. School B, however, only offers the two highest tracks, and77.4% of their lottery losing students attend schools with the same school tracks, while 22.6% attendschools that also offer lower, pre-vocational, tracks, implying that this mechanism should not operatein school B. As shown in Table 4.v, the fourth year outcomes are not significantly different for schoolA and school B, suggesting that school policies are more relevant than avoiding to change schools.

77

Tab

le4.4:

Descriptive

statistics

andlotterybalan

ceDep

endent

Mean(SD)for

Mean(SD)for

Balan

cing

P-values

Balan

cing

tests

P-values

variab

lelotterylosers

lotterywinners

testslottery

lotteryfor

question

naire

respon

dents

(1)

(2)

(3)

(4)

(5)

(6)

Citoscore

540.438

541.517

0.240

0.413

0.217

0.565

(5.229)

(4.380)

(0.293)

(0.376)

Pre-voc.IV

0.231

0.084

0.010

0.620

0.006

0.805

prim

aryscho

olad

vice

(0.422)

(0.277)

(0.020)

(0.025)

Pre-voc.IV

/seniorgen.

0.109

0.052

-0.011

0.567

-0.015

0.527

prim

aryscho

olad

vice

(0.312)

(0.223)

(0.019)

(0.024)

Senior

gen.

0.231

0.271

0.005

0.900

0.067

0.152

prim

aryscho

olad

vice

(0.422)

(0.445)

(0.038)

(0.047)

Senior

gen./p

re-uni.

0.224

0.307

0.008

0.825

-0.032

0.482

prim

aryscho

olad

vice

(0.418)

(0.462)

(0.036)

(0.045)

Pre-uni.

0.206

0.286

-0.012

0.730

-0.026

0.564

prim

aryscho

olad

vice

(0.405)

(0.452)

(0.034)

(0.044)

Boy

0.315

0.367

0.028

0.504

0.016

0.756

(0.465)

(0.483)

(0.042)

(0.052)

Dutch

0.685

0.766

-0.016

0.662

-0.055

0.218

(0.465)

(0.424)

(0.037)

(0.045)

Non

-western

migrant

0.209

0.132

-0.003

0.914

0.047

0.200

(0.407)

(0.338)

(0.030)

(0.037)

Western

migrant

0.106

0.102

0.019

0.450

0.008

0.802

(0.308)

(0.303)

(0.026)

(0.031)

(Tab

leco

ntin

ues

onne

xtpa

ge)

78

Tab

le4.4:

(con

tinu

ed)

Dep

endent

Mean(SD)for

Mean(SD)for

Balan

cing

P-values

Balan

cing

tests

P-values

variab

lelotterylosers

lotterywinne

rstestslottery

lotteryfor

question

naire

respon

dents

(1)

(2)

(3)

(4)

(5)

(6)

Age

atap

plication

12.087

12.109

0.079

0.047

0.082

0.085

(0.446)

(0.415)

(0.040)

(0.048)

Disad

vantaged

0.542

0.315

-0.023

0.559

-0.020

0.683

neighb

orho

od(0.499)

(0.465)

(0.039)

(0.049)

Num

berof

stud

ents

321

479

800

415

Note:

Colum

ns(1)an

d(2)displaythemeans

andstan

dard

deviations

forstud

ents

losing

andwinning

alotteryforaMon

tessorischo

ol.Colum

ns(3)an

d(4)

repo

rtsepa

rate

regression

coeffi

cients

andthep-values

ofthevariab

lesindicatedin

each

row

onan

indicatorvariab

leequa

lling

0ifthestud

entlost

thelottery

andequa

lling

1ifthestud

entwon

thelottery.

The

sampleis

restricted

tostud

ents

participatingon

thelottery.

Colum

ns(5)an

d(6)repo

rtsimila

rregression

coeffi

cients

forthesampleof

lotterypa

rticipatingstud

ents

who

respon

dedto

thequ

estion

naire.

Allregression

sinclud

elotteryfix

edeff

ects.Rob

uststan

dard

errors

arerepo

rted

inpa

rentheses.

79

Table 4.5: Results on academic outcomesOutcome Mean (SD) N (1) (2)

In 4th grade of advised track or better 0.752 625 -0.049 -0.022

after four years (0.432) (0.048) (0.047)

In 4th grade of advised track or better 0.895 525 -0.079** -0.080**

after four years (cond. on being in 4th grade) (0.307) (0.038) (0.037)

Obtained degree on time 0.671 383 -0.049 0.000

on advised track or higher (0.470) (0.064) (0.060)

Obtained degree on advised track or higher 0.777 376 -0.017 0.008

with at most one year delay (0.417) (0.058) (0.057)

Obtained degree on time on advised track 0.047 381 -0.035 -0.036

with at least a 7.5 on average (0.212) (0.030) (0.030)

Obtained degree on time on advised track 0.517 377 -0.047 -0.015

with passing grade for Dutch (0.500) (0.069) (0.068)

Obtained degree on time on advised track 0.561 378 0.022 0.057

with passing grade for English (0.497) (0.069) (0.068)

Obtained degree on time on advised track 0.500 308 -0.035 0.029

with passing grade for Math (0.501) (0.078) (0.075)

Enrolled in science program 0.184 483 -0.069 -0.030

(0.388) (0.053) (0.052)

Parental support 8.664 408 0.441 0.406

(2.406) (0.302) (0.308)

Participated in commercial 0.491 407 0.115* 0.112*

school support (0.501) (0.064) (0.066)

Controls !

Note: Each row reports two IV regressions with winning the lottery as an IV for ever attendinga Montessori school. The sample is restricted to students participating on the lottery. The firstcolumn displays the means and standard deviations for these lottery participating students. Columns(1) and (2) report IV estimates without and with controls. Robust standard errors are reported inparentheses. Alternative specifications of the standard errors (e.g. clustering on school attended inthe first year or clustering on school attended in the first year*school advise*year) yield similar results.All regressions include lottery fixed effects. Controls include gender, being a nonwestern or westernmigrant, student age, living in a disadvantaged neighborhood, cito testscore, making another test thancito, no information on primary school test-score and primary school advice. Exams are graded on a10-point scale, where grades above 5.5 are considered a pass. ***p<0.01 **p<0.05 *p<0.10.

80

track with one year delay. It might be that Montessori students catch up by doing thehigher school track after their first exams, which also yields a degree with one yeardelay.

With respect to grades, it turns out that Montessori education has no significantimpact on obtaining a degree with high average grades, nor on obtaining a degree withpassing grades for Dutch, English and Math. Further, Montessori education does notaffect the probability of being enrolled in a science program.

It is conceivable that the limited differences in academic outcomes in either directionare created by selective parental compensation. Parents might send badly achievingstudents to commercial school support activities (such as private tutoring) or decideto invest more time themselves, while they do not interfere when a student is doingwell. The results in Table 4.5 show that there is no significant difference in the amountof parental support between Montessori and non-Montessori students. Montessori stu-dents do, however, receive more commercial school support, although the coefficient isonly significant at the 10% level.

An alternative explanation is that the coefficients are imprecisely estimated. Look-ing at the signs of the coefficients, it should be noted that all coefficients on additionalsupport are positive, while 13 out of the 18 coefficients for the academic outcomesare negative. Since the standard errors are rather large, the possibility of substantialnegative effects cannot be excluded.

Overall, the results indicate that Montessori students achieve just as well as studentsattending regular secondary education.

4.6.2 The effects of Montessori education on socio-emotional

outcomes

While the previous section focussed on academic achievement, the results in Table4.6 focus on socio-emotional functioning. Even though Montessori education aims toimprove students’ independence and motivation, the results show no positive effects ofMontessori education on these outcomes. There are no significant differences betweenMontessori and non-Montessori students on the measures of independence at school andmotivation towards learning tasks. Contrary to the expectations, Montessori educationhas a negative impact on independence out of school. Exploring the sub-scales (resultsnot shown), it turns out that this result is mainly caused by the sub-scale of evaluativethinking. Evaluative thinking measures the extent to which students think about theiractions and decisions before acting.

On the other measures of socio-emotional functioning, there are few significanteffects of Montessori education. The exception is well-being at school, which indicatesthat Montessori students enjoy their school better and have a better relationship with

81

Table 4.6: Results on socio-emotional outcomesOutcome Mean (SD) N (1) (2)

Independence at school 73.520 408 -0.008 0.095

(7.619) (0.917) (0.938)

Independence out of school 83.475 408 -2.253** -1.923**

(7.936) (0.975) (0.974)

Motivation towards learning tasks 17.895 409 -0.369 -0.269

(4.494) (0.552) (0.555)

Well-being at school 37.875 409 1.302 1.462*

(6.570) (0.823) (0.830)

Interest in society 23.627 408 0.770 0.660

(3.994) (0.484) (0.506)

Number of times being late, truant or sent 21.593 407 -3.022 -5.833

out of class in the past 6 months (39.622) (4.019) (4.297)

Smoked cigarettes or drank alcohol 0.597 407 0.084 0.082

in the past month (0.491) (0.052) (0.054)

Used drugs of any kind 0.373 407 0.048 0.014

in the past 12 months (0.484) (0.057) (0.058)

Ever got a fine or was suspected of a 0.270 407 0.041 0.024

criminal offense by the police (0.445) (0.054) (0.051)

Controls !

Note: Each row reports two IV regressions with winning the lottery as an IV for ever attendinga Montessori school. The sample is restricted to students participating on the lottery. The firstcolumn displays the means and standard deviations for these lottery participating students. Columns(1) and (2) report IV estimates without and with controls. Robust standard errors are reported inparentheses. Alternative specifications of the standard errors (e.g. clustering on school attended inthe first year or clustering on school attended in the first year*school advise*year) yield similar results.All regressions include lottery fixed effects. Controls include gender, being a nonwestern or westernmigrant, student age, living in a disadvantaged neighborhood, cito testscore, making another test thancito, no information on primary school test-score and primary school advice. ***p<0.01 **p<0.05*p<0.10.

82

their teachers. Montessori education does not significantly affect problematic behaviornor students’ interest in society.

Including the academic outcomes, 20 different variables were tested in the analysesabove. At a 5% significance level, you would expect one variable to turn out signifi-cantly by chance: a Type I error. When applying the Bonferroni and the Bonferroni-Holm method to control for multiple testing, none of the null hypotheses can be re-jected. Even though the Bonferroni and Bonferroni-Holm methods are very conserva-tive, the results should be interpreted with caution. The main message to take out ofthe results is that the outcomes of Montessori students are similar to the outcomes ofnon-Montessori students.

4.6.3 Heterogeneous effects

The results presented so far are estimates based on two Montessori secondary schools.Although both schools are Montessori schools, it might be that one school attainsbetter outcomes than the other. Table 4.v in Appendix 4.A.3 shows heterogeneouseffects for school A and school B on the main outcomes considered in this study.17

It turns out that the effects on academic achievement after four years of secondaryeducation are not significantly different between the two schools.18 The finding thatMontessori students more often participate in commercial school support, however, isdriven by students at school B.

With respect to socio-emotional functioning, independence at school has the onlysignificant interaction effect. It indicates that students educated at school B scoresignificantly lower on independence at school than their non-Montessori counterpartswhile no such effect is found for students participating in school admission lotteries forschool A. Since 12 out of 14 interaction effects are insignificant, the general conclusionis that the effects of Montessori education are similar for school A and school B.

Next to differences between schools, there might be differential effects of Montessorieducation for different groups of students. For example, it could be that girls do betterat Montessori schools, since girls are generally more able to work independently thanboys. For the same reason, it might also be that students in the higher school tracksdo better in Montessori education than students in the lower school tracks.

Tables 4.vi and 4.vii of Appendix 4.A.3 show additional differential effects of Montes-sori education for boys and girls and for different school tracks. Regarding gender, Table

17The interaction effects for the other outcome measures (on grades and problematic behavior) areinsignificant for school A and school B and for boys and girls. For the different school tracks, itturns out that compared to senior general secondary/pre-university students, the pre-vocational IVstudents in Montessori education more often obtain a passing grade on English and Math. Further,these students are less often late, truant or sent out of class.

18Differential effects on exam measures are not available because the students participating inadmission lotteries for school B are too young for secondary school exams.

83

4.vi shows that none of the interaction effects is significant at the 5% level. Therefore,the null-hypothesis of equal effects of Montessori education for boys and girls cannotbe rejected, which implies that the effects of Montessori education are similar for boysand girls.

For the different school tracks, some of the interaction effects are significant. Theeffects indicate that Montessori education has a more positive impact on pre-vocationalIV students than on students at the academic tracks. Compared to senior generalsecondary/pre-university students, pre-vocational IV students more often obtain theirdegree on time on the advised track or higher. The main effects on this variable arenot significant. Further, the positive effect of Montessori education on well-being atschool seems to be driven by pre-vocational IV students.

4.6.4 Difference between lottery participants and students with

priority

As mentioned in Section 4.4.1, not every student attending the Montessori schoolsparticipated in a lottery. Both schools yield priority to students having brothers orsisters at the school and to children of employees. Moreover, school B conducts lotteriesbased on students’ residence, while school A yields priority to students from Montessoriprimary schools.

It might be that students who got placed with priority are different from lotteryparticipating students. Table 4.7 shows a-priori differences between students who par-ticipated in the lottery and students who got placed with priority. The table shows thaton most background characteristics, there are no differences between lottery participa-tors and students with priority. The differences that are found indicate that comparedto students with priority, students who participate in the lottery are less often boy andmore often have a non-western migrant background.

More important than background characteristics is the fact that school A yieldspriority to students from Montessori primary schools. As the majority of the lotteryparticipating students subscribed for school A, the effects of Montessori education aremainly identified on a sample of students who did not attend Montessori primaryschools. It is possible that the effects of Montessori secondary education are morepositive for students who were educated in Montessori primary schools: students fromMontessori primary schools will have more experience with independent working andchoice in their school tasks.

Table 4.8 reports differences in the main outcomes between Montessori students whowon the lottery and students who were placed with priority. It focusses on winningstudents as students who lost the lottery are generally educated at regular educationschools. Although the results are descriptive, the results in the table indicate that the

84

Table 4.7: Differences in background characteristics between students par-ticipating on the lottery and students being placed with priorityDependent Mean (SD) for Balancing test P-values

variable students with priority lottery participation

(1) (2) (3)

Citoscore 541.0 0.128 0.497

(5.421) (0.189)

Pre-voc. IV 0.107 0.012 0.328

primary school advice (0.310) (0.012)

Pre-voc. IV/senior gen. 0.082 -0.019 0.094

primary school advice (0.274) (0.012)

Senior gen. 0.248 0.031 0.127

primary school advice (0.432) (0.021)

Senior gen./pre-uni. 0.280 0.011 0.607

primary school advice (0.449) (0.020)

Pre-uni. 0.283 -0.034 0.082

primary school advice (0.451) (0.020)

Boy 0.408 -0.081 0.001

(0.492) (0.024)

Dutch 0.744 -0.029 0.184

(0.437) (0.022)

Non-western migrant 0.132 0.039 0.031

(0.338) (0.018)

Western migrant 0.125 -0.010 0.502

(0.331) (0.016)

Age at application 12.086 -0.003 0.900

(0.439) (0.021)

Disadvantaged 0.501 0.018 0.432

neighborhood (0.500) (0.023)

Number of students 1322 2122Note: Column (1) displays the means and standard deviations for students who got placed withpriority. Columns (2) and (3) report separate regression coefficients and the p-values of the dependentvariables indicated in each row on an indicator variable equalling 0 if the student was placed withpriority and equalling 1 if the student participated in the lottery. The regressions are on the full sample,including all students who participated in the lottery and all students who got placed with priority.All regressions include lottery fixed effects. Robust standard errors are reported in parentheses.

85

outcomes of students with priority are not significantly different from the outcomes ofstudents who won a school admission lottery.19

As the priority rules for school A are likely to create a larger difference betweenpriority and non-priority students, it is interesting to note that the results are similarwhen only focussing on students at school A (results not shown). This implies thatMontessori students who were educated at regular primary schools achieve similar totheir secondary education peers who were educated at Montessori primary schools.Since students from Montessori primary schools do not get priority in school B, furthersupport on this observation is provided by Table 4.v. This table shows that the effectsof Montessori education are similar for students in school A and school B.

Overall, the results in this subsection indicate that it is unlikely that the results onthe effects of Montessori secondary education are driven by the priority rules.

4.7 Conclusions

In this study, the causal effects of Montessori education are investigated by exploit-ing school admission lotteries for two Montessori secondary schools. It is found thatMontessori education has little impact on students’ academic achievement. Montes-sori students obtain their secondary school degree without delay at the same rate andwith similar grades as non-Montessori students, although the route towards the examsis somewhat different. After four years of secondary education, Montessori studentsmore often attend a lower school track while non-Montessori students are more oftengrade-retained.

On most socio-emotional outcomes, there is no significant effect of Montessori ed-ucation. Most notably, there are no positive effects of Montessori education on in-dependence and motivation, although these are the main characteristics Montessorieducation claims to foster. Regarding independence out of school, Montessori studentseven score worse than their non-Montessori counterparts.

As the results in this study are based on two Montessori schools, one might worrythat the results cannot be generalized to other Montessori secondary schools. It isreassuring that both schools in this study yield similar outcomes. Still, the results arebased on Dutch Montessori secondary schools that conduct school admission lotteries.Since the demand for these schools is high, it might be that these are especially highquality Montessori programs. In that case, the results can be interpreted as the effectsof Montessori education in high quality Montessori programs.

Montessori secondary education programs are less standardized than Montessoriprimary education programs, and the results of this study should not be generalized

19The results for the other outcome measures (on grades and problematic behavior) are insignificantas well.

86

Table 4.8: Differences in outcomes between students winning the lotteryand students being placed with priorityOutcome (1)

Achievement

In 4th grade of advised track or better after four years -0.050*

(0.029)

In 4th grade of advised track or better after four years (cond. on being in 4th grade) -0.029

(0.024)

Obtained degree on time on advised track or higher -0.031

(0.039)

Obtained degree on advised track or higher with at most one year delay -0.027

(0.034)

Enrolled in science program 0.020

(0.028)

Parental support 0.263

(0.257)

Participated in commercial school support 0.064

(0.054)

Socio-emotional functioning

Independence at school -0.992

(0.856)

Independence out of school -0.195

(0.855)

Motivation towards learning tasks -0.531

(0.474)

Well-being at school -0.123

(0.704)

Interest in society -0.164

(0.441)Note: Each coefficient reports an OLS regression regressing the dependent variable in that row on anindicator equalling 1 if the student participated in the lottery. The regressions include the students whogot placed with priority and the students who won the lottery for the Montessori school. The numberof students varies from 807 to 1381 for the academic outcomes and from 401 to 403 on the variablesmeasured by the questionnaire. Robust standard errors are reported in parentheses. All regressionsinclude lottery fixed effects and student controls. Controls include gender, being a nonwestern orwestern migrant, student age, living in a disadvantaged neighborhood, cito testscore, making anothertest than cito, no information on primary school test-score and primary school advice. Exams aregraded on a 10-point scale, where grades above 5.5 are considered a pass. ***p<0.01 **p<0.05*p<0.10.

87

to the small number of rather radical Montessori boarding schools with ’Erdkinder’programs. The schools considered in this study are, however, part of a long traditionof Montessori secondary education. Montessori secondary schools in other countriesare based on the same philosophy and have similar approaches with respect to choiceof activities and field projects.

Another issue is that the results are based on students applying for Montessorischools. It is likely that this is a selective sample of students, as students at Montessorischools are generally of higher socioeconomic status and less often from ethnic minori-ties (Dohrmann et al., 2007; Rathunde & Csikszentmihalyi, 2005a). It is possible thatthe effects of Montessori education are different for less advantaged groups of students.Further, the results for one of the schools are based on students who did not attendMontessori primary schools. From a critic’s point of view, you could argue that ifMontessori education has a negative impact, the effects might be especially negativefor students who did not attend Montessori primary schools. These students are un-familiar with the Montessori education system and are probably not used to a largeamount of choice and independent working. The outcomes of these students, however,turn out to be similar to the outcomes of the Montessori secondary education studentswho were educated at a Montessori primary school.

Despite its limitations, this is the first study to investigate the causal effects ofMontessori secondary education. Overall, the results of this study indicate that Montes-sori education provides an alternative route to similar student outcomes. This is im-portant information for parents, students and policy makers, as it does not supportpopular claims on the effects of Montessori education. Critics are not supported in theirconcern that Montessori education has a negative impact on student achievement. Foradvocates, on the other hand, there is no evidence that Montessori education has apositive effect on independence and motivation.

88

4.A Appendix

4.A.1 Details on questionnaire sampling and response

In this appendix, the sampling procedure for the questionnaire and the questionnaireresponse are described. Because of the large difference in the number of lottery losingstudents, a different sampling procedure was used for school A and school B. Table4.i reports the number of lottery winners, losers and priority students selected for thequestionnaire in each year, school and school track.

For school A, all 540 students participating in the school admission lotteries wereselected to receive the questionnaire. Besides that, a stratified random sample of 290priority students was drawn. This sample is stratified by school year and school track,weighted on the number of lottery winners plus 10%.

Since the number of students losing the lottery is much lower in school B, the ques-tionnaire sample at school B was restricted to lottery participating students. Because

Table 4.i: Number of students selected for the questionnaireNo questionnaire Questionnaire

School year Lottery winner Priority Lottery winner Lottery loser Priority Total

School A: Senior general secondary / pre-university education

2003/2004 114 50 46 55 265

2004/2005 184 16 59 19 278

2005/2006 141 47 9 55 252

2009/2010 117 37 42 46 242

2010/2011 145 29 17 33 224

Total 701 179 173 208 1261

School A: Pre-vocational IV education

2003/2004 34 17 45 19 115

2004/2005 47 10 30 13 100

2007/2008 52 10 29 10 101

2008/2009 9 29 3 33 74

2009/2010 36 5 10 7 58

Total 178 71 117 82 448

School B: Senior general secondary/pre-university education

2008/2009 28 60 24 15 127

2009/2010 77 16 6 4 103

2010/2011 25 43 13 9 90

2011/2012 50 34 6 3 93

Total 180 153 49 31 413

Grand Total 180 1032 299 321 290 2122Note: This table reports the number of students selected for the questionnaire by school, school track,year and lottery status.

89

the number of winning students is much larger than the number of losing students, aweighted random sample of winning students was used. This sample is weighted on theyearly number of lottery losers plus 40%, with a minimum of 6 students. This way, atotal of 80 students applying to school B were selected for the questionnaire, bringingthe total number of students selected for the questionnaire to 910.20

To protect the privacy of the students, the invitations for the questionnaire weresent out via the Dutch educational administration. Four different letters were sent: forMontessori and non-Montessori students and for students above and under 16. The lastdistinction was made because parental permission is required for research participationby students under 16 years of age. 505 students were sent the letter for Montessoristudents of 16 years and older, 98 students got the letter for Montessori students under16. 268 students got the letter for lottery losing students of 16 years and older, 31students got the letter for lottery losing students under 16.

The online questionnaire was filled out by 609 students (67%), including 415 lotteryparticipators. As described in Section 4.4.3, there was selective nonresponse to thequestionnaire. Characteristics of responders and non responders are summarized inTable 4.ii.

20In fact, the original number of questionnaires requested to be sent out was 915 instead of 910. 5students were dropped from the sample later on because of missing values on ethnicity and primaryschool advice. It turned out that 13 students did not have accurate address information, so 902 letterswere sent. Which students have inaccurate address information is unknown.

90

Table 4.ii: Selectivity in nonresponse to the questionnaireNon-responders Responders P-value

School B 0.053 0.105 0.009

Admission for year 2003/2004 0.359 0.204 0.000

Admission for year 2004/2005 0.216 0.135 0.002

Admission for year 2005/2006 0.116 0.125 0.712

Admission for year 2007/2008 0.057 0.053 0.805

Admission for year 2008/2009 0.083 0.130 0.037

Admission for year 2009/2010 0.093 0.212 0.000

Admission for year 2010/2011 0.076 0.128 0.020

Admission for year 2011/2012 0.000 0.015 0.034

Lottery participant 0.681 0.681 0.991

Lottery loser 0.405 0.327 0.020

Citoscore 540.2 540.8 0.115

Pre-voc. IV primary school advice 0.263 0.130 0.000

Pre-voc. IV/senior gen. primary school advice 0.100 0.113 0.535

Senior gen. primary school advice 0.249 0.227 0.450

Senior gen./pre-uni. primary school advice 0.223 0.271 0.116

Pre-uni. primary school advice 0.166 0.259 0.002

Boy 0.392 0.322 0.036

Dutch 0.651 0.723 0.027

Non-western migrant 0.219 0.164 0.043

Western migrant 0.130 0.113 0.476

Age at application 12.166 12.071 0.002

Disadvantaged neighborhood 0.535 0.498 0.290

Number of students 301 609Note: The first two columns report the mean of questionnaire non-responders and questionnaireresponders for the characteristic mentioned in that row. The p-values refer to two-group mean com-parison t-tests comparing students who did and did not respond on the questionnaire.

91

4.A.2 First stage coefficients

Tables 4.iii and 4.iv report the first stage coefficients belonging to the IV regressionsin Tables 4.5 and 4.6. For all variables, the first stage coefficients are large (between0.826 and 0.858) and significant at the 1% level. The partial F-statistics are high aswell, ranging from 492.78 to 1083.18, well above the rule of thumb of a minimum of10. These results confirm that the lottery is indeed a relevant instrument for presenceat the Montessori secondary school.

Table 4.iii: First stages achievementOutcome (1) (2)

In 4th grade of advised track or better 0.844*** 0.843***

after four years (0.026) (0.026)

In 4th grade of advised track or better 0.848*** 0.848***

after four years (cond. on being in 4th grade) (0.027) (0.028)

Obtained degree on time 0.857*** 0.858***

on advised track or higher (0.029) (0.031)

Obtained degree on advised track or higher 0.867*** 0.869***

with at most one year delay (0.027) (0.030)

Obtained degree on time on advised track 0.856*** 0.858***

with at least a 7.5 on average (0.029) (0.031)

Obtained degree on time on advised track 0.856*** 0.857***

with passing grade for Dutch (0.029) (0.031)

Obtained degree on time on advised track 0.856*** 0.857***

with passing grade for English (0.029) (0.031)

Obtained degree on time on advised track 0.837*** 0.838***

with passing grade for Math (0.034) (0.038)

Enrolled in science program 0.845*** 0.846***

(0.029) (0.030)

Parental support 0.835*** 0.827***

(0.030) (0.031)

Participated in commercial 0.834*** 0.826***

school support (0.030) (0.031)

Controls !

Note: Each row reports two sets of first stages of IV regressions with winning the lottery as an IVfor ever attending a Montessori school. The sample is restricted to students participating on thelottery. Robust standard errors are reported in parentheses. All regressions include lottery fixedeffects. Controls include gender, being a nonwestern or western migrant, student age, living in adisadvantaged neighborhood, cito testscore, making another test than cito, no information on primaryschool test-score and primary school advice. ***p<0.01 **p<0.05 *p<0.10.

92

Table 4.iv: First stages socio-emotional functioningOutcome (1) (2)

Independence at school 0.835*** 0.827***

(0.030) (0.031)

Independence out of school 0.835*** 0.827***

(0.030) (0.031)

Motivation towards learning tasks 0.835*** 0.827***

(0.030) (0.031)

Well-being at school 0.835*** 0.827***

(0.030) (0.031)

Interest in society 0.835*** 0.827***

(0.030) (0.031)

Number of times being late, truant or sent 0.834*** 0.826***

out of class in the past 6 months (0.030) (0.031)

Smoked cigarettes or drank alcohol 0.834*** 0.826***

in the past month (0.030) (0.031)

Used drugs of any kind 0.834*** 0.826***

in the past 12 months (0.030) (0.031)

Ever got a fine or was suspected of a 0.834*** 0.826***

criminal offense by the police (0.030) (0.031)

Controls !

Note: Each row reports two sets of first stages of IV regressions with winning the lottery as an IVfor ever attending a Montessori school. The sample is restricted to students participating on thelottery. Robust standard errors are reported in parentheses. All regressions include lottery fixedeffects. Controls include gender, being a nonwestern or western migrant, student age, living in adisadvantaged neighborhood, cito testscore, making another test than cito, no information on primaryschool test-score and primary school advice. ***p<0.01 **p<0.05 *p<0.10.

93

4.A.3 Heterogeneous effects

Table 4.v: Differential results for school A and school BTreatment Interaction Treatment

Outcome effect school A effect effect school B

(1) (2) (3)

Achievement

In 4th grade of advised track or better -0.027 0.036 0.010

after four years (0.049) (0.136) (0.131)

In 4th grade of advised track or better -0.076* -0.038 -0.114***

after four years (cond. on being in 4th grade) (0.041) (0.055) (0.044)

Enrolled in science program -0.018 -0.082 -0.100

(0.052) (0.161) (0.155)

Parental support 0.329 0.473 0.801

(0.343) (0.716) (0.642)

Participated in commercial 0.048 0.386*** 0.435***

school support (0.072) (0.147) (0.132)(Table continues on next page)

94

Table 4.v: (continued)Treatment Interaction Treatment

Outcome effect school A effect effect school B

(1) (2) (3)

Socio-emotional functioning

Independence at school 0.879 -4.778** -3.899**

(1.048) (2.165) (1.913)

Independence out of school -1.623 -1.828 -3.451*

(1.082) (2.186) (1.948)

Motivation towards learning tasks -0.094 -1.061 -1.155

(0.627) (1.220) (1.054)

Well-being at school 1.827* -2.223 -0.396

(0.941) (1.757) (1.503)

Interest in society 0.763 -0.628 0.135

(0.557) (1.169) (1.058)

Number of times being late, truant or sent -7.131 7.880 0.749

out of class in the past 6 months (5.008) (6.700) (4.655)

Smoked cigarettes or drank alcohol 0.062 0.121 0.183

in the past month (0.060) (0.129) (0.116)

Used drugs of any kind 0.016 -0.012 0.004

in the past 12 months (0.067) (0.099) (0.075)

Ever got a fine or was suspected of a 0.025 -0.008 0.017

criminal offense by the police (0.059) (0.097) (0.076)Note: Each row reports two IV regressions with winning the lottery as an IV for ever attending aMontessori school. Columns (1) and (2) display the treatment effect for school A and the interactioneffect on school B. Column (3) is another IV regression describing the treatment effect for school B.The sample is restricted to students participating on the lottery. Differential effects on exam measuresare not available because the students participating in admission lotteries for school B are too youngto have done secondary school exams. Robust standard errors are reported in parentheses. Allregressions include lottery fixed effects and controls. Controls include gender, being a nonwestern orwestern migrant, student age, living in a disadvantaged neighborhood, cito testscore, making anothertest than cito, no information on primary school test-score and primary school advice. Exams aregraded on a 10-point scale, where grades above 5.5 are considered a pass. ***p<0.01 **p<0.05*p<0.10.

95

Table 4.vi: Differential results for boys and girlsTreatment Interaction Treatment

Outcome effect girls effect effect boys

(1) (2) (3)

Achievement

In 4th grade of advised track or better 0.016 -0.115 -0.099

after four years (0.054) (0.083) (0.074)

In 4th grade of advised track or better -0.058 -0.073 -0.131**

after four years (cond. on being in 4th grade) (0.042) (0.064) (0.057)

Obtained degree on time 0.010 -0.030 -0.020

on advised track or higher (0.071) (0.118) (0.101)

Obtained degree on advised track or higher 0.017 -0.029 -0.012

with at most one year delay (0.064) (0.110) (0.098)

Enrolled in science program -0.061 0.095 0.034

(0.053) (0.095) (0.092)

Parental support 0.478 -0.215 0.264

(0.366) (0.591) (0.500)

Participated in commercial 0.164** -0.153 0.011

school support (0.080) (0.125) (0.103)

Socio-emotional functioning

Independence at school 0.383 -0.858 -0.476

(1.132) (1.884) (1.559)

Independence out of school -1.148 -2.308 -3.456**

(1.143) (2.031) (1.721)

Motivation towards learning tasks 0.404 -2.003* -1.599*

(0.657) (1.083) (0.910)

Well-being at school 1.815* -1.051 0.764

(1.044) (1.616) (1.271)

Interest in society 0.320 1.011 1.331

(0.593) (1.006) (0.866)Note: Each row reports two IV regressions with winning the lottery as an IV for ever attending aMontessori school. Columns (1) and (2) display the treatment effect for girls and the interactioneffect on boys. Column (3) is another IV regression describing the treatment effect for boys. Thesample is restricted to students participating on the lottery. Robust standard errors are reportedin parentheses. All regressions include lottery fixed effects and controls. Controls include gender,being a nonwestern or western migrant, student age, living in a disadvantaged neighborhood, citotestscore, making another test than cito, no information on primary school test-score and primaryschool advice. Exams are graded on a 10-point scale, where grades above 5.5 are considered a pass.***p<0.01 **p<0.05 *p<0.10.

96

Table 4.vii: Differential results for pre-vocational IV and academic studentsTreatment Interaction Treatment effect

Outcome effect pre-voc. IV effect senior gen./pre-uni.

(1) (2) (3)

Achievement

In 4th grade of advised track or better 0.048 -0.103 -0.054

after four years (0.084) (0.101) (0.057)

In 4th grade of advised track or better -0.039 -0.060 -0.099**

after four years (cond. on being in 4th grade) (0.073) (0.085) (0.043)

Obtained degree on time 0.142 -0.248** -0.106

on advised track or higher (0.092) (0.120) (0.079)

Obtained degree on advised track or higher 0.079 -0.124 -0.044

with at most one year delay (0.091) (0.117) (0.073)

Enrolled in science program 0.015 -0.054 -0.039

(0.131) (0.144) (0.057)

Parental support 0.293 0.153 0.445

(0.684) (0.760) (0.339)

Participated in commercial -0.054 0.224 0.170**

school support (0.131) (0.152) (0.076)

Socio-emotional functioning

Independence at school -0.142 0.318 0.176

(1.856) (2.111) (1.067)

Independence out of school -2.006 0.112 -1.894*

(2.107) (2.381) (1.095)

Motivation towards learning tasks 0.599 -1.166 -0.567

(1.215) (1.366) (0.621)

Well-being at school 4.324** -3.847* 0.477

(1.781) (1.996) (0.915)

Interest in society 1.070 -0.551 0.519

(0.967) (1.139) (0.596)Note: Each row reports two IV regressions with winning the lottery as an IV for ever attending aMontessori school. Columns (1) and (2) display the treatment effect for pre-vocational IV students andthe interaction effect on senior general/pre-university students. Column (3) is another IV regressiondescribing the treatment effect for senior general/pre-university students. The sample is restrictedto students participating on the lottery. Robust standard errors are reported in parentheses. Allregressions include lottery fixed effects and controls. Controls include gender, being a nonwestern orwestern migrant, student age, living in a disadvantaged neighborhood, cito testscore, making anothertest than cito, no information on primary school test-score and primary school advice. Exams aregraded on a 10-point scale, where grades above 5.5 are considered a pass. ***p<0.01 **p<0.05*p<0.10.

97

Chapter 5

Special needs students in regulareducation: Do they affect theirclassmates?

5.1 Introduction

Since the 1990’s, there has been an international trend towards educating children withSpecial Educational Needs (SEN) in regular schools instead of in special schools (e.g.Ainscow & César, 2006). This development, often referred to as ‘Inclusive Education’,‘Mainstreaming’ or ‘Integration’, has not gone unnoticed: there have been fierce debatesabout the desirability of educating children with SEN in regular classes.

One of the main concerns is that inclusion may have negative impact on the achieve-ment of students without SEN (from here on referred to as ‘regular’ students). Manyteachers and parents are afraid that regular students get distracted by the behaviorof SEN students and that SEN students need more teacher attention at the expenseof regular students. On the other hand, people in favor of inclusive education statethat there is more teacher support in inclusive classrooms, which could have a positiveimpact on the regular students.

This chapter uses data from centralized exams in Dutch secondary education anda standardized national test in Dutch primary education to investigate whether thepresence of students with Special Educational Needs affects the achievement of regularstudents. Since the presence of SEN students is related to school and peer charac-teristics, three different strategies are pursued to deal with selection issues. First, acentral feature of the Dutch secondary school system is exploited: students take a lim-ited number of courses within the secondary school program. Since different studentstake different courses, there is within-student-between-course variation in the numberof SEN students. Using student fixed effects, I compare student achievement in courses

99

with varying numbers of SEN students. Second, I follow a widely used approach in theliterature on peer effects (e.g. Hoxby, 2000c; Lavy et al., 2012a) by using school fixedeffects that exploit within school cohort-to-cohort variation in the number of SENstudents. Third, additional evidence for primary education is obtained by utilizingneighborhood variation in the number of SEN students.

The results consistently show that the number of SEN students is unrelated tostudent achievement, indicating that the presence of SEN students does not adverselyor favorably affect the achievement of regular students. This finding is not driven by alack of precision: the standard errors are small and the point estimates are close to zero.Distinguishing between different types of SEN does not change the results: there is nodifferential effect of the presence of students with visual problems, hearing problems,physical and mental disabilities or behavioral problems. Studying cohort level data,no evidence is found for heterogeneous effects of the presence of SEN students acrossthe ability distribution.

These findings contrast with recent findings that disruptive or disabled peers nega-tively affect academic achievement (e.g. Figlio, 2007; Aizer, 2008; Fletcher, 2010; Lavyet al., 2012a). It is quite likely that the difference is due to differences in the edu-cational policy context. This study investigates the presence of students with severeand diagnosed special educational needs. The presence of such students in Dutch reg-ular education classes is highly subsidized, with a per student budget more than twicethe amount of the regular student budget. Although this cannot be studied directly,it seems that these additional funds are currently sufficient to avoid negative exter-nalities of the presence of SEN students in regular primary and secondary educationclassrooms.

The chapter is structured as follows. Section 5.2 discusses related literature. Section5.3 provides background information on the Dutch education context and on inclusiveeducation in the Netherlands. Sections 5.4 and 5.5 describe the data and empiricalstrategies and in section 5.6, the results are presented and discussed. Section 5.7concludes.

5.2 Related literature

Educational research into the effects of inclusive education on regular students mainlycompares the achievement of regular students in more and less inclusive classes (e.g.Kalambouka et al., 2007; Ruijs et al., 2010; Farrell et al., 2007; Salend & GarrickDuhaney, 1999). With these general comparisons, causal inferences on the effects ofinclusion cannot be drawn.

In the education economics literature, there have been a few studies pursuing novelidentification strategies to identify the causal impact of SEN students on regular stu-

100

dents. Hanushek et al. (2002) use longitudinal data on mathematics achievement inprimary school. They investigate grade level variation in the proportion of SEN stu-dents, using student- and school-by-grade fixed effects. They find no evidence that thepresence of SEN students harms the achievement of regular students; an increase in theproportion of students with ‘other’ disabilities (i.e. not learning disabled, emotionallydisturbed or speech impaired) even increases test score gains for regular students. Usingan IV strategy with boys having girls’ names as an instrument for disruptive behav-ior, Figlio (2007) finds that disciplinary problems are related to reduced mathematicsscores and an increase in disruptive behavior of peers in middle school.

Friesen et al. (2010) exploit grade level variation in the proportion of SEN studentsin primary school, using school fixed effects. They find a negative coefficient for in-cluding students with learning and behavioral difficulties on numeracy and reading testscores, but find a positive coefficient for the inclusion of students with other disabil-ities. All results, however, are statistically insignificant. Fletcher (2010) investigatesclass level inclusion of students with emotional problems in early primary school. Heuses school and student fixed effects specifications, and finds that regular students ininclusive classes score lower on reading and mathematics tests.

Lavy et al. (2012a) investigate the grade level presence of low-achieving studentsin middle and high school. Using school fixed effects and correcting for school-specificlinear time trends in low-achievers, they find that a higher proportion of low-achievingpeers has a small negative effect on exam scores, especially for students from low socio-economic background. Using a similar empirical approach, Carrell & Hoekstra (2010)find that students exposed to domestic violence decrease peer achievement and increaseproblematic behavior.

Using student fixed effects regressions, Aizer (2008) finds that the presence of stu-dents with undiagnosed ADD harms peer achievement, mainly for boys. Once theADD is diagnosed and behavior improves, peer achievement also improves.

The issue of the impact of inclusive education on regular students is closely relatedto the literature on peer effects. There, somewhat mixed results are found (e.g. Ammer-mueller & Pischke, 2009; Booij et al., 2014; Duflo et al., 2011; Hoxby, 2000c; Sacerdote,2001), although it seems that having academically strong peers has a positive effect onstudent achievement.

This study contributes to the current literature in two ways. First, I pursue anidentification strategy that is novel for the inclusive education literature, exploitinga key feature of the Dutch secondary school system. Since students take a personalselection of courses, I am able to use within-student-between-course variation in thenumber of SEN students with a student fixed effect strategy. In a slightly differentcontext, Lavy et al. (2012b) use within student between course variation to study abilitypeer effects. Second, as will be described in section 5.3.2, Dutch regular education

101

schools get substantially compensated for educating students with special educationalneeds. It is interesting to investigate the effects of inclusive education in a context withsignificant additional funds for students with special educational needs.

5.3 The Dutch context

5.3.1 Primary and secondary education in the Netherlands

In the Netherlands, school choice is free: students and parents are not restricted bycatchment areas. Virtually all primary and secondary schools are completely publiclyfunded, including religious and special program schools. The government funding ofschools is to a large extent dependent on student numbers, in which money follows thestudent. The per student funding is nationally determined and averages to €6030 perstudent in primary education and €7730 per student in secondary education for 2012(Ministerie van Onderwijs, 2012a).

Schools can get additional funding for students from disadvantaged backgrounds.In primary education, the ’weighted student funding’ amounts to 0.3 or 1.2 times theregular per student funding, dependent on the educational background of the par-ents. In secondary education, schools can get €750 for each student from disadvan-taged neighborhoods, if the number of such students exceeds a substantial threshold(Staatscourant, 2011). Further, secondary schools can apply for funding for additionalsupport to students who lag behind significantly when starting secondary education inthe vocational tracks. In this chapter, student level eligibility for these arrangementsis used as control variables.

Dutch primary education starts at age 4 and lasts until age 12. Students are gener-ally educated by one or two teachers for an entire school year. At the end of primaryeducation, most students take a standardized national test called the ’citotoets’. To-gether with the advice of the primary school teacher, this test determines which tracka student should take in secondary education. The outcomes of this high stakes testare used as a measure of student performance in primary education.

Secondary education in the Netherlands lasts four to six years. The secondaryschool system is highly tracked, the exact length depends on the school track. Thelowest tracks (pre-vocational secondary education, vmbo) last four years, and giveaccess to vocational education programs. Within the pre-vocational track, there arefour different levels, each giving access to different levels of vocational education pro-grams. From low to high, their names are basis (10.9%), kader (13.6%), gl (2.9%)and tl (24.1%), with the percentage of students taking each school track reported inparentheses. For clarity reasons, the levels are denoted by the numbers I to IV, withI being the lowest level. The intermediate track (senior general secondary education,

102

havo, 27.7%) takes five years, and gives access to higher professional education. Thehighest track (pre-university education, vwo, 20.9%) takes six years, and gives accessto university education (Eurydice, 2009). Not all secondary schools offer all schooltracks. Dependent on student achievement and school policies, students can changetrack during secondary education. Also, they can decide to take a higher track afterfinishing a lower track.1

In the second half of secondary education, students choose their exam courses. Partof the exam courses are clustered in a specialization, such as “Science and Technology”or “Economics and Society”. Other courses are obliged for all students (e.g. Dutch,English) and part of the courses can be freely chosen. The courses are taught byspecialized teachers: students have different teachers for each course. Secondary schoolshave to follow national curriculum guidelines, and students take centrally determinednational exams at the end of secondary school. The national exams count for 50% ofstudents’ final grades, the other 50% is determined by school specific exams taken inthe last two or three years of secondary education.

5.3.2 Inclusive education

In the Netherlands, special schools coexist next to inclusive education. In total, about3,7% of the Dutch primary school students and about 5% of the Dutch secondary schoolstudents are identified as having severe Special Educational Needs (DUO, 2010). Thesestudents are eligible for special education schools, or can get ’backpack’ funding. The’backpack’ is a pupil-bound-budget, meant to fund the additional support needed toeducate children with severe special educational needs in regular schools. The ’back-pack’ was introduced to facilitate inclusive education, giving parents the choice betweenregular and special schools.

The additional funding is substantial: for primary education, it ranges from about€7050 for students with severe social, emotional or behavioral problems, to €13.550for deaf children. On top of the budget for the regular primary school, special ed-ucation schools get around €5000 for an included student, in order to provide addi-tional specialist support to the regular primary schools. In secondary education, theyearly ’backpack’ budget for the regular school is about €3500. The affiliated spe-cial education school receives between €2923 and €5234 to provide specialist support(Staatscourant, 2012).2 The use of the pupil-bound-budget depends on the needs of

1In that case, students enter the higher track in the year before the final exams. When doingpre-university education after senior general secondary education, for example, students enter in year5 of pre-university education. When graduating at once, they have their pre-university degree after 7years of secondary education instead of 6.

2When a student is also eligible for additional support for students who lag behind in the vocationaltrack, the backpack funding decreases by €735 to €1617. The average budget for students with severespecial educational needs in special education schools amounts to €22.600 per student (Ministerie van

103

the student. For example, it can include physical adaptations in the class, remedialteaching support or the presence of a teaching assistant. Currently, about 2% of thechildren in regular primary education and 1% of the children in regular secondary ed-ucation gets educated with ’backpack’ funding. This chapter focusses on the inclusionof students with ’backpack’ funding.

Whether a student is eligible for special education or ’backpack’ funding is decidedby a regional committee; parents have to apply for it. The committee requests officialreports on the disability of the student, made, for instance, by a certified psychologist.The student needs to score at least two standard deviations below average on relevanttests and the disability of the student needs to be a stable trait. When the committeedecides that a child is eligible, parents can choose for a special or a regular school.Parents can subscribe their child to any school, although schools can reject studentswith SEN when they have a good motivation to do so. When a child is accepted to aregular school, it mainly follows education in the regular classroom.

The special education schools and the specialist support for students with severeSEN are organized in four clusters, dependent on the type of special needs.3 Thefirst cluster is meant for students with visual handicaps: a visual disability causingproblems in educational participation. The second cluster focusses on communicativehandicaps, including students with significant hearing loss or students with very weakcommunicative abilities. The third cluster is for students with physical and cognitivehandicaps, including students with severe learning disabilities, physically handicappedstudents and students with a long-term illness. The fourth cluster is for studentswith severe social, emotional or behavioral problems, including students with formallydiagnosed4 behavioral disorders, developmental disorders and psychiatric problems.It is possible that students with different types of problems have different effects onregular education students. Therefore, I study whether there are differential effects ofincluding students with different types of special educational needs.

Onderwijs, 2012b). As of 2014, there will be substantial reforms in special education; the backpackfunding will be abolished.

3Children with less severe SEN can be referred to non-regular education schools as well, such asspecial primary schools (‘speciaal basisonderwijs’) or practical education (‘praktijkonderwijs’). Theseschools are different from the special education schools (’speciaal onderwijs’) and are not taken intoaccount in this study.

4Formally diagnosed refers to disorders fitting in standardized classification systems such as DSM-IV and ICD-10.

104

5.4 Data

5.4.1 Data primary education

For primary education, the data used in this study are data on all Dutch students leav-ing primary education in 2009, 2010 and 2011. The data are provided by DUO, thegovernment organization that finances schools and administrates educational data forall students and schools in the Netherlands. Only students in regular primary schoolsare taken into account. The data include information on citotest scores,5 receiving’backpack’ funding and on background characteristics including gender, weighted stu-dent funding, ethnicity and student post code area.

The information on backpack funding is used to compute the number of SEN stu-dents per cohort within schools. Descriptive statistics on SEN and non-SEN studentsare reported in Table 5.1. It turns out that SEN students are more often male, Dutchand from non-disadvantaged backgrounds. SEN students have a lower participationrate for the citotest, and when they do participate, they score lower. Since we areinterested in the effect of the number of SEN students on regular students, all SENstudents are dropped from the sample.

As can be seen in the table, 18% of the students in cito participating schools donot take the citotest. Descriptive statistics on differences between students taking andnot taking the citotest are reported in the 4th and 5th column of Table 5.1. The tableshows that students who do not take the citotest are slightly more disadvantaged thanstudents who do: they are more often students with weighted student funding, lessoften Dutch and they are slightly older. This is likely to be caused by the fact thatschools are free to decide which students participate in the citotest. When schoolsexclude bad performing students, they increase their average testscores at the sametime. Because of this, cito participation is used as an outcome next to citoscore. Thecitoscores are standardized to ease interpretation.

The selection of students is somewhat different for the analyses using neighborhoodinformation. The descriptive statistics for this sample are also listed in Table 5.1. Forthe neighborhood analyses, we are interested in students who would be in the finalgrade of primary education when following the typical primary school path. A cohortis defined as all students in a 4-digit post code area who are 11 years old at October 1st2009, 2010 or 2011. A 4-digit post code area is an administrative neighborhood areawith an average of 4153 inhabitants (Statistics Netherlands, 2013). Since October 1st is

5Although the citotest is taken by the majority of primary school students, it is not obliged. About15% of the schools does not take the citotest. These schools are slightly different from schools thatdo take the citotest: they have slightly more SEN students, more Dutch students and fewer weightedstudents. Since there is no other standardized test available, these schools are dropped from thesample. 1 case was dropped because of missing gender information.

105

Tab

le5.1:

Descriptive

statistics

forstudents

inprimaryed

ucation

Stud

ents

infin

algrad

eof

prim

aryeducation

Difference

betw

eenSE

NDifferencesby

cito

participation

Neigh

borhoo

d

andno

n-SE

Nstud

ents

forno

n-SE

Nstud

ents

sample

non-SE

NSE

Nwithcito

witho

utcito

Boy

0.49

0.76***

0.49

0.49

0.49

Dutch

0.78

0.85***

0.78

0.76***

0.78

Non

-western

non-Dutch

0.16

0.09***

0.16

0.17***

0.16

Western

non-Dutch

0.06

0.05**

0.06

0.06***

0.06

Unk

nownethn

icity

0.00

0.00**

0.00

0.01***

0.00

Noad

dition

alfund

ing

0.86

0.89***

0.86

0.84***

0.86

0.30

fund

ingweigh

t0.08

0.08***

0.08

0.09***

0.08

1.20

fund

ingweigh

t0.06

0.03***

0.06

0.07***

0.06

Age

atOctob

er1st

11.13

11.34***

11.13

11.16***

11

Shareof

cito

participation

0.82

0.75***

0.57

Meanstan

dardized

citoscore

0.00

-0.30***

0.11

SDstan

dardized

citoscore

(1.00)

(1.05)

(0.95)

Num

berof

stud

ents

462,227

8,775

377,228

84,999

522,095

Shareof

stud

ents

0.98

0.02

0.82

0.18

Coh

ortsize

26.77

50.10

%of

SEN

stud

ents

percoho

rt2.05%

4.95%

Num

berof

SEN

stud

ents

percoho

rt0.50

2.63

SDnu

mbe

rof

SEN

percoho

rt(0.83)

(3.24)

Num

berof

scho

ols

5,958

Num

berof

SEN

stud

ents

neighb

orho

odsample

30,274

Note:

The

twocolumns

oncito

participation

referto

stud

ents

inscho

olstaking

thecitotest

forat

leaston

estud

entin

thesample.

The

numbe

rsforthe

neighb

orho

odsamplereferto

non-SE

Nstud

ents.The

starsreferto

thep-values

oftw

o-grou

pmeancompa

risont-tests.

***p

<0.01

**p<

0.05

*p<0.10.

106

a weak cutoff date6 and since it is common to repeat or skip classes in the Netherlands,only 79.9% of the regular education students is actually in the final grade of primaryeducation at age 11. 19.1% of the students is in the penultimate grade. As describedbefore, the citotest is taken at the end of primary education. Since repeaters are 12years old in the final grade, their citoscores are not taken into account when usingneighborhood information. Therefore, the average citoscore is somewhat higher in thissample.7

Students at schools not participating on the citotest are taken into account inthe neighborhood analyses. SEN students are defined as all students with a specialeducation indication, which includes students with backpack funding and studentseducated at special education schools.8 Again, SEN students are dropped from thesample.

5.4.2 Data secondary education

For secondary education, administrative data on all Dutch students taking secondaryschool exams in 2009, 2010 and 2011 are used. The data are provided by DUO. Onlystudents in regular secondary education schools are taken into account. The centralexam grade for each course is available for each student. To ease interpretation, theexam grades are standardized within school tracks. When a student is in the sampletwice because of failing the exams the first time or because of attaining a higher schooltrack, each observation is treated as a separate entry. Next to exam grades, the datainclude a rich set of background characteristics: gender, age, ethnicity, living in adisadvantaged neighborhood, missing neighborhood information and getting additionalsupport in the pre-vocational tracks.

Further, the data include information on receiving ’backpack’ funding.9 This infor-mation is used to compute the number of SEN students per cohort. Since students geteducated within their own school track, each school track within a school is treated sep-arately. Descriptive statistics on the number of SEN and non-SEN students per schooltrack are reported in Table 5.2. Note that the percentage of SEN students gets lowerfor higher school tracks. Table 5.3 shows descriptive statistics for students with andwithout SEN. Given the severity of the special educational needs, it is surprising thatstudents with SEN have higher exam grades than students without SEN. A potential

6Virtually all students start primary education at age 4, but the first two grades are similar tokindergarten. October 1st refers to a cutoff for 6 year olds, when the primary school curriculum starts.

7Repeaters are by definition low-achieving students.8Students in other types of non-regular primary education (’speciaal basisonderwijs’) are not taken

into account in the analyses. Changing the definition of SEN to include these students does not changethe results.

9For 0.03% of the students, the information on receiving ’backpack’ funding is missing or incon-sistent. These cases are dropped from the sample. Additionally, 0.21% of the students do not haveinformation on central exam grades. These students are not taken into account in the analysis.

107

Tab

le5.2:

Number

ofstudents

andschoo

lsper

secondaryed

ucation

schoo

ltrack

Stud

ents

Stud

ents

Percentage

Meannu

mbe

rSD

numbe

rNum

berof

Num

berof

witho

utSE

NwithSE

NSE

Nstud

ents

SEN

percoho

rtSE

Npe

rcoho

rtscho

ols

coho

rts

Pre-voc.I

56,048

949

1.66%

0.79

1.18

441

1,220

Pre-voc.II

70,283

861

1.21%

0.69

1.04

451

1,252

Pre-voc.III

14,955

172

1.14%

0.22

0.55

299

768

Pre-voc.IV

125,023

1,750

1.38%

0.85

1.54

741

2,068

Senior

gen.

143,847

1,203

0.83%

0.84

1.15

500

1,446

Pre-uni.

108,829

537

0.49%

0.36

0.69

516

1,497

Total

518,985

5,472

1.04%

0.67

1.17

1,036

8,251

Note:

The

percentage

ofSE

Nstud

ents

istheoverallpe

rcentage

ofSE

Nstud

ents

inthesample.

The

meanan

dSD

ofthenu

mbe

rof

SEN

stud

ents

percoho

rtreferto

themeanan

dSD

over

coho

rtswithinscho

ols,

usingon

eob

servationpe

rscho

olpe

rcoho

rt.

108

explanation is that schools are tracking special needs students somewhat below theiractual academic level. Further, SEN students are more often male, Dutch and fromnon-disadvantaged neighborhoods. Since we are interested in the effect of the numberof SEN students on regular students, all SEN students are dropped from the sample.

5.5 Empirical strategy

5.5.1 Empirical strategy 1: Student fixed effects

In this study, three strategies are used to identify the effect of the presence of SEN stu-dents on regular education students. The first strategy is for secondary education only,and is based on the fact that secondary school students follow a personal selection ofcourses. Using student fixed effects, I exploit within-student-between-course variationin the number of SEN students. Basically, this strategy compares student achievementin courses with varying numbers of SEN students:

yicst = δ1SENcst +W ′cstγ1 + ζ1,i + κ1,c + ε1,icst (5.1)

Here, yicst indicates the standardized exam grade of student i in course c in schools in year t. SENcst indicates the number of students with special educational needs incourse c in school s in year t, which makes δ1 the parameter of interest. W ′

cst is a vectorof peer characteristics in course c in school s in year t, such as the percentage of boys inthe course. ζ1,i is an individual fixed effect, picking up student constant characteristics.κ1,c is a course fixed effect, which picks up general differences between courses. ε1,icst isan individual and course specific error term, which is assumed to be exogenous apartfrom the individual and course fixed effects. To account for the possibility that coursecharacteristics change between years, equation 5.2 is estimated. In this equation, thecourse fixed effect κ1,c is replaced for a course*year fixed effect λ2,ct.

yicst = δ2SENcst +W ′cstγ2 + ζ2,i + λ2,ct + ε2,icst (5.2)

Since the course content is different for each school track, secondary school tracks areanalyzed separately. By using this student fixed effects strategy, a number of potentialselection problems are solved. The student fixed effects capture all student constantcharacteristics, including the student’s background and ability, the characteristics ofthe school and the average cohort quality.

What is left is variation in exam grades within students between different courses:students generally have varying talents and different grades for different courses. The

109

Tab

le5.3:

Descriptivesforstudents

withan

dwithou

tSEN

insecondaryed

ucation

Pre-voc.I

Pre-voc.II

Pre-voc.III

Pre-voc.IV

Senior

gen.

Pre-uni.

non-SE

NSE

Nno

n-SE

NSE

Nno

n-SE

NSE

Nno

n-SE

NSE

Nno

n-SE

NSE

Nno

n-SE

NSE

N

Std.

exam

grad

e0.00

0.17***

0.00

0.31***

0.00

0.32***

0.00

0.40***

0.00

0.38***

0.00

0.23***

(SD)

(1.00)

(1.07)

(1.00)

(1.10)

(1.00)

(0.96)

(1.00)

(1.10)

(1.00)

(1.11)

(1.00)

(0.98)

Add

itiona

lsup

port

0.58

0.63***

0.24

0.34***

0.06

0.10**

0.03

0.11***

Boy

0.57

0.75***

0.53

0.77***

0.48

0.74***

0.50

0.79***

0.48

0.73***

0.46

0.68***

Dutch

0.68

0.83***

0.75

0.87***

0.85

0.88

0.78

0.87***

0.83

0.90***

0.84

0.90***

Surina

mese

0.04

0.01***

0.04

0.01***

0.02

0.01

0.03

0.01***

0.02

0.01***

0.02

0.01

Arube

an0.02

0.02

0.01

0.01

0.01

0.00

0.01

0.01

0.01

0.00**

0.01

0.01

Turkish

0.08

0.02***

0.05

0.02***

0.03

0.02

0.04

0.01***

0.02

0.01***

0.01

0.01

Moroccan

0.07

0.03***

0.05

0.02***

0.02

0.01

0.03

0.01***

0.02

0.01***

0.01

0.00

Non

-western

0.06

0.02***

0.05

0.03***

0.03

0.02

0.05

0.03***

0.05

0.03***

0.05

0.02***

Western

0.05

0.06

0.05

0.05

0.04

0.06

0.06

0.06

0.06

0.05

0.07

0.05**

Disad

vantaged

0.22

0.11***

0.16

0.09***

0.09

0.08

0.13

0.11***

0.10

0.09

0.09

0.07

neighb

orho

od

Mis.neighb

orho

od0.00

0.00**

0.00

0.00

0.00

0.01

0.01

0.01

0.01

0.00

0.00

0.00

inform

ation

Age

15.61

15.71***

15.47

15.68***

15.32

15.53***

15.37

15.58***

16.58

16.86***

17.16

17.43***

Num

berof

courses

7.74

7.55***

8.09

8.06***

9.39

9.37

9.25

9.20*

10.56

10.55

13.40

13.11***

Coh

ortsize

66.60

62.97***

76.07

73.49**

42.38

41.40

83.14

74.82***

120.21

116.94**

90.69

89.21

Note:

The

starsreferto

thep-values

ofatw

o-grou

pmeancompa

risont-test

compa

ring

SEN

andno

n-SE

Nstud

ents.***p

<0.01

**p<

0.05

*p<0.10.

110

Figure 5.1: Variation in the number of SEN students per course

Note: Only courses that are taken by more than 1000 students are displayed in the figure.

question is whether these differences are systematically related to the number of SENstudents in a course. Nevertheless, there might also be differences in course character-istics: some courses may be more difficult than others or some courses may be moreinteresting for certain groups of students. Therefore, it is important to control forcourse and course*year fixed effects. In Figure 5.1, it can be seen that there is indeedvariation in the number of SEN students between courses. This figure is for the lowestschool track (pre-vocational I) only, the figures for the other school tracks are similarand reported in Figure 5.i in Appendix 5.A.

An important assumption in this estimation strategy is that the presence of SENstudents in a course is random within individuals: students should not select intocourses because of (avoiding) certain peers. Given the importance of course choicefor future education and employment possibilities, this is very likely. Even if someregular students are avoiding SEN students because they are more affected by negativeexternalities, this will bias the results in a predictable manner. Since the student fixedeffect strategy exploits within student variation in exam grades, all students are exposedto SEN students in their obliged courses. This negatively affects the achievement in

111

these courses for regular students avoiding SEN students. Their achievement in non-obliged courses will not be harmed, creating a larger difference in exam grades betweencourses with more or less SEN students. This would bias the results towards findingnegative effects.

Because some courses are obliged while others are not, and some courses are ingeneral more popular than others, the number of students in a course varies withinschool cohorts. For the larger courses, cohorts are split into several classes. Therefore,not all regular students will actually be exposed to SEN students when they are presentin a certain course. Since the probability of being educated with the SEN student(s)is larger when a course is smaller, course size is included as a cohort control. Twoalternative strategies, using the percentage of students instead of the number of SENstudents and restricting the sample to courses with 30 or fewer students, yield similarresults.10

5.5.2 Empirical strategy 2: School fixed effects

The student fixed effects can only be used for secondary education, since primary educa-tion students are educated in stable classes. Therefore, I also take a more conventionalapproach using school fixed effects. This strategy utilizes within school cohort-to-cohortvariation in the number of SEN students using the following specification:

yist = X′

istβ3 + P′

stγ3 + δ3SENst + µ3,s + ν3,t + ε3,ist (5.3)

Here, yist indicates the standardized citoscore or exam grade of student i in schools in year t. X ′

ist is a vector of student covariates, such as gender and ethnicity. Pst is avector of peer characteristics in school s in year t, such as the proportion of studentsfrom a disadvantaged neighborhood. SENst indicates the number of students withSpecial Educational Needs in school s in year t, which makes δ3 the parameter ofinterest. µ3,s is the school fixed effect, which will pick up time invariant differencesbetween schools. ν3,t is a year fixed effect and ε3,ist is an individual specific error term,which is assumed to be exogenous apart from the school and year fixed effects. Sincethe school fixed effects will only pick up time invariant differences between schools,equation 5.4 is also estimated. In this equation, a linear time trend in the numberof SEN students within a school controls for the possibility that some schools haveincreasing or decreasing numbers of SEN students over time.

10In practice, 30 students is often considered the maximum class size in secondary education. Resultsare reported in Tables 5.vii, 5.viii, 5.ix and 5.x in Appendix 5.A.

112

yist = X′

istβ4 + P′

stγ4 + δ4SENst + θ4PREDSENst + µ4,s + ν4,t + ε4,ist (5.4)

This strategy solves a number of potential selection problems. As described insubsection 5.3.2, Dutch schools can decide to reject a student with SEN. If weakerschools are more likely to reject, stronger schools and stronger classes have a highernumber of SEN students. By using the school fixed effects, I account for this type ofsystematic and time invariant differences between schools.

Second, within schools, the assignment of SEN students to classes may be non-random: teachers and principals may take the current class composition or teachercharacteristics into account. For example, when one class is more easily distractedthan another class within a grade, it can be attractive to place a SEN student withbehavioral problems in the latter class. The school fixed effect strategy solves this bylooking at the effect of inclusive education at the grade level instead of at the classlevel.

With respect to the overall cohort quality, it is assumed that the general cohortquality in a school is independent of the presence of SEN students. Teachers andprincipals should not take characteristics of the whole cohort into account in theirplacement decision. Table 5.4 shows joint balancing tests for primary education andall tracks of secondary education combined. Results for the individual school tracks arereported in Table 5.i of Appendix 5.A. The tables show that within schools, cohort-to-cohort variation in the number of SEN students is not strongly associated to changesin the observable characteristics of the non-SEN students. The first columns for eachtype of education show OLS estimates. These estimates show a positive correlationbetween the number of SEN students and background characteristics, indicating thatSEN students are included in more advantaged schools. When adding school fixedeffects, the coefficients become considerably smaller, indicating that these differencesare mainly due to variation between schools. Adding the school time trend decreases thecoefficients further. The joint balancing tests generally indicate that the backgroundcharacteristics are unrelated to the number of SEN students once controlling for schoolfixed effects and the school time trend. While some coefficients remain statisticallysignificant, their practical importance is small. For example, when looking at all schoollevels for secondary education, one more SEN student in the cohort relates to a 0.9%decrease in the probability of being Turkish and a 0.8% decrease in the probability ofliving in a disadvantaged neighborhood.

Further, it is assumed that without ’backpack’ funding, students with SEN areplaced in special education. When SEN students would otherwise be present as ’prob-

113

Table 5.4: Joint balancing tests for the number of SEN students in thecohort

Primary education

Boy -0.005 -0.001 0.000

(0.003) (0.002) (0.001)

Age -0.011** -0.008*** -0.004**

(0.005) (0.002) (0.002)

Western non-Dutch -0.001 0.005 0.000

(0.011) (0.005) (0.003)

Non-western non-Dutch -0.120*** -0.003 -0.002

(0.015) (0.004) (0.002)

Unknown ethnicity -0.084* -0.010 0.005

(0.046) (0.017) (0.010)

0.30 funding weight -0.107*** 0.001 0.001

(0.011) (0.004) (0.003)

1.20 funding weight -0.153*** 0.001 -0.001

(0.016) (0.005) (0.003)

Year fixed effects ! ! !

School fixed effects ! !

School specific time trend !

Number of students 462,227 462,227 462,227

Number of schools 5,958 5,958 5,958

F-statistic 28.705 2.280 1.357

p-value 0.000 0.026 0.219

Df (7,5957) (7,5957) (7,5957)(Table continues on next page)

114

Table 5.4: (continued)Secondary education (all school tracks)

Boy 0.027*** -0.003 -0.000

(0.008) (0.003) (0.002)

Age -0.037** 0.001 0.000

(0.015) (0.002) (0.001)

Surinamese -0.292*** -0.002 -0.004

(0.035) (0.006) (0.004)

Arubean -0.195*** -0.006 -0.003

(0.031) (0.010) (0.007)

Turkish -0.185*** -0.014** -0.009**

(0.029) (0.006) (0.004)

Moroccan -0.281*** -0.010 -0.001

(0.035) (0.007) (0.005)

Non-western -0.121*** -0.014*** -0.006**

(0.018) (0.005) (0.003)

Western -0.057*** -0.007 -0.003

(0.014) (0.004) (0.003)

Disadvantaged -0.152*** -0.015*** -0.008**

neighborhood (0.032) (0.005) (0.003)

Missing neighborhood 0.017 -0.006 -0.008

information (0.053) (0.015) (0.011)

Number of courses -0.062*** 0.031*** -0.010

(0.007) (0.009) (0.006)

Additional support 0.117** 0.006 0.006

(0.053) (0.009) (0.004)

Year fixed effects ! ! !

School fixed effects ! !

School specific time trend !

Number of students 518,985 518,985 518,985

Number of schools 1,036 1,036

F-statistic 13.541 2.933 1.949

p-value 0.000 0.001 0.026

Df (12, 1035) (12, 1035) (12, 1035)Note: Each column represents a regression of the number of SEN students in the cohort on studentbackground characteristics. The school specific time trend refers to a linear trend with the predictedvalues of the number of SEN students in a certain year. F-statistics, p-values and degrees of freedomat the bottom of the table refer to F-tests on the joint significance of gender, age, ethnicity andweighted student funding in primary education, and gender, age, ethnicity, disadvantaged neighbor-hood, number of courses and additional support in pre-vocational education for secondary education.Standard errors are reported in parentheses. Standard errors are robust and clustered at the schoollevel. ***p<0.01 **p<0.05 *p<0.10.

115

lematic’ students in regular schools, the quality of inclusive cohorts would be overes-timated by leaving out such ’problematic’ students. Table 5.5 shows that accountingfor school fixed effects, individual and cohort controls and a time trend in the numberof SEN students, the number of SEN students is positively related to cohort size. Forinstance, one additional SEN student is related to a 0.985 student increase in primaryschool cohort size. The fact that all coefficients are positive and close to 1 indicatesthat SEN students are additional to the regular cohort, instead of relabeling a ’prob-lematic’ student. Since students with ’backpack’ funding are by definition eligible forspecial education and need to be formally diagnosed, this finding is as expected.

Since the school fixed effects strategy exploits year to year variation in the numberof SEN students, it is important that the number of SEN students differs betweencohorts in the same school. Table 5.6 shows the number of SEN students in 2009 and2010 for primary and secondary education schools. The secondary education data isaggregated over all school tracks. As can be seen, the number of SEN students in aschool is not constant over school years. There are few SEN students in each cohort,in most cases, the variation comes from going from 0 to 1 SEN student.11

Next to the school fixed effects using student level data, the data is aggregated tothe cohort level to investigate heterogeneous effects of the presence of SEN studentson the achievement of regular students:

yst = X′

stβ5 + P′

stγ5 + δ5SENst + θ5PREDSENst + µ5,s + ν5,t + ε5,st (5.5)

Examples of the outcome variables in these regressions are the lowest and higheststandardized mean exam grade and the standard deviation of the standardized examgrades in the cohort. To account for differences in school size, the regressions areweighted by the number of observations per school.

5.5.3 Empirical strategy 3: Exploiting neighborhood variation

As described in subsection 5.5.2, the school fixed effects strategy hinges on the assump-tion that the quality of cohorts within schools is independent of the presence of SENstudents. While the balancing tests generally indicate that the observed backgroundcharacteristics are unrelated to the number of SEN students once controlling for schoolfixed effects and a school time trend, the possibility that the presence of SEN studentsis related to unobserved aspects of within-school cohort quality cannot be excluded.

11As in the student fixed effect strategy, students in bigger schools are potentially not exposed tothe SEN student when the SEN student is in another class. Robustness checks using the percentageinstead of the number of SEN students and restricting the sample to include only small schools yieldsimilar results. Results are reported in Tables 5.xi, 5.xii, 5.xiii and 5.xiv in Appendix 5.A.

116

Tab

le5.5:

Coh

ortsize

andthenu

mber

ofSEN

students

inthecohort

(1)

(2)

(3)

(4)

(5)

NClusters

Primaryeducation

6.722***

1.144***

1.143***

1.165***

0.985***

462,227

5,958

(0.380)

(0.122)

(0.122)

(0.122)

(0.176)

Pre-voc.I

8.722***

2.078***

2.078***

2.127***

0.750

56,048

441

(1.743)

(0.662)

(0.662)

(0.660)

(0.569)

Pre-voc.II

9.557***

2.502***

2.502***

2.570***

1.904***

70,283

451

(1.219)

(0.535)

(0.536)

(0.545)

(0.716)

Pre-voc.III

19.537***

2.052**

2.030**

1.848*

1.518

14,955

299

(2.516)

(1.004)

(1.002)

(1.021)

(1.671)

Pre-voc.IV

7.354***

2.281***

2.280***

2.176***

1.690***

125,023

741

(1.271)

(0.397)

(0.397)

(0.394)

(0.487)

Senior

gen.

8.627***

0.890

0.890

0.931

0.643

143,847

500

(1.273)

(0.605)

(0.605)

(0.594)

(0.881)

Pre-uni.

11.575***

3.366***

3.363***

3.230***

2.733**

108,829

516

(3.110)

(1.129)

(1.130)

(1.099)

(1.120)

Yearfix

edeff

ects

!!

!!

!

Scho

olfix

edeff

ects

!!

!!

Individu

alcontrols

!!

!

Coh

ortlevelc

ontrols

!!

Scho

olspecifictimetrend

!

Note:

Eachcoeffi

cientrepresents

asepa

rate

regression

withthenu

mbe

rof

SEN

stud

ents

asindepe

ndentvariab

lean

dthetotaln

umbe

rof

stud

ents

inthecoho

rtas

depe

ndentvariab

le.Individu

alcontrols

includ

egend

er,age,

ethn

icityan

dstud

entweigh

tforprim

aryeducation,

andad

dition

alsupp

ortin

pre-vo

cation

aleducation,

gend

er,a

ge,e

thnicity,d

isad

vantaged

neighb

orho

odan

dnu

mbe

rof

coursesforsecond

aryed

ucation.

Coh

ortmeancontrols

includ

estud

entindividu

alcontrols

averaged

bystud

ents’scho

ol,trackan

dyear.The

scho

olspecifictimetrendrefers

toalin

eartrendwiththepredictedvalues

ofthenu

mbe

rof

SEN

stud

ents

inacertainyear.

Stan

dard

errors

arerepo

rted

inpa

rentheses.

Stan

dard

errors

arerobu

stan

dclusteredat

thescho

ollevel.

***p

<0.01

**p<

0.05

*p<0.10.

117

Tab

le5.6:

Variation

inthenu

mber

ofSEN

students

over

time

Num

berof

SEN

stud

ents

2010

Primaryeducation

Second

aryed

ucation(allscho

oltracks)

01

23

45

>5

01

23

45

>5

02,782

896

210

5913

41

01,188

412

115

419

51

Num

ber

1763

392

124

529

50

1278

176

7434

93

5

ofSE

N2

179

108

6219

61

12

5763

2621

165

2

stud

ents

338

2518

62

20

319

237

123

42

2009

48

125

31

11

43

33

23

13

50

31

10

00

51

03

23

10

>5

22

32

00

2>

51

02

01

03

Note:

Thistablerepo

rtsthenu

mbe

rof

SEN

stud

ents

in2009

and2010

forprim

aryan

dsecond

aryeducationscho

ols.

118

More specific, it might be that SEN students are more easily referred to special educa-tion when the general quality of the school cohort is lower. In that case, we might findno effects of the presence of SEN students in the school fixed effects strategy, whilethe achievement of inclusive cohorts might have been better without the SEN studentspresent.

The student fixed effect strategy yields credible estimates that do not depend onthis assumption for secondary education. Given the educational context, this strategyis not feasible for primary education.12 Instead, I utilize neighborhood variation in thenumber of SEN students in two ways: using an IV strategy and using neighborhoodfixed effects. The idea behind using neighborhood information is that within neigh-borhoods, the presence of an additional SEN student is likely to be exogenous. It isunlikely that parents will move because of the presence of an additional SEN student inthe neighborhood cohort. Also, schools cannot influence the number of SEN studentsin their neighborhood, while they could potentially influence the presence of SEN stu-dents in a school cohort. On the other hand, even though there is free school choice,most primary school students go to school in their own neighborhood. To be exact,71% of the students go to a school in their own 4-digit post code area. As can be seenin Table 5.1, the average neighborhood cohort size is 50.1, while the average schoolcohort size is 26.8.

5.5.3.1 Neighborhood IV

In the instrumental variable strategy, the number of SEN students in a neighborhoodcohort is used as an instrument for the number of SEN students in a school cohort. Theidea is that schools cannot influence the presence of SEN students in neighborhoods,while it is a good predictor of the number of SEN students in school cohorts. In thisstrategy, we need to assume that the potential effects of the number of SEN studentsin the neighborhood operate via the schools. Again, we are interested in the effect ofthe number of SEN students on student achievement:

yist = X′

istβ6 + P′

stγ6 + δ6 ˆSENst + µ6,s + ν6,t + ε6,ist (5.6)

As before, yist indicates citoscore or cito participation of student i in school s inyear t. X ′

ist , P ′st , µ6,s , ν6,t and ε6,ist indicate student covariates, peer characteristics,

school fixed effects, year fixed effects and an individual specific error term. ˆSEN st

indicates the predicted values for the number of students with Special EducationalNeeds in school s in year t. The first stage predicted values are defined as:

12Reversely, the neighborhood strategies are not feasible for secondary education, since secondaryschool students are more mobile.

119

ˆSEN st = X′

istβ̂7 + P′

stγ̂7 + π̂7SENna + µ̂7,s + ν̂7,t (5.7)

Where SENna indicates the number of SEN students in neighborhood n in neigh-borhood cohort a. Cohort a is defined as all 11 year old students at October 1st ofyear t.

5.5.3.2 Neighborhood fixed effects

The neighborhood fixed effects strategy exploits neighborhood variation in a differentway:

yina = X′

inaβ8 + P′

naγ8 + δ8SENna + η8,n + υ8,a + ε8,ina (5.8)

Here, yina indicates citoscore or cito participation for student i in neighborhoodn in neighborhood cohort a. X ′

ina is a vector of student covariates, Pna is a vector ofpeer characteristics. SENna indicates the number of students with Special EducationalNeeds in neighborhood n in cohort a, which makes δ8 the parameter of interest. η8,nand υ8,a are neighborhood and cohort fixed effects.

While this strategy is less direct than the IV strategy, it does not assume that thereis no direct effect of SEN students in the neighborhood on student achievement. Itinvestigates whether variation in the number of SEN students in a neighborhood affectsthe achievement of students in that neighborhood. If schools are more inclined to referSEN students to special education in difficult or low quality cohorts, the number of SENstudents in the school will be lower in such cohorts. This would bias the results. At theneighborhood level, however, the number of SEN students in a cohort is fixed. Whenresults at the neighborhood level (including both ’backpack’ and special educationstudents) are similar to the school fixed effects results, this provides additional evidencethat schools are not selectively referring to special education.

5.6 Results

5.6.1 Results for empirical strategy 1: Student fixed effects

Results for the student fixed effect strategy are reported in Table 5.7. Each cellin Table 5.7 represents a separate regression of the standardized exam grades on thenumber of SEN students in the course. The columns represent 5 different specificationsfor each school track: column 1 are OLS estimates including year fixed effects, columns

120

Tab

le5.7:

Estim

ates

ofthe

effect

ofthe

number

ofSEN

students

onstan

dardized

centralexam

grad

es(studentfixedeff

ects)

(1)

(2)

(3)

(4)

(5)

NN

Observation

sStud

ents

Pre-voc.I

0.009

0.054***

-0.025**

-0.025**

-0.020*

271,376

55,957

(0.008)

(0.012)

(0.012)

(0.012)

(0.010)

Pre-voc.II

-0.005

-0.033***

-0.017*

-0.021**

-0.021**

355,213

70,277

(0.007)

(0.009)

(0.010)

(0.010)

(0.010)

Pre-voc.III

0.046***

-0.005

-0.007

-0.009

0.007

93,314

14,960

(0.014)

(0.017)

(0.017)

(0.017)

(0.019)

Pre-voc.IV

0.009*

0.026***

0.000

0.000

-0.002

780,815

125,018

(0.005)

(0.005)

(0.005)

(0.005)

(0.005)

Senior

gen.

-0.016***

-0.037***

-0.010*

-0.010*

-0.003

972,532

143,844

(0.005)

(0.006)

(0.005)

(0.005)

(0.005)

Pre-uni.

0.021*

-0.026***

-0.001

0.004

0.005

820,551

108,829

(0.011)

(0.007)

(0.007)

(0.006)

(0.006)

Stud

entfix

edeff

ects

!!

!!

Cou

rsefix

edeff

ects

!

Cou

rse*year

fixed

effects

!!

Coh

ort-levelc

ontrols

!

Note:

Eachcoeffi

cientrepresents

asepa

rate

regression

withthenu

mbe

rof

SEN

stud

ents

asindepe

ndentvariab

le.Coh

ortcontrols

includ

e(cou

rse)

coho

rtsize,

percentage

ofstud

ents

withad

dition

alsupp

ortin

pre-vo

cation

aleducation,

percentage

ofbo

ys,pe

rcentage

ofstud

ents

from

diffe

rent

ethn

icities,

percentage

ofstud

ents

from

disadv

antagedneighb

orho

ods,meanagean

dmeannu

mbe

rof

courses.

Stan

dard

errors

arerepo

rted

inpa

rentheses.

Stan

dard

errors

arerobu

stan

dclusteredat

thescho

ollevel.***p

<0.01

**p<

0.05

*p<0.10.

121

2 and 3 add student and course fixed effects. In column 4, course fixed effects arereplaced by course*year fixed effects. Column 5 adds cohort level controls.

In general, the coefficients are small and inconsistent in sign. The significant coeffi-cients in column 2 contrast to considerably smaller and mainly insignificant coefficientsin column 3. This difference indicates that course characteristics drive the associationbetween the number of SEN students and exam grades in column 2. For pre-vocationalI and IV, more SEN students are present in courses with higher mean exam grades,while for pre-vocational II, senior general secondary and pre-university education, moreSEN students are present in courses with lower exam grades. When adding course*yearfixed effects and cohort level controls, the coefficients remain similar to the coefficientsin column 3.

The results in the last three columns of Table 5.7 generally indicate that the numberof SEN students in the course is unrelated to a students’ standardized exam grade.Although the point estimates are precise (the standard errors are small), the coefficientsare generally insignificant and inconsistent in sign. Even for the significant coefficients,the practical relevance is low: for pre-vocational II, one additional SEN student in thecourse is related to a 2.1% of a standard deviation decrease in the exam grade for thatcourse. The lower bounds of the 95% confidence intervals for the estimates in column 5range from -0.040 to -0.005. In terms of a 1 SD increase in the number of SEN students,the lower bound of these confidence intervals range from -0.047 to -0.005, which makesthe lower bound similar in size to the point estimates found in Lavy et al. (2012a).

Table 5.8 makes a distinction between the inclusion of students with different typesof special educational needs. The table only shows the results of the full models. Resultsfor models without student fixed effects, course*year fixed effects and cohort controlsare reported in Table 5.ii of Appendix 5.A. It turns out that there is no differentialeffect for the inclusion of students with different types of SEN, the coefficients are closeto zero, precise and insignificant.13

Overall, the results of the student fixed effect strategy indicate that the presenceof SEN students does not help or harm the achievement of regular students. Whenaccounting for general differences between courses, the exam grades of regular studentsdo not differ between courses with more or less SEN students.

5.6.2 Results for empirical strategy 2: School fixed effects

Table 5.9 shows the results for the school fixed effects strategy. Each cell in thetable represents a separate regression of cito participation, citoscore or standardizedmean exam grades on the number of SEN students in the school cohort. Column 1

13Moreover, there might be differences in the effects of including male or female SEN students, asboys are more likely to exhibit externalizing behavior. Distinguishing between male and female SENstudents does not change the results.

122

Table 5.8: Estimates of the effect of the number of specific types of SENstudents on standardized central exam grades (student fixed effects)

Pre-voc. I Pre-voc. II Pre-voc. III Pre-voc. IV Senior gen. Pre-uni.

Visual problems -0.083 0.038 0.126 0.020 -0.029 -0.006

(0.063) (0.056) (0.078) (0.025) (0.018) (0.018)

Hearing problems -0.030 -0.052** -0.044 -0.015 -0.022 0.020

(0.023) (0.025) (0.064) (0.018) (0.016) (0.023)

Physical and mental -0.020 -0.036 0.005 0.020 -0.001 0.020

disabilities (0.023) (0.023) (0.050) (0.012) (0.010) (0.014)

Behavioral -0.017 -0.014 0.011 -0.004 -0.000 0.000

problems (0.012) (0.011) (0.023) (0.006) (0.006) (0.008)

N observations 271,376 355,213 93,314 780,815 972,532 820,551

N students 55,957 70,277 14,960 125,018 143,844 108,829Note: Each column represents a separate regression with the number of specific SEN students asindependent variables. All regressions include student fixed effects, course*year fixed effects and cohortcontrols. Cohort controls include (course) cohort size, percentage of students with additional supportin pre-vocational education, percentage of boys, percentage of students from different ethnicities,percentage of students from disadvantaged neighborhoods, mean age and mean number of courses.Standard errors are reported in parentheses. Standard errors are robust and clustered at the schoollevel. ***p<0.01 **p<0.05 *p<0.10.

shows OLS estimates with year fixed effects, column 2 includes school fixed effects. Incolumns 3 and 4, individual and cohort level controls are added. Column 5 includes aschool specific time trend in the number of SEN students in the cohort.

Adding the school fixed effects reduces the magnitude of the coefficients, indicatingthat selection plays a significant role in the coefficients in column 1. In column 2 tocolumn 5, the coefficients are small, precise and generally insignificant. The coefficientsindicate that the presence of SEN students in the cohort does not affect the exam gradesof regular students. For example, one additional SEN student in the cohort decreasesthe mean exam grade for pre-university education students by 0.9% of a standarddeviation.

Table 5.10 shows that there are no differential effects in the inclusion of studentswith different types of special educational needs. The table only shows results formodels with year fixed effects, school fixed effects, individual and cohort controls,and a school specific time trend in the number of SEN students. Results for moreparsimonious models are reported in Tables 5.iii and 5.iv of Appendix 5.A. Again, thecoefficients are close to zero, precise and insignificant.14

There might be differential effects of the inclusion of students with SEN on regulareducation students with different abilities. For example, since SEN students are likely

14Again, there might also be differences in the effects of including male or female SEN students, asboys are more likely to exhibit externalizing behavior. Distinguishing between male and female SENstudents does not change the results.

123

Tab

le5.9:

Estim

ates

oftheeff

ectof

thenu

mber

ofSEN

students

onstudentachievement(schoo

lfixedeff

ects)

(1)

(2)

(3)

(4)

(5)

NClusters

Citopa

rticipation

0.004

-0.000

-0.000

0.000

0.003

462,227

5,958

(Primaryeducation)

(0.003)

(0.004)

(0.004)

(0.004)

(0.004)

Stan

dardized

citoscore

0.041***

-0.001

-0.002

-0.000

-0.001

377,135

5,861

(Primaryeducation)

(0.004)

(0.003)

(0.003)

(0.003)

(0.004)

Pre-voc.I

-0.012

0.005

0.004

0.003

-0.001

56,048

441

(0.010)

(0.006)

(0.007)

(0.007)

(0.009)

Pre-voc.II

0.006

-0.000

-0.002

0.003

0.005

70,283

451

(0.010)

(0.008)

(0.008)

(0.008)

(0.010)

Pre-voc.III

0.092***

0.004

-0.004

-0.009

0.020

14,955

299

(0.031)

(0.026)

(0.024)

(0.024)

(0.042)

Pre-voc.IV

0.014**

0.002

-0.000

0.003

0.003

125,023

741

(0.007)

(0.006)

(0.006)

(0.006)

(0.008)

Senior

gen.

-0.002

-0.010*

-0.011**

-0.010*

-0.005

143,847

500

(0.007)

(0.005)

(0.005)

(0.005)

(0.007)

Pre-uni.

0.048***

-0.008

-0.007

-0.000

-0.009

108,829

516

(0.015)

(0.008)

(0.009)

(0.007)

(0.009)

Yearfix

edeff

ects

!!

!!

!

Scho

olfix

edeff

ects

!!

!!

Individu

alcontrols

!!

!

Coh

ortlevelc

ontrols

!!

Scho

olspecifictimetrend

!

Note:

Eachcoeffi

cientrepresents

asepa

rate

regression

withthenu

mbe

rof

SEN

stud

ents

asindepe

ndentvariab

le.Individu

alcontrolsinclud

egend

er,a

ge,e

thnicity

andstud

entweigh

tforprim

aryeducation,

andad

dition

alsupp

ortin

pre-vo

cation

aleducation,

gend

er,a

ge,e

thnicity,d

isad

vantaged

neighb

orho

odan

dnu

mbe

rof

coursesforsecond

aryeducation.

Coh

ortmeancontrolsinclud

ecoho

rtsize

andstud

entindividu

alcontrolsaveraged

bystud

ents’schoo

l,trackan

dyear.Stan

dard

errors

arerepo

rted

inpa

rentheses.

Stan

dard

errors

arerobu

stan

dclusteredat

thescho

ollevel.

***p

<0.01

**p<

0.05

*p<0.10.

124

Tab

le5.10

:Estim

ates

oftheeff

ectof

thenu

mber

ofspecificSEN

students

onstudentachievement(schoo

lfixedeff

ects)

Primaryeducation

Second

aryeducation

Citopa

rticipation

Citoscore

Pre-voc.I

Pre-voc.II

Pre-voc.III

Pre-voc.IV

Senior

gen.

Pre-uni.

Visua

lproblem

s0.017

0.003

0.044

0.070

-0.274

0.011

0.019

-0.012

(0.030)

(0.030)

(0.092)

(0.109)

(0.174)

(0.040)

(0.028)

(0.032)

Hearing

prob

lems

-0.009

-0.010

-0.017

0.018

-0.057

0.001

-0.024

0.010

(0.012)

(0.012)

(0.017)

(0.026)

(0.100)

(0.022)

(0.023)

(0.038)

Phy

sicala

ndmental

-0.001

-0.004

0.036

-0.008

0.080

0.036

-0.002

-0.004

disabilities

(0.010)

(0.009)

(0.026)

(0.032)

(0.086)

(0.026)

(0.015)

(0.015)

Behavioralp

roblem

s0.007

0.001

-0.004

0.005

0.015

-0.005

-0.005

-0.013

(0.006)

(0.005)

(0.010)

(0.013)

(0.056)

(0.007)

(0.009)

(0.012)

Nstud

ents

462,227

377,135

56,048

70,283

14,955

125,023

143,847

108,829

Nscho

ols

5,958

5,861

441

451

299

741

500

516

Note:

Eachcolumnrepresents

asepa

rate

regression

withthenu

mbe

rof

specificSE

Nstud

ents

asindepe

ndentvariab

les.

Allregression

sinclud

eyear

fixed

effects,

scho

olfix

edeff

ects,individu

alan

dcoho

rtcontrols

andascho

olspecifictimetrendin

thenu

mbe

rof

SEN

stud

ents.Individu

alcontrols

includ

egend

er,age,

ethn

icityan

dstud

entweigh

tforprim

aryeducation,

andad

dition

alsupp

ortin

pre-vo

cation

aleducation,

gend

er,a

ge,e

thnicity,d

isad

vantaged

neighb

orho

odan

dnu

mbe

rof

coursesforsecond

aryeducation.

Coh

ortmeancontrols

includ

ecoho

rtsize

andstud

entindividu

alcontrols

averaged

bystud

ents’scho

ol,trackan

dyear.Stan

dard

errors

arerepo

rted

inpa

rentheses.

Stan

dard

errors

arerobu

stan

dclusteredat

thescho

ollevel.

***p

<0.01

**p<

0.05

*p<0.10.

125

to need more attention, teachers might expect more independency from the non-SENstudents. This might harm the achievement of low-achieving students, while high-achieving students potentially benefit, driving the average effects to zero. To checkfor this type of differential effects, I aggregated the data to cohorts within schools andcomputed summary measures of the standardized exam grades.

The results in Table 5.11 show results for different summary measures, such as themean, standard deviation and the 10th and 90th percentile of the standardized examgrades in cohorts within schools. The results in Table 5.11 are for models with yearfixed effects, school fixed effects, cohort controls and a school specific time trend inthe number of SEN students. Results for more parsimonious models are reported inTables 5.v and 5.vi of Appendix 5.A. Again, the coefficients are small and insignificant.From this table, there is no evidence for differential effects for high and low achievingstudents within a school track. Considering that the effects of inclusive education donot differ for the different school tracks either, it seems that the effects of inclusiveeducation do not differ for high and low achieving students.

Overall, the results of the school fixed effect strategy are very similar to the resultsfrom the student fixed effect strategy. There is no evidence that the presence of SENstudents helps or harms the achievement of the regular students in their cohort.

5.6.3 Results for empirical strategy 3: Exploiting neighborhood

variation

5.6.3.1 Neighborhood IV

Table 5.12 shows the results for the IV strategy. The number of SEN students in theneighborhood cohort turns out to be a good predictor of the number of SEN studentsin school cohorts. The first stage is always significant and the partial F-statistic is wellabove the rule of thumb of a minimum of 10. Consistent to the earlier results, thecoefficients for the effect of SEN students on cito participation and citoscore are smalland insignificant. At the 1% level, there is one significant coefficient for citoscore, whenonly taking into account year fixed effects. When exploiting within school variationusing school fixed effects, this difference vanishes.

5.6.3.2 Neighborhood fixed effects

Table 5.13 shows results for the neighborhood fixed effect strategy. The coefficientsare small, precise and insignificant, indicating that the number of SEN students in theneighborhood cohort does not affect student achievement. These results are similar tothe results for the school fixed effects models, providing additional evidence that thezero results cannot be explained by more referral to special education in weaker school

126

Tab

le5.11

:Estim

ates

oftheeff

ectof

thenu

mber

ofSEN

students

onstudentachievementat

thecohortlevel

Primaryeducation

Second

aryeducation

Citopa

rticipation

Citoscore

Pre-voc.I

Pre-voc.II

Pre-voc.III

Pre-voc.IV

Senior

gen.

Pre-uni.

Mean

0.002

0.004

-0.000

0.005

0.028

0.001

-0.006

-0.009

(0.002)

(0.005)

(0.009)

(0.010)

(0.043)

(0.008)

(0.007)

(0.009)

SD-0.000

-0.007

0.003

0.002

0.007*

-0.002

-0.003

(0.003)

(0.006)

(0.005)

(0.016)

(0.004)

(0.004)

(0.006)

10th

percentile

0.005

0.001

0.005

0.048

-0.010

-0.001

-0.004

(0.008)

(0.014)

(0.013)

(0.052)

(0.010)

(0.011)

(0.012)

50th

percentile

0.002

-0.002

0.000

0.030

0.008

-0.007

-0.005

(0.006)

(0.010)

(0.011)

(0.043)

(0.008)

(0.006)

(0.010)

90th

percentile

0.006

-0.009

0.010

0.045

0.004

-0.009

-0.011

(0.005)

(0.012)

(0.013)

(0.056)

(0.011)

(0.009)

(0.018)

Nstud

ents

462,227

377,135

56,048

70,283

14,955

125,023

143,847

108,829

Nscho

ols

5,958

5,861

441

451

299

741

500

516

Note:

Eachcoeffi

cientrepresents

asepa

rate

regression

withthenu

mbe

rof

SEN

stud

ents

asindepe

ndentvariab

le.Allregression

sinclud

eyear

fixed

effects,schoo

lfix

edeff

ects,c

ohortcontrols

andascho

olspecifictimetrendin

thenu

mbe

rof

SEN

stud

ents.Coh

ortcontrols

includ

ecoho

rtsize,p

ercentageof

boys,m

eanage,

percentage

ofstud

ents

from

diffe

rent

ethn

icitiesan

dpe

rcentage

ofstud

ents

withweigh

tedstud

entfund

ingforprim

aryeducation,

andcoho

rtsize,p

ercentageof

stud

ents

withad

dition

alsupp

ortin

pre-vo

cation

aleducation,

percentage

ofbo

ys,pe

rcentage

ofstud

ents

from

diffe

rent

ethn

icities,

percentage

ofstud

ents

from

disadv

antagedneighb

orho

ods,

meanagean

dmeannu

mbe

rof

coursesforsecond

aryeducation.

Regressions

areweigh

tedby

themeancoho

rtsize

withinscho

ols.

Stan

dard

errors

arerepo

rted

inpa

rentheses.

Stan

dard

errors

arerobu

stan

dclusteredat

thescho

ollevel.***p

<0.01

**p<

0.05

*p<0.10.

127

Table 5.12: IV Estimates of the effect of the number of SEN students inprimary education

(1) (2) (3) (4)

Cito participation 0.029* 0.010 0.010 0.013

(0.016) (0.027) (0.027) (0.027)

First stage coefficient 0.049*** 0.049*** 0.049*** 0.048***

(0.003) (0.005) (0.005) (0.005)

Partial F statistic first stage 206.55 113.11 113.10 114.44

Standardized citoscore -0.144*** 0.015 0.018 0.019

(0.028) (0.027) (0.025) (0.025)

First stage coefficient 0.050*** 0.049*** 0.049*** 0.048***

(0.004) (0.005) (0.005) (0.005)

Partial F statistic first stage 177.13 89.88 89.88 89.91

Year fixed effects ! ! ! !

School fixed effects ! ! !

Individual controls ! !

Cohort level controls !

Cito participation Standardized citoscore

Number of students 460,823 Number of students 376,154

Number of schools 5,957 Number of schools 5,859Note: Instrument: number of SEN students in the neighborhood. Each column represents two separateIV regressions, one for standardized citoscore, one for cito participation. Individual controls includegender, age, ethnicity and student weight. Cohort mean controls include cohort size and studentindividual controls averaged by students’ school and year. Standard errors are reported in parentheses.Standard errors are robust and clustered at the school level. ***p<0.01 **p<0.05 *p<0.10.

128

Table 5.13: Effect of the number of SEN students in the post code area forprimary education students

(1) (2) (3) (4) (5)

Cito participation -0.001 0.000 0.000 -0.000 0.001

(0.002) (0.001) (0.001) (0.001) (0.001)

Standardized citoscore -0.006*** 0.002 0.002 0.002 0.001

(0.002) (0.001) (0.001) (0.001) (0.002)

Year fixed effects ! ! ! ! !

Post code area fixed effects ! ! ! !

Individual controls ! ! !

Cohort level controls ! !

Post code area time trend !

Cito participation Standardized citoscore

Number of students 522,095 Number of students 295,249

Number of post code areas 3,917 Number of post code areas 3,668Note: Each coefficient represents a separate regression with the number of SEN students as indepen-dent variable. Individual controls include gender, age, ethnicity and student weight. Cohort levelcontrols includes (post code area) cohort size and student individual controls averaged by students’post code area and year. Standard errors are reported in parentheses. Standard errors are robust andclustered at the level of the post code area. ***p<0.01 **p<0.05 *p<0.10.

cohorts.

5.7 Conclusions

In this chapter, effects of including students with Special Educational Needs on regularstudents are investigated. Three different strategies are used to account for the selectionproblems associated to this research question: I exploit within-student-between-coursevariation in the number of SEN students using student fixed effects, I exploit within-school-between-cohort variation using school fixed effects and I exploit neighborhoodvariation using IV and neighborhood fixed effects. The results consistently show nosignificant effects of inclusive education. The number of SEN students is neither sys-tematically related to regular students’ exam grades in a course or in a cohort insecondary education, nor systematically related to participation and achievement onthe standardized national test in primary education. There are no differential effectsby the type of special educational needs, or for high and low achieving regular students.

These results contrast with earlier findings that the inclusion of students with spe-cial educational needs harms the achievement and the behavior of regular students(Figlio, 2007; Fletcher, 2010; Lavy et al., 2012a). A possible explanation is the factthat inclusive education is highly subsidized in the Netherlands: the budget for a stu-dent with severe SEN is around double the budget of a regular student. It seems that

129

the current level of additional funding for SEN students is sufficient to avoid negativeexternalities of inclusion on the achievement of regular students.

Interestingly, Friesen et al. (2010) also find insignificant effects of inclusive educationin the Canadian context, where schools also receive substantial additional fundingfor educating students with special educational needs. Although this claim cannotbe studied directly, these results seem to indicate that additional funding can offsetnegative effects of the presence of SEN students. In that sense, the findings from thisstudy are also interesting in a broader education economics perspective, where thegeneral evidence on the effectiveness of additional resources is mixed (e.g. Krueger,2003; Hanushek, 2006).

From a policy perspective, it is important to stress that these findings are based onthe current situation on inclusive education in the Netherlands. When the additionalsupport for the inclusion of SEN students is increased or decreased, or when the pop-ulation of included students changes, the impact on regular students might change. Inthat respect, this study adds an interesting nuance to the inclusive education debate:in a situation with substantial additional funding, inclusive education does not harmthe achievement of regular students.

130

5.A Appendix

Figure 5.i: Variation in the number of SEN students per course

Note: Only courses that are taken by more than 1000 students are displayed in the figure.

131

Table 5.i: Joint balancing tests for the number of SEN students in thecohort

Pre-voc. I Pre-voc. II

Additional support 0.096** 0.001 0.007 0.096** 0.008 0.012**

(0.041) (0.011) (0.007) (0.039) (0.009) (0.005)

Boy 0.063** 0.000 0.001 0.058** 0.002 0.003

(0.025) (0.009) (0.006) (0.025) (0.008) (0.005)

Age -0.045* 0.011* 0.004 -0.032** -0.002 -0.003

(0.024) (0.007) (0.004) (0.014) (0.006) (0.004)

Surinamese -0.413*** -0.011 -0.010 -0.420*** -0.011 0.009

(0.072) (0.016) (0.010) (0.052) (0.013) (0.008)

Arubean -0.269*** -0.020 -0.014 -0.305*** -0.002 0.010

(0.073) (0.022) (0.015) (0.056) (0.021) (0.015)

Turkish -0.216*** -0.015 -0.006 -0.181*** -0.023* 0.003

(0.059) (0.014) (0.010) (0.051) (0.012) (0.009)

Moroccan -0.347*** -0.031** -0.009 -0.291*** -0.021 0.000

(0.069) (0.014) (0.010) (0.052) (0.015) (0.009)

Non-Western -0.209*** -0.026* -0.012 -0.204*** -0.015 0.003

(0.057) (0.014) (0.009) (0.036) (0.013) (0.008)

Western -0.122** -0.021 -0.019* -0.116*** -0.004 0.002

(0.053) (0.015) (0.011) (0.030) (0.011) (0.007)

Disadvantaged -0.273*** -0.020 -0.007 -0.239*** -0.011 -0.005

neighborhood (0.087) (0.012) (0.006) (0.055) (0.010) (0.007)

Missing neighborhood -0.012 0.039 0.054 0.184 0.079 0.040

information (0.126) (0.062) (0.044) (0.148) (0.059) (0.037)

Number of courses -0.003 0.004 0.002 -0.225*** -0.008 0.003

(0.016) (0.004) (0.003) (0.081) (0.025) (0.017)

Year fixed effects ! ! ! ! ! !

School fixed effects ! ! ! !

School specific time trend ! !

Number of students 56,048 56,048 56,048 70,283 70,283 70,283

Number of schools 441 441 441 451 451 451

F-statistic 5.980 1.455 1.049 7.499 0.859 0.937

p-value 0.000 0.138 0.402 0.000 0.589 0.509

Df (12, 440) (12, 440) (12, 440) (12, 450) (12, 450) (12, 450)(Table continues on next page)

132

Table 5.i: (continued)Pre-voc. III Pre-voc. IV

Additional support -0.121 -0.013 -0.010 0.758*** 0.005 -0.011

(0.108) (0.015) (0.013) (0.254) (0.047) (0.017)

Boy 0.009 0.015* 0.005 0.006 -0.014*** -0.006

(0.028) (0.009) (0.006) (0.012) (0.005) (0.004)

Age 0.017 0.009 0.006 0.034 -0.002 0.003

(0.036) (0.007) (0.006) (0.022) (0.005) (0.003)

Surinamese -0.249** 0.003 0.004 -0.231*** 0.003 -0.002

(0.096) (0.027) (0.012) (0.072) (0.014) (0.010)

Arubean -0.336*** -0.004 0.053** -0.198*** -0.010 -0.006

(0.098) (0.035) (0.024) (0.052) (0.025) (0.018)

Turkish -0.204* -0.004 -0.003 -0.254*** -0.007 -0.010

(0.113) (0.022) (0.010) (0.052) (0.013) (0.008)

Moroccan -0.265*** -0.053*** -0.026* -0.313*** 0.005 -0.001

(0.100) (0.019) (0.015) (0.072) (0.013) (0.010)

Non-Western -0.173* -0.017 0.007 -0.171*** -0.009 0.004

(0.102) (0.020) (0.014) (0.037) (0.011) (0.008)

Western -0.064 0.010 0.021 -0.060** -0.016 -0.006

(0.040) (0.019) (0.018) (0.025) (0.012) (0.008)

Disadvantaged 0.043 0.017 -0.002 -0.109* -0.034** -0.011

neighborhood (0.200) (0.016) (0.005) (0.061) (0.014) (0.008)

Missing neighborhood -0.055 0.027 0.046 -0.206*** -0.039 -0.038**

information (0.097) (0.053) (0.037) (0.074) (0.026) (0.019)

Number of courses 0.110 0.020 0.012 0.008 0.026* 0.002

(0.078) (0.014) (0.010) (0.026) (0.016) (0.009)

Year fixed effects ! ! ! ! ! !

School fixed effects ! ! ! !

School specific time trend ! !

Number of students 14,955 14,955 14,955 125,023 125,023 125,023

Number of schools 299 299 299 741 741 741

F-statistic 3.528 1.589 0.911 3.847 1.658 1.065

p-value 0.000 0.094 0.537 0.000 0.072 0.387

Df (12, 298) (12, 298) (12, 298) (12, 740) (12, 740) (12, 740)(Table continues on next page)

133

Table 5.i: (continued)Senior gen. Pre-uni.

Additional support

Boy 0.000 -0.000 -0.000 0.015** 0.001 0.003

(0.008) (0.005) (0.004) (0.007) (0.003) (0.002)

Age -0.013 -0.000 -0.001 -0.018* 0.002 -0.001

(0.012) (0.004) (0.002) (0.010) (0.004) (0.002)

Surinamese -0.330*** -0.006 -0.009 -0.092*** 0.014 -0.012

(0.050) (0.013) (0.009) (0.030) (0.013) (0.008)

Arubean -0.156*** -0.022 -0.011 0.032 0.037* 0.007

(0.047) (0.026) (0.016) (0.043) (0.021) (0.015)

Turkish -0.189*** -0.015 -0.027*** -0.072** -0.014 -0.006

(0.047) (0.016) (0.009) (0.032) (0.014) (0.010)

Moroccan -0.256*** 0.013 0.012 -0.148*** -0.016 0.004

(0.057) (0.016) (0.012) (0.036) (0.017) (0.011)

Non-Western -0.141*** -0.014 -0.017*** -0.013 -0.013 -0.011*

(0.025) (0.010) (0.006) (0.017) (0.008) (0.006)

Western -0.055*** 0.003 0.006 -0.013 -0.008 -0.005

(0.021) (0.009) (0.006) (0.017) (0.006) (0.004)

Disadvantaged -0.138*** -0.013 -0.013** -0.049 0.002 -0.001

neighborhood (0.050) (0.009) (0.006) (0.033) (0.009) (0.006)

Missing neighborhood 0.071 -0.030 -0.039* 0.092* -0.002 0.012

information (0.107) (0.036) (0.023) (0.049) (0.027) (0.015)

Number of courses 0.041 0.034 0.014 -0.016 -0.005 -0.026*

(0.065) (0.023) (0.014) (0.021) (0.017) (0.014)

Year fixed effects ! ! ! ! ! !

School fixed effects ! ! ! !

School specific time trend ! !

Number of students 143,847 143,847 143,847 108,829 108,829 108,829

Number of schools 500 500 500 516 516 516

F-statistic 5.214 1.081 2.623 3.304 0.824 1.168

p-value 0.000 0.375 0.003 0.000 0.616 0.307

Df (11, 499) (11, 499) (11, 499) (11, 515) (11, 515) (11, 515)Note: Each column represents a regression of the number of SEN students in the cohort on studentbackground characteristics. The school specific time trend refers to a linear trend with the predictedvalues of the number of SEN students in a certain year. F-statistics, p-values and degrees of freedom atthe bottom of the table refer to F-tests on the joint significance of additional support in pre-vocationaleducation, gender, age, ethnicity, disadvantaged neighborhood and number of courses. Standard errorsare reported in parentheses. Standard errors are robust and clustered at the school level. ***p<0.01**p<0.05 *p<0.10.

134

Table 5.ii: Student fixed effect estimates of the effect of the number ofspecific types of SEN students on standardized central exam grades fordifferent secondary school tracks

(1) (2) (3) (4) (5)

Pre-voc. I

Visual problems 0.064 0.057 -0.076 -0.082 -0.083

(0.055) (0.072) (0.060) (0.064) (0.063)

Hearing problems -0.017 0.045 -0.035 -0.036 -0.030

(0.014) (0.031) (0.024) (0.025) (0.023)

Physical and mental -0.011 0.067*** -0.030 -0.026 -0.020

disabilities (0.018) (0.024) (0.023) (0.024) (0.023)

Behavioral problems 0.017* 0.053*** -0.021 -0.022* -0.017

(0.010) (0.013) (0.013) (0.013) (0.012)

Pre-voc. II

Visual problems -0.004 -0.004 0.023 0.037 0.038

(0.061) (0.058) (0.058) (0.055) (0.056)

Hearing problems 0.036* -0.063** -0.057** -0.050** -0.052**

(0.021) (0.027) (0.027) (0.025) (0.025)

Physical and mental -0.023 -0.038* -0.033 -0.036 -0.036

disabilities (0.020) (0.023) (0.024) (0.024) (0.023)

Behavioral problems -0.008 -0.028*** -0.009 -0.015 -0.014

(0.009) (0.011) (0.011) (0.011) (0.011)

Pre-voc. III

Visual problems 0.171*** 0.162*** 0.151*** 0.114 0.126

(0.064) (0.048) (0.048) (0.085) (0.078)

Hearing problems 0.126** -0.082* -0.062 -0.064 -0.044

(0.054) (0.046) (0.056) (0.063) (0.064)

Physical and mental 0.137*** 0.042 0.028 -0.011 0.005

disabilities (0.045) (0.042) (0.045) (0.051) (0.050)

Behavioral problems 0.023 -0.009 -0.011 -0.006 0.011

(0.014) (0.023) (0.022) (0.021) (0.023)

Student fixed effects ! ! ! !

Course fixed effects !

Course*year fixed effects ! !

Cohort-level controls !

(Table continues on next page)

135

Table 5.ii: (continued)(1) (2) (3) (4) (5)

Pre-voc. IV

Visual problems 0.037 0.041 0.013 0.024 0.020

(0.039) (0.026) (0.024) (0.025) (0.025)

Hearing problems -0.021 0.018 -0.010 -0.012 -0.015

(0.022) (0.018) (0.017) (0.017) (0.018)

Physical and mental 0.023* 0.070*** 0.027** 0.022* 0.020

disabilities (0.013) (0.013) (0.013) (0.012) (0.012)

Behavioral problems 0.010* 0.019*** -0.003 -0.003 -0.004

(0.005) (0.006) (0.006) (0.006) (0.006)

Senior gen.

Visual problems -0.037** -0.050** -0.044** -0.040** -0.029

(0.018) (0.020) (0.019) (0.018) (0.018)

Hearing problems -0.020 -0.057*** -0.026 -0.027 -0.022

(0.017) (0.019) (0.017) (0.017) (0.016)

Physical and mental -0.001 -0.027** -0.010 -0.009 -0.001

disabilities (0.009) (0.012) (0.010) (0.010) (0.010)

Behavioral problems -0.019*** -0.037*** -0.006 -0.006 -0.000

(0.006) (0.009) (0.006) (0.006) (0.006)

Pre-uni.

Visual problems 0.054* -0.040* -0.002 -0.008 -0.006

(0.033) (0.022) (0.021) (0.018) (0.018)

Hearing problems -0.013 -0.007 0.015 0.018 0.020

(0.026) (0.025) (0.022) (0.022) (0.023)

Physical and mental 0.009 0.000 0.012 0.018 0.020

disabilities (0.029) (0.016) (0.016) (0.014) (0.014)

Behavioral problems 0.023 -0.035*** -0.008 -0.001 0.000

(0.014) (0.008) (0.008) (0.008) (0.008)

Student fixed effects ! ! ! !

Course fixed effects !

Course*year fixed effects ! !

Cohort-level controls !

Note: Each column represents a separate regression with the number of specific SEN students asindependent variables. Cohort controls include cohort (course) size, percentage of students with addi-tional support in pre-vocational education, percentage of boys, percentage of students from differentethnicities, percentage of students from disadvantaged neighborhoods, mean age and mean number ofcourses. Standard errors are reported in parentheses. Standard errors are robust and clustered at theschool level. ***p<0.01 **p<0.05 *p<0.10.

136

Tab

le5.iii:Schoo

lfixedeff

ectestimates

oftheeff

ectof

thenu

mber

ofspecifictypes

ofSEN

students

oncito

participation

andcitoscore(primaryed

ucation

)(1)

(2)

(3)

(4)

(5)

(1)

(2)

(3)

(4)

(5)

Citopa

rticipation

Citoscore

Visua

lproblem

s0.004

-0.012

-0.012

-0.011

0.017

0.012

-0.003

-0.006

-0.004

0.003

(0.023)

(0.023)

(0.023)

(0.023)

(0.030)

(0.029)

(0.023)

(0.022)

(0.022)

(0.030)

Hearing

prob

lems

0.004

-0.005

-0.005

-0.005

-0.009

0.011

-0.001

-0.000

0.001

-0.010

(0.009)

(0.010)

(0.010)

(0.010)

(0.012)

(0.012)

(0.010)

(0.009)

(0.009)

(0.012)

Phy

sicala

ndmental

-0.014

-0.006

-0.006

-0.006

-0.001

0.024***

0.004

0.003

0.005

-0.004

disabilities

(0.010)

(0.008)

(0.008)

(0.008)

(0.010)

(0.009)

(0.008)

(0.007)

(0.007)

(0.009)

Behavioralp

roblem

s0.010**

0.002

0.002

0.003

0.007

0.051***

-0.002

-0.004

-0.002

0.001

(0.004)

(0.005)

(0.005)

(0.005)

(0.006)

(0.005)

(0.004)

(0.004)

(0.004)

(0.005)

Yearfix

edeff

ects

!!

!!

!!

!!

!!

Scho

olfix

edeff

ects

!!

!!

!!

!!

Individu

alcontrols

!!

!!

!!

Coh

ortlevelc

ontrols

!!

!!

Scho

olspec.timetrend

!!

Note:

Eachcolumnrepresents

asepa

rate

regression

withthenu

mbe

rof

specificSE

Nstud

ents

asindepe

ndentvariab

les.

Individu

alcontrols

includ

egend

er,a

ge,

ethn

icityan

dstud

entweigh

t.Coh

ortmeancontrols

includ

ecoho

rtsize

andstud

entindividu

alcontrols

averaged

bystud

ents’scho

olan

dyear.Stan

dard

errors

arerepo

rted

inpa

rentheses.

Stan

dard

errors

arerobu

stan

dclusteredat

thescho

ollevel.

***p

<0.01

**p<

0.05

*p<0.10.

137

Table 5.iv: School fixed effect estimates of the effect of the number ofspecific types of SEN students on standardized central exam grades fordifferent secondary school tracks

(1) (2) (3) (4) (5)

Pre-voc. I

Visual problems 0.020 0.024 0.042 0.039 0.044

(0.057) (0.064) (0.063) (0.065) (0.092)

Hearing problems -0.044*** -0.045*** -0.043*** -0.043*** -0.017

(0.016) (0.012) (0.012) (0.013) (0.017)

Physical and mental -0.048* 0.044** 0.037* 0.037* 0.036

disabilities (0.025) (0.020) (0.020) (0.019) (0.026)

Behavioral problems -0.001 0.010 0.008 0.008 -0.004

(0.013) (0.007) (0.008) (0.008) (0.010)

Pre-voc. II

Visual problems -0.019 0.050 0.053 0.056 0.070

(0.091) (0.082) (0.078) (0.077) (0.109)

Hearing problems 0.076*** 0.024 0.026 0.034 0.018

(0.026) (0.021) (0.021) (0.021) (0.026)

Physical and mental -0.020 -0.034 -0.033 -0.027 -0.008

disabilities (0.027) (0.025) (0.024) (0.024) (0.032)

Behavioral problems 0.002 0.003 -0.000 0.004 0.005

(0.012) (0.009) (0.008) (0.008) (0.013)

Pre-voc. III

Visual problems 0.225*** -0.136 -0.134 -0.112 -0.274

(0.055) (0.085) (0.083) (0.140) (0.174)

Hearing problems 0.220*** 0.075 0.057 0.034 -0.057

(0.082) (0.070) (0.069) (0.085) (0.100)

Physical and mental 0.178** 0.058 0.058 0.043 0.080

disabilities (0.078) (0.066) (0.062) (0.064) (0.086)

Behavioral problems 0.066** -0.015 -0.024 -0.025 0.015

(0.028) (0.032) (0.031) (0.030) (0.056)

Year fixed effects ! ! ! ! !

School fixed effects ! ! ! !

Individual controls ! ! !

Cohort level controls ! !

School spec. time trend !

(Table continues on next page)

138

Table 5.iv: (continued)(1) (2) (3) (4) (5)

Pre-voc. IV

Visual problems 0.042 0.015 0.028 0.024 0.011

(0.043) (0.029) (0.031) (0.029) (0.040)

Hearing problems -0.033 -0.016 -0.021 -0.015 0.001

(0.029) (0.017) (0.017) (0.017) (0.022)

Physical and mental 0.020 0.024 0.016 0.023 0.036

disabilities (0.019) (0.017) (0.017) (0.017) (0.026)

Behavioral problems 0.017** -0.001 -0.003 -0.000 -0.005

(0.008) (0.006) (0.006) (0.006) (0.007)

Senior gen.

Visual problems -0.011 -0.006 -0.009 -0.007 0.019

(0.030) (0.022) (0.021) (0.021) (0.028)

Hearing problems -0.008 0.005 0.005 0.005 -0.024

(0.024) (0.022) (0.021) (0.021) (0.023)

Physical and mental 0.022* -0.006 -0.009 -0.006 -0.002

disabilities (0.012) (0.013) (0.013) (0.013) (0.015)

Behavioral problems -0.008 -0.013* -0.014** -0.013** -0.005

(0.008) (0.006) (0.007) (0.006) (0.009)

Pre-uni.

Visual problems 0.100** -0.034 0.016 -0.022 -0.012

(0.046) (0.026) (0.027) (0.025) (0.032)

Hearing problems 0.012 -0.015 -0.003 -0.011 0.010

(0.033) (0.030) (0.031) (0.027) (0.038)

Physical and mental 0.022 0.002 -0.006 0.008 -0.004

disabilities (0.038) (0.016) (0.016) (0.015) (0.015)

Behavioral problems 0.054*** -0.007 -0.010 0.001 -0.013

(0.019) (0.010) (0.011) (0.010) (0.012)

Year fixed effects ! ! ! ! !

School fixed effects ! ! ! !

Individual controls ! ! !

Cohort level controls ! !

School spec. time trend !

Note: Each column represents a separate regression with the number of specific SEN students asindependent variables. Individual controls include additional support in pre-vocational education,gender, age, ethnicity, disadvantaged neighborhood and number of courses. Cohort mean controlsinclude cohort size and student individual controls averaged by students’ school, track and year.Standard errors are reported in parentheses. Standard errors are robust and clustered at the schoollevel. ***p<0.01 **p<0.05 *p<0.10.

139

Table 5.v: Estimates of the effect of the number of SEN students on citoparticipation and citoscore at the cohort level (primary education)

(1) (2) (3) (4)

Mean cito participation 0.001 -0.001 0.001 0.002

(0.001) (0.002) (0.002) (0.002)

Mean citoscore 0.041*** -0.000 0.003 0.004

(0.004) (0.003) (0.003) (0.005)

SD -0.005** -0.001 -0.002 -0.000

(0.002) (0.002) (0.002) (0.003)

10th percentile 0.039*** -0.001 0.006 0.005

(0.007) (0.006) (0.006) (0.008)

50th percentile 0.049*** -0.002 0.000 0.002

(0.005) (0.004) (0.004) (0.006)

90th percentile 0.036*** 0.004 0.003 0.006

(0.003) (0.003) (0.003) (0.005)

Year fixed effects ! ! ! !

School fixed effects ! ! !

Cohort level controls ! !

School specific time trend !

Note: Each coefficient represents a separate regression with the number of SEN students as indepen-dent variable. The first row of coefficients refers to mean cito participation, all other coefficients referto aggregated measures of citoscore. Cohort controls include cohort size, percentage of boys, percent-age of students from different ethnicities, percentage of weighted students and mean age. Regressionsare weighted by the mean cohort size within schools. Standard errors are reported in parentheses.Standard errors are robust and clustered at the school level. ***p<0.01 **p<0.05 *p<0.10.

140

Table 5.vi: Estimates of the effect of the number of SEN students on stan-dardized central exam grades at the cohort level (secondary education)

(1) (2) (3) (4)

Pre-voc. I

Mean -0.012 0.006 0.004 -0.000

(0.010) (0.007) (0.007) (0.009)

SD 0.002 -0.006 -0.007* -0.007

(0.003) (0.004) (0.004) (0.006)

10th percentile -0.019 0.011 0.010 0.001

(0.012) (0.010) (0.011) (0.014)

50th percentile -0.009 0.005 0.001 -0.002

(0.009) (0.008) (0.008) (0.010)

90th percentile -0.009 0.004 0.001 -0.009

(0.010) (0.008) (0.008) (0.012)

Pre-voc. II

Mean 0.002 -0.001 0.003 0.005

(0.010) (0.008) (0.008) (0.010)

SD 0.004 0.001 -0.000 0.003

(0.003) (0.003) (0.003) (0.005)

10th percentile -0.002 -0.001 0.004 0.005

(0.013) (0.009) (0.010) (0.013)

50th percentile 0.001 -0.002 0.002 0.000

(0.010) (0.009) (0.009) (0.011)

90th percentile 0.005 0.000 0.002 0.010

(0.011) (0.010) (0.010) (0.013)

Pre-voc. III

Mean 0.100*** 0.011 -0.004 0.028

(0.033) (0.026) (0.025) (0.043)

SD 0.026*** 0.005 0.006 0.002

(0.009) (0.010) (0.010) (0.016)

10th percentile 0.065** -0.006 -0.021 0.048

(0.030) (0.033) (0.032) (0.052)

50th percentile 0.092*** 0.005 -0.007 0.030

(0.034) (0.026) (0.026) (0.043)

90th percentile 0.164*** 0.061* 0.042 0.045

(0.042) (0.035) (0.035) (0.056)

Year fixed effects ! ! ! !

School fixed effects ! ! !

Cohort level controls ! !

School specific time trend !

(Table continues on next page)

141

Table 5.vi: (continued)(1) (2) (3) (4)

Pre-voc. IV

Mean 0.012* 0.002 0.002 0.001

(0.007) (0.006) (0.006) (0.008)

SD 0.007** 0.006** 0.005* 0.007*

(0.003) (0.003) (0.003) (0.004)

10th percentile 0.007 -0.003 -0.002 -0.010

(0.009) (0.009) (0.009) (0.010)

50th percentile 0.011 0.006 0.007 0.008

(0.007) (0.006) (0.006) (0.008)

90th percentile 0.023*** 0.007 0.004 0.004

(0.007) (0.007) (0.007) (0.011)

Senior gen.

Mean -0.003 -0.009* -0.010** -0.006

(0.007) (0.005) (0.005) (0.007)

SD -0.004** -0.001 -0.000 -0.002

(0.002) (0.002) (0.002) (0.004)

10th percentile 0.003 -0.007 -0.008 -0.001

(0.009) (0.007) (0.007) (0.011)

50th percentile -0.003 -0.011** -0.012** -0.007

(0.007) (0.005) (0.005) (0.006)

90th percentile -0.009 -0.010 -0.010 -0.009

(0.007) (0.007) (0.007) (0.009)

Pre-uni.

Mean 0.042*** -0.009 -0.000 -0.009

(0.015) (0.008) (0.007) (0.009)

SD 0.006 0.002 0.002 -0.003

(0.004) (0.004) (0.004) (0.006)

10th percentile 0.032** -0.015* -0.005 -0.004

(0.013) (0.009) (0.009) (0.012)

50th percentile 0.039*** -0.006 0.002 -0.005

(0.014) (0.008) (0.008) (0.010)

90th percentile 0.052** -0.006 0.006 -0.011

(0.021) (0.014) (0.014) (0.018)

Year fixed effects ! ! ! !

School fixed effects ! ! !

Cohort level controls ! !

School specific time trend !

Note: Each coefficient represents a separate regression with the number of SEN students as indepen-dent variable. Cohort controls include cohort size, percentage of students with additional supportin pre-vocational education, percentage of boys, percentage of students from different ethnicities,percentage of students from disadvantaged neighborhoods, mean age and mean number of courses.Regressions are weighted by the mean cohort size within schools. Standard errors are reported inparentheses. Standard errors are robust and clustered at the school level. ***p<0.01 **p<0.05*p<0.10. 142

Tab

le5.vii:

Estim

ates

oftheeff

ectof

thepercentageof

SEN

students

onstan

dardized

centralexam

grad

es(student

fixedeff

ects)

(1)

(2)

(3)

(4)

(5)

NN

Observation

sStud

ents

Pre-voc.I

0.001

0.003**

0.002*

0.002*

0.002

271,376

55,957

(0.002)

(0.001)

(0.001)

(0.001)

(0.001)

Pre-voc.II

0.003

-0.005***

0.000

0.000

0.000

355,213

70,277

(0.002)

(0.002)

(0.002)

(0.002)

(0.002)

Pre-voc.III

0.009***

0.003

0.005**

0.005**

0.005*

93,314

14,960

(0.003)

(0.003)

(0.002)

(0.003)

(0.003)

Pre-voc.IV

0.002

-0.002

0.002

0.001

0.002

780,815

125,018

(0.002)

(0.001)

(0.001)

(0.001)

(0.001)

Senior

gen.

-0.001

-0.004

0.001

0.001

0.001

972,532

143,844

(0.003)

(0.002)

(0.002)

(0.002)

(0.002)

Pre-uni.

0.004

-0.005**

-0.001

-0.001

-0.001

820,551

108,829

(0.003)

(0.002)

(0.002)

(0.002)

(0.002)

Stud

entfix

edeff

ects

!!

!!

Cou

rsefix

edeff

ects

!

Cou

rse*year

fixed

effects

!!

Coh

ort-levelc

ontrols

!

Note:

Eachcoeffi

cientrepresents

asepa

rate

regression

withthepe

rcentage

ofSE

Nstud

ents

asindepe

ndentvariab

le.Coh

ortcontrols

includ

e(cou

rse)

coho

rtsize,p

ercentageof

stud

ents

withad

dition

alsupp

ortin

pre-vo

cation

aleducation,

percentage

ofbo

ys,p

ercentageof

stud

ents

from

diffe

rent

ethn

icities,pe

rcentage

ofstud

ents

from

disadv

antagedneighb

orho

ods,

meanagean

dmeannu

mbe

rof

courses.

Stan

dard

errors

arerepo

rted

inpa

renthe

ses.

Stan

dard

errors

arerobu

stan

dclusteredat

thescho

ollevel.***p

<0.01

**p<

0.05

*p<0.10.

143

Table 5.viii: Estimates of the effect of the percentage of specific types ofSEN students on standardized central exam grades (student fixed effects)

Pre-voc. I Pre-voc. II Pre-voc. III Pre-voc. IV Senior gen. Pre-uni.

Visual problems -0.003 0.007 0.042*** 0.005 -0.010 -0.006

(0.006) (0.011) (0.006) (0.008) (0.007) (0.006)

Hearing problems 0.005 -0.002 -0.005 -0.001 -0.007 -0.008

(0.003) (0.004) (0.005) (0.004) (0.007) (0.005)

Physical and mental 0.003 -0.002 0.010 0.003 0.003 0.004

disabilities (0.003) (0.004) (0.007) (0.003) (0.003) (0.004)

Behavioral problems 0.001 0.001 0.004 0.001 0.001 -0.001

(0.002) (0.002) (0.003) (0.001) (0.002) (0.003)

N observations 271,376 355,213 93,314 780,815 972,532 820,551

N students 55,957 70,277 14,960 125,018 143,844 108,829Note: Each column represents a separate regression with the percentage of specific SEN students asindependent variables. All regressions include student fixed effects, course*year fixed effects and cohortcontrols. Cohort controls include (course) cohort size, percentage of students with additional supportin pre-vocational education, percentage of boys, percentage of students from different ethnicities,percentage of students from disadvantaged neighborhoods, mean age and mean number of courses.Standard errors are reported in parentheses. Standard errors are robust and clustered at the schoollevel. ***p<0.01 **p<0.05 *p<0.10.

144

Tab

le5.ix:Estim

ates

oftheeff

ectof

thenu

mber

ofSEN

students

onstan

dardized

centralexam

grad

esin

courses

with

30or

fewer

students

(studentfixedeff

ects)

(1)

(2)

(3)

(4)

(5)

NN

Observation

sStud

ents

Pre-voc.I

0.022*

0.019

-0.013

-0.012

0.004

114,775

50,508

(0.012)

(0.020)

(0.020)

(0.019)

(0.020)

Pre-voc.II

0.005

-0.099***

-0.024

-0.024

-0.001

120,496

58,808

(0.013)

(0.023)

(0.023)

(0.023)

(0.024)

Pre-voc.III

0.093***

0.033

0.033

0.030

0.039

60,381

13,912

(0.024)

(0.028)

(0.026)

(0.028)

(0.027)

Pre-voc.IV

0.017**

0.001

0.008

0.009

0.014

200,351

92,669

(0.008)

(0.010)

(0.009)

(0.009)

(0.009)

Senior

gen.

-0.001

-0.010

0.001

-0.002

0.001

172,441

85,807

(0.013)

(0.017)

(0.016)

(0.015)

(0.015)

Pre-uni.

-0.016

-0.043**

-0.025

-0.020

-0.009

254,943

83,249

(0.018)

(0.019)

(0.017)

(0.017)

(0.017)

Stud

entfix

edeff

ects

!!

!!

Cou

rsefix

edeff

ects

!

Cou

rse*year

fixed

effects

!!

Coh

ort-levelc

ontrols

!

Note:

Eachcoeffi

cientrepresents

asepa

rate

regression

withthenu

mbe

rof

SEN

stud

ents

asindepe

ndentvariab

le.Coh

ortcontrols

includ

e(cou

rse)

coho

rtsize,

percentage

ofstud

ents

withad

dition

alsupp

ortin

pre-vo

cation

aleducation,

percentage

ofbo

ys,pe

rcentage

ofstud

ents

from

diffe

rent

ethn

icities,

percentage

ofstud

ents

from

disadv

antagedneighb

orho

ods,meanagean

dmeannu

mbe

rof

courses.

Stan

dard

errors

arerepo

rted

inpa

rentheses.

Stan

dard

errors

arerobu

stan

dclusteredat

thescho

ollevel.***p

<0.01

**p<

0.05

*p<0.10.

145

Table 5.x: Estimates of the effect of the number of specific types of SENstudents on standardized central exam grades in courses with 30 or fewerstudents (student fixed effects)

Pre-voc. I Pre-voc. II Pre-voc. III Pre-voc. IV Senior gen. Pre-uni.

Visual problems 0.040 0.082 0.413*** 0.026 -0.078 -0.124**

(0.101) (0.152) (0.120) (0.045) (0.058) (0.052)

Hearing problems 0.027 -0.043 -0.098 -0.025 0.013 -0.106

(0.053) (0.059) (0.108) (0.033) (0.055) (0.069)

Physical and mental 0.083* -0.045 0.084 0.045 -0.033 0.058**

disabilities (0.046) (0.061) (0.072) (0.028) (0.023) (0.029)

Behavioral problems -0.023 0.015 0.034 0.013 0.023 -0.007

(0.024) (0.027) (0.030) (0.010) (0.019) (0.023)

N observations 114,775 120,496 60,381 200,351 172,441 254,943

N students 50,508 58,808 13,912 92,669 85,807 83,249Note: Each column represents a separate regression with the number of specific SEN students asindependent variables. All regressions include student fixed effects, course*year fixed effects and cohortcontrols. Cohort controls include (course) cohort size, percentage of students with additional supportin pre-vocational education, percentage of boys, percentage of students from different ethnicities,percentage of students from disadvantaged neighborhoods, mean age and mean number of courses.Standard errors are reported in parentheses. Standard errors are robust and clustered at the schoollevel. ***p<0.01 **p<0.05 *p<0.10.

146

Tab

le5.xi:Estim

ates

oftheeff

ectof

thepercentageof

SEN

students

onstudentachievement(schoo

lfixedeff

ects)

(1)

(2)

(3)

(4)

(5)

NClusters

Citopa

rticipation

-0.001

-0.001

-0.001

-0.001

0.000

462,227

5,958

(Primaryeducation)

(0.001)

(0.001)

(0.001)

(0.001)

(0.001)

Stan

dardized

citoscore

0.006***

0.000

-0.000

-0.000

-0.000

377,135

5,861

(Primaryeducation)

(0.001)

(0.001)

(0.001)

(0.001)

(0.001)

Pre-voc.I

0.000

-0.001

-0.002

-0.001

-0.001

56,048

441

(0.004)

(0.004)

(0.004)

(0.004)

(0.005)

Pre-voc.II

0.006

-0.001

-0.001

-0.000

0.004

70,283

451

(0.006)

(0.005)

(0.005)

(0.005)

(0.007)

Pre-voc.III

0.020***

0.005

0.002

0.003

0.010

14,955

299

(0.007)

(0.005)

(0.005)

(0.005)

(0.008)

Pre-voc.IV

0.007**

-0.001

-0.002

-0.001

-0.000

125,023

741

(0.003)

(0.003)

(0.003)

(0.003)

(0.003)

Senior

gen.

-0.003

-0.010*

-0.011**

-0.011**

-0.006

143,847

500

(0.007)

(0.005)

(0.005)

(0.005)

(0.006)

Pre-uni.

0.013

-0.003

-0.003

-0.003

-0.006

108,829

516

(0.009)

(0.007)

(0.007)

(0.006)

(0.008)

Yearfix

edeff

ects

!!

!!

!

Scho

olfix

edeff

ects

!!

!!

Individu

alcontrols

!!

!

Coh

ortlevelc

ontrols

!!

Scho

olspecifictimetrend

!

Note:

Eachcoeffi

cientrepresents

asepa

rate

regression

withthepe

rcentage

ofSE

Nstud

ents

asindepe

ndentvariab

le.Individu

alcontrols

includ

egend

er,age,

ethn

icityan

dstud

entweigh

tforprim

aryeducation,

andad

dition

alsupp

ortin

pre-vo

cation

aleducation,

gend

er,a

ge,e

thnicity,d

isad

vantaged

neighb

orho

odan

dnu

mbe

rof

coursesforsecond

aryeducation.

Coh

ortmeancontrols

includ

ecoho

rtsize

andstud

entindividu

alcontrols

averaged

bystud

ents’scho

ol,trackan

dyear.Stan

dard

errors

arerepo

rted

inpa

rentheses.

Stan

dard

errors

arerobu

stan

dclusteredat

thescho

ollevel.

***p

<0.01

**p<

0.05

*p<0.10.

147

Tab

le5.xii:

Estim

ates

oftheeff

ectof

thepercentageof

specificSEN

students

onstudentachievement(schoo

lfixed

effects

mod

els)

Primaryeducation

Second

aryeducation

Citopa

rticipation

Citoscore

Pre-voc.I

Pre-voc.II

Pre-voc.III

Pre-voc.IV

Senior

gen.

Pre-uni.

Visua

lproblem

s0.003

0.006

0.019

-0.001

-0.092

-0.025

0.028

-0.010

(0.006)

(0.008)

(0.041)

(0.077)

(0.107)

(0.023)

(0.031)

(0.024)

Hearing

prob

lems

-0.002

-0.002

-0.011

-0.004

0.013

-0.001

-0.032

0.012

(0.003)

(0.003)

(0.011)

(0.020)

(0.026)

(0.016)

(0.024)

(0.042)

Phy

sicala

ndmental

-0.002

-0.003

0.012

-0.011

0.026*

0.019*

-0.005

0.001

disabilities

(0.002)

(0.003)

(0.013)

(0.019)

(0.016)

(0.011)

(0.015)

(0.012)

Behavioralp

roblem

s0.001

0.000

-0.003

0.008

0.004

-0.004

-0.005

-0.008

(0.001)

(0.001)

(0.006)

(0.007)

(0.009)

(0.004)

(0.008)

(0.010)

Nstud

ents

462,227

377,135

56,048

70,283

14,955

125,023

143,847

108,829

Nscho

ols

5,958

5,861

441

451

299

741

500

516

Note:

Eachcolumnrepresents

asepa

rate

regression

withthepe

rcentage

ofspecificSE

Nstud

ents

asindepe

ndentvariab

les.

Allregression

sinclud

eyear

fixed

effects,scho

olfix

edeff

ects,individu

alan

dcoho

rtcontrols

andascho

olspecifictimetrendin

thenu

mbe

rof

SEN

stud

ents.Individu

alcontrols

includ

egend

er,

age,

ethn

icityan

dstud

entweigh

tforprim

aryeducation,

andad

dition

alsupp

ortin

pre-vo

cation

aleducation,

gend

er,a

ge,e

thnicity,d

isad

vantaged

neighb

orho

odan

dnu

mbe

rof

coursesforsecond

aryeducation.

Coh

ortmeancontrolsinclud

ecoho

rtsize

andstud

entindividu

alcontrolsaveraged

bystud

ents’schoo

l,trackan

dyear.Stan

dard

errors

arerepo

rted

inpa

rentheses.

Stan

dard

errors

arerobu

stan

dclusteredat

thescho

ollevel.

***p

<0.01

**p<

0.05

*p<0.10.

148

Tab

le5.xiii:Estim

ates

oftheeff

ectof

thenu

mber

ofSEN

students

onstudentachievementin

smallschoo

ls(schoo

lfixedeff

ects)

(1)

(2)

(3)

(4)

(5)

NClusters

Citopa

rticipation

-0.009

-0.008

-0.009

-0.007

0.013

128,065

2,920

(Primaryeducation)

(0.006)

(0.009)

(0.009)

(0.009)

(0.011)

Stan

dardized

citoscore

0.034***

-0.003

-0.007

-0.006

0.005

104,323

2,671

(Primaryeducation)

(0.009)

(0.009)

(0.009)

(0.009)

(0.012)

Pre-voc.I

-0.020

-0.036

-0.031

-0.031

-0.026

6,929

195

(0.025)

(0.032)

(0.030)

(0.030)

(0.034)

Pre-voc.II

0.002

-0.015

-0.011

-0.024

0.021

5,376

145

(0.036)

(0.051)

(0.050)

(0.052)

(0.074)

Pre-voc.III

0.138***

-0.022

-0.033

-0.028

-0.022

6,974

264

(0.045)

(0.043)

(0.043)

(0.042)

(0.061)

Pre-voc.IV

0.048**

-0.037**

-0.041**

-0.023

-0.016

8,218

227

(0.023)

(0.017)

(0.016)

(0.018)

(0.019)

Senior

gen.

-0.316***

-0.261**

-0.281**

-0.408**

-0.385*

1,292

41

(0.089)

(0.110)

(0.114)

(0.166)

(0.199)

Pre-uni.

0.112

0.049

0.019

0.008

-0.068

3,039

76

(0.122)

(0.105)

(0.094)

(0.092)

(0.117)

Yearfix

edeff

ects

!!

!!

!

Scho

olfix

edeff

ects

!!

!!

Individu

alcontrols

!!

!

Coh

ortlevelc

ontrols

!!

Scho

olspecifictimetrend

!

Note:

Eachcoeffi

cientrepresents

asepa

rate

regression

withthenu

mbe

rof

SEN

stud

ents

asindepe

ndentvariab

le.Sm

allschoo

lsaredefin

edas

scho

olswithcoho

rts

withbe

tween20

and30

stud

ents

inprim

aryeduc

ation,

andcoho

rtswith30

orfewer

stud

ents

insecond

aryeducation.

The

diffe

rent

defin

itionis

chosen

because

very

smallcoho

rtsin

prim

aryeducationarelik

elyto

bein

thesameclassas

ayoun

gercoho

rt.Individu

alcontrols

includ

egend

er,age,

ethn

icityan

dstud

ent

weigh

tforprim

aryeducation,

andad

dition

alsupp

ortin

pre-vo

cation

aleducation,

gend

er,a

ge,e

thnicity,d

isad

vantaged

neighb

orho

odan

dnu

mbe

rof

coursesfor

second

aryeducation.

Coh

ortmeancontrolsinclud

ecoho

rtsize

andstud

entindividu

alcontrolsaveraged

bystud

ents’schoo

l,trackan

dyear.Stan

dard

errors

are

repo

rted

inpa

rentheses.

Stan

dard

errors

arerobu

stan

dclusteredat

thescho

ollevel.***p

<0.01

**p<

0.05

*p<0.10.

149

Tab

le5.xiv:

Estim

ates

oftheeff

ectof

thenu

mber

ofspecificSEN

students

onstudentachievementin

smallschoo

ls(schoo

lfixedeff

ects

mod

els)

Primaryeducation

Second

aryeducation

Citopa

rticipation

Citoscore

Pre-voc.I

Pre-voc.II

Pre-voc.III

Pre-voc.IV

Senior

gen.

Pre-uni.

Visua

lproblem

s0.035

0.114*

-0.337

0.426***

-0.039

0.361

-0.261

(0.056)

(0.067)

(0.208)

(0.157)

(0.128)

(0.313)

(0.362)

Hearing

prob

lems

-0.017

0.022

0.006

-0.419

0.291

0.061

-0.800**

-0.711

(0.024)

(0.030)

(0.108)

(0.253)

(0.313)

(0.149)

(0.334)

(0.440)

Phy

sicala

ndmental

0.010

-0.030

-0.081

-0.183

0.182

-0.005

-1.042***

0.042

disabilities

(0.023)

(0.025)

(0.098)

(0.113)

(0.138)

(0.092)

(0.237)

(0.113)

Behavioralp

roblem

s0.020

0.008

-0.011

0.129*

-0.130*

-0.024

0.116

-0.080

(0.014)

(0.016)

(0.040)

(0.075)

(0.071)

(0.021)

(0.271)

(0.132)

Nstud

ents

128,065

104,323

6,929

5,376

6,974

8,218

1,292

3,039

Nscho

ols

2,920

2,671

195

145

264

227

4176

Note:

Eachcolumnrepresents

asepa

rate

regression

withthenu

mbe

rof

specificSE

Nstud

ents

asindepe

ndentvariab

les.

Allregression

sinclud

eyear

fixed

effects,

scho

olfix

edeff

ects,individu

alan

dcoho

rtcontrols

andascho

olspecifictimetrendin

thenu

mbe

rof

SEN

stud

ents.Individu

alcontrols

includ

egend

er,age,

ethn

icityan

dstud

entweigh

tforprim

aryeducation,

andad

dition

alsupp

ortin

pre-vo

cation

aleducation,

gend

er,a

ge,e

thnicity,d

isad

vantaged

neighb

orho

odan

dnu

mbe

rof

coursesforsecond

aryeducation.

Coh

ortmeancontrols

includ

ecoho

rtsize

andstud

entindividu

alcontrols

averaged

bystud

ents’scho

ol,trackan

dyear.Stan

dard

errors

arerepo

rted

inpa

rentheses.

Stan

dard

errors

arerobu

stan

dclusteredat

thescho

ollevel.

***p

<0.01

**p<

0.05

*p<0.10.

150

Chapter 6

Summary

This thesis consists of four empirical studies investigating questions in the field of theeconomics of education. This chapter summarizes the conclusions from the previouschapters and discusses their relevance for the education field. Remarkably, all chap-ters provide evidence against at least one popular argument in the public debate oneducation.

Chapter 2 investigates secondary school choice in Amsterdam. Specifically, it stud-ies whether information about school quality, distance, peers and conducting a lotteryin the previous year predict students’ school choices. The results in this chapter yieldsome indications that students avoid schools that had a school admission lottery in theprevious year. Further, the findings are consistent with the notion that there are peereffects in secondary school choice: students appear to prefer schools that are chosenby many of their former classmates in primary education. Moreover, students pre-fer schools close to their home while information about school quality and changes inschool quality are not consistent predictors of school choice.

The findings on the relevance of school quality in school choice contrast to the oftenmade ’quality-argument’ in the education debate on school choice. While the Nether-lands has a long tradition of free school choice, there is a public debate on whetherschool choice should be increased or not in countries with more restrictive policies, forinstance in the US and the UK. One of the main arguments in this discussion is thatchoice creates competition between schools. It is argued that parents and studentsprefer high quality schools, such that school choice creates an incentive for schools toimprove their quality, which would improve the overall quality of education. For thischannel to work, it is important that students and parents indeed base their choiceson school quality. The findings in Chapter 2 indicate that this mechanism might notoperate; at least not in the Amsterdam context with the current Dutch school qualitymeasures.

An important consequence of having free school choice in Amsterdam is that some

151

schools are oversubscribed: the number of students choosing a school sometimes exceedsthe number of students that can be enrolled. The oversubscribed schools are not allowedto ’cherry-pick’ students: they have to conduct school admission lotteries to allocatethe available places. Students who lose the school admission lottery in the first roundof the application process have to find a school that still has seats available, whichimplies that the school of their second or third choice may no longer be an option.

Every year, the school admission lotteries are prominently covered and criticizedin the local news. Chapter 3 investigates whether losing a school admission lotteryactually affects students’ school performance. Comparing students who win and losea school admission lottery, it turns out that students’ academic achievement is notharmed by losing a school admission lottery. After four years of secondary education,lottery losing students even seem to achieve somewhat better than lottery winningstudents. As most students in this study are too young to have finished secondaryschool, it is yet unclear whether the results carry over to the final exams. Nevertheless,the results can be seen as provisional reassurance for (parents of) students losing schooladmission lotteries: after four years of secondary education, the academic achievementof lottery losing students measure up to those of lottery winning students.

Chapter 4 also investigates school admission lotteries, albeit in a different context:it exploits admission lotteries to study the effects of Montessori secondary education.Montessori secondary education differs from regular education by allowing students alarge amount of choice in their school days. Although many students in many differ-ent countries are educated using the Montessori pedagogy, little is known about theeffectiveness of Montessori methods, especially in secondary education.

For parents, students and policy makers, however, this is very important informa-tion. When Montessori education has positive effects on students, schools might beimproved by a broader implementation of Montessori methods. On the other hand,if Montessori education negatively affects students, students and parents could betteropt for regular education schools.

The results of Chapter 4 indicate that Montessori education provides an alternativeroute towards towards similar student outcomes. Montessori students obtain theirsecondary school degree without delay at the same rate and with similar grades asnon-Montessori students, although the route towards the exams is somewhat different.This contrasts with claims that Montessori education may harm academic achievementby a lack of structure and academic standards. On the other hand, students exposedto Montessori secondary education do not score better on measures of independenceand motivation, even though these are the main characteristics Montessori educationclaims to foster.

Chapter 5 is loosely related to Chapter 2 by focussing on classmates. While Chap-

152

ter 2 studied peer effects in school choice, this chapter focusses on the classmates ofstudents with special educational needs. The presence of students with special educa-tional needs (such as visual or behavioral problems) in regular education classrooms,also called inclusive education, is an often discussed issue in the education field. Onthe one hand, teachers and parents are worried that students get distracted by thebehavior of students with special educational needs and argue that special needs stu-dents demand more teacher attention at the expense of regular students. Proponentsof inclusive education, however, state that there is more teacher support in inclusiveclassrooms, which could positively affect the regular education students.

Chapter 5 studies the impact of inclusive education on the academic achievement ofstudents without special educational needs. It is found that under the Dutch ’backpack’policy, the presence of students with special educational needs does not adversely orfavorably affect the academic achievement of regular education students. This contraststo the popular beliefs described above and to earlier research findings. Internationaleconomic research on this topic generally finds that the presence of students with specialeducational needs harms the academic achievement of their regular education peers.A possible explanation for this difference is the fact that inclusive education is highlysubsidized in the Netherlands.

It is important to stress that the results in Chapter 5 are based on the Dutch’backpack’ policy. When Dutch schools radically change their inclusion policies in thewake of the new policy called ’passend onderwijs’, the impact on regular educationstudents might change. Nevertheless, this study adds an interesting nuance to theinclusive education debate: in a situation with substantial additional funding, inclusiveeducation does not harm the academic achievement of regular education students.

153

Bibliography

Abdulkadiroğlu, A., Angrist, J., & Pathak, P. (2014). The elite illusion: Achievementeffects at Boston and New York exam schools. Econometrica, 82(1), 137–196.

Abdulkadiroğlu, A., Pathak, P. A., & Roth, A. E. (2009). Strategy-proofness versusefficiency in matching with indifferences: Redesigning the NYC high school match.American Economic Review, 99(5), 1954–1978.

Abdulkadiroğlu, A., Pathak, P. A., Roth, A. E., & Sönmez, T. (2005). The Bostonpublic school match. American Economic Review, 95(2), 368–371.

Abdulkadiroğlu, A. & Sönmez, T. (2003). School choice: A mechanism design approach.American Economic Review, 93(3), 729–747.

Ainscow, M. & César, M. (2006). Inclusive education ten years after Salamanca: Settingthe agenda. European Journal of Psychology of Education, 21(3), 231–238.

Aizer, A. (2008). Peer effects and human capital accumulation: The externalities ofADD. NBER Working Paper 14354, National Bureau of Economic Research.

Ammermueller, A. & Pischke, J. S. (2009). Peer effects in European primary schools:Evidence from the Progress in International Reading Literacy Study. Journal ofLabor Economics, 27(3), 315–348.

Anderson, R. A., Worthington, L., Anderson, W. T., & Jennings, G. (1994). Thedevelopment of an autonomy scale. Contemporary Family Therapy, 16(4), 329–345.

Angrist, J. D. (2014). The perils of peer effects. Labour Economics, 30, 98–108.

Association Montessori Internationale (2013). Montessori movement. Retrieved 12August, 2013, from http://www.montessori-ami.org/.

Beckert, T. E. (2007). Cognitive autonomy and self-evaluation in adolescence: Aconceptual investigation and instrument development. North American Journal ofPsychology, 9(3), 579–594.

155

Böhlmark, A. & Lindahl, M. (2013). Independent schools and long-run educationaloutcomes: Evidence from Sweden’s large scale voucher reform. IZA Discussion Paper6683, Institute for the Study of Labor.

Booij, A., Leuven, E., & Oosterbeek, H. (2014). Ability peer effects in university:Evidence from a randomized experiment. Working paper, University of Amsterdam.

Buser, T., Niederle, M., & Oosterbeek, H. (2014). Gender, competitiveness and careerchoices. The Quarterly Journal of Economics, 129(3), 1409–1447.

Calff, J. (1980). Van pionier tot mammoet: Het Amsterdams Montessori Lyceum 1930-1980. Amsterdam: Stichting Montessori 50.

Calsamiglia, C. & Guell, M. (2014). The illusion of school choice: empirical evidencefrom Barcelona. IZA Discussion Paper 8202, Institute for the Study of Labor.

Carrell, S. E. & Hoekstra, M. L. (2010). Externalities in the classroom: How childrenexposed to domestic violence affect everyone’s kids. American Economic Journal:Applied Economics, 2(1), 211–228.

CBS (2009). Jaarboek Onderwijs in Cijfers [Education in Numbers Yearbook]. TheHague: Statistics Netherlands.

CBS (2012). Voortgezet onderwijs; vertraging opgesplitst naarop- en afstroom [secondary education: delay split up to go-ing to higher or lower school tracks]. Retrieved fromhttp://statline.cbs.nl/StatWeb/Download/VO_vertraging_inclu_161213150400.html.

Chattin-McNichols, J. (1992). The Montessori controversy. New York: Delmar.

Clark, D. (2010). Selective schools and academic achievement. The B.E. Journal ofEconomic Analysis & Policy, 10(1).

Cullen, J. B., Jacob, B. A., & Levitt, S. (2006). The effect of school choice on partici-pants: Evidence from randomized lotteries. Econometrica, 74(5), 1191–1230.

De Giorgi, G., Pellizzari, M., & Redaelli, S. (2010). Identification of social interactionsthrough partially overlapping peer groups. American Economic Journal: AppliedEconomics, 2(2), 241–275.

Deming, D. J. (2014). Using school choice lotteries to test measures of school effective-ness. NBER Working Paper 19803, National Bureau of Economic Research.

156

Dobbie, W. & Fryer, R. G. (2011). Exam high schools and academic achievement:Evidence from New York City. NBER Working Paper 17286, National Bureau ofEconomic Research.

Dohrmann, K. R., Nishida, T. K., Gartner, A., Lipsky, D. K., & Grimm, K. J. (2007).High school outcomes for students in a public Montessori program. Journal of Re-search in Childhood Education, 22(2), 205–217.

Duflo, E., Dupas, P., & Kremer, M. (2011). Peer effects, teacher incentives, and theimpact of tracking: Evidence from a randomized evaluation in Kenya. AmericanEconomic Review, 101(5), 1739–1774.

DUO (2010). Factsheet Passend Onderwijs, November 2010 [Factsheet suitable educa-tion, November 2010]. Den Haag: Ministerie van Onderwijs, Cultuur en Wetenschap.

Ergin, H. & Sönmez, T. (2006). Games of school choice under the Boston mechanism.Journal of Public Economics, 90(1), 215–237.

Eurydice (2009). Organisation of the education system in the Netherlands 2008/2009.Brussels: EU Education, Audiovisual and Culture Executive Agency.

Farrell, P., Dyson, A., Polat, F., Hutcheson, G., & Gallannaugh, F. (2007). Inclu-sion and achievement in mainstream schools. European Journal of Special NeedsEducation, 22(2), 131–145.

Figlio, D. N. (2007). Boys named Sue: Disruptive children and their peers. EducationFinance and Policy, 2(4), 376–394.

Fletcher, J. (2010). Spillover effects of inclusion of classmates with emotional problemson test scores in early elementary school. Journal of Policy Analysis and Manage-ment, 29(1), 69–83.

Friesen, J., Hickey, R., & Krauth, B. (2010). Disabled peers and academic achievement.Education Finance and Policy, 5(3), 317–348.

Hanushek, E. A. (2006). School resources. In E. Hanushek & F. Welch (Eds.), Handbookof the Economics of Education (Vol. 2) (pp. 865 – 908). Amsterdam: Elsevier.

Hanushek, E. A., Kain, J. F., & Rivkin, S. G. (2002). Inferring program effects forspecial populations: Does special education raise achievement for students with dis-abilities? The Review of Economics and Statistics, 84(4), 584–599.

Hastings, J. S., Kane, T. J., & Staiger, D. O. (2009). Heterogeneous preferences andthe efficacy of public school choice. Working paper. Formerly NBER Working PapersNo. 11805 and No. 12145.

157

Hastings, J. S. & Weinstein, J. M. (2008). Information, school choice, and academicachievement: Evidence from two experiments. The Quarterly Journal of Economics,123(4), 1373–1414.

Hoekstra, M. (2009). The effect of attending the flagship state university on earnings:A discontinuity-based approach. The Review of Economics and Statistics, 91(4),717–724.

Hoxby, C. M. (2000a). Does competition among public schools benefit students andtaxpayers? American Economic Review, 90(5), 1209–1238.

Hoxby, C. M. (2000b). The effects of class size on student achievement: New evidencefrom population variation. The Quarterly Journal of Economics, 115(4), 1239–1285.

Hoxby, C. M. (2000c). Peer effects in the classroom: Learning from gender and racevariation. NBER Working Paper 7876, National Bureau of Economic Research.

Hoxby, C. M. (2007). Does competition among public schools benefit students andtaxpayers? Reply. American Economic Review, 97(5), 2038–2055.

Hsieh, C. & Urquiola, M. (2006). The effects of generalized school choice on achieve-ment and stratification: Evidence from Chile’s voucher program. Journal of PublicEconomics, 90(8-9), 1477–1503.

Kalambouka, A., Farrell, P., Dyson, A., & Kaplan, I. (2007). The impact of placingpupils with special educational needs in mainstream schools on the achievement oftheir peers. Educational Research, 49(4), 365–382.

Karnes, M. B., Shwedel, A. M., & Williams, M. B. (1983). A comparison of fiveapproaches for educating young children from low-income homes. In Consortium forLongitudinal Studies (Ed.), As the twig is bent: Lasting effects of preschool programs(pp. 133–170). Hillsdale: Lawrence Erlbaum Associates.

Karnes, M. B., Teska, J. A., & Hodgins, A. S. (1970). The effects of four programsof classroom intervention on the intellectual and language development of 4-year-olddisadvantaged children. American Journal of Orthopsychiatry, 40(1), 58–76.

Kesten, O. (2010). School choice with consent. The Quarterly Journal of Economics,125(3), 1297–1348.

Kirabo Jackson, C. (2010). Do students benefit from attending better schools? Evi-dence from rule-based student assignments in Trinidad and Tobago. The EconomicJournal, 120(549), 1399–1429.

158

Koning, P. & Van der Wiel, K. (2013). Ranking the schools: How quality informa-tion affects school choice in the Netherlands. Journal of the European EconomicAssociation, 11(2), 466–493.

Krueger, A. B. (2003). Economic considerations and class size. The Economic Journal,113(485), F34–F63.

Ladd, H. F., Fiske, E. B., & Ruijs, N. M. (2011). Does parental choice foster segregatedschools? Insights from the Netherlands. In M. Berends, M. Cannata, & E. Goldring(Eds.), School Choice and School Improvement (pp. 233–253). Cambridge: HarvardEducation Press.

Lavy, V., Paserman, M. D., & Schlosser, A. (2012a). Inside the black box of ability peereffects : Evidence from variation in the proportion of low achievers in the classroom.The Economic Journal, 122(559), 208–237.

Lavy, V., Silva, O., & Weinhart, F. (2012b). The good, the bad and the average:Evidence on ability peer effects in schools. Journal of Labor Economics, 30(2), 367–414.

Leuven, E., Oosterbeek, H., & Klaauw, B. (2010). The effect of financial rewardson students’ achievement: Evidence from a randomized experiment. Journal of theEuropean Economic Association, 8(6), 1243–1265.

Lillard, A. & Else-Quest, N. (2006). The early years: Evaluating Montessori education.Science, 313(5795), 1893–1894.

Lillard, A. S. (2005). Montessori: The science behind the genius. New York: OxfordUniversity Press.

Lillard, A. S. (2013). Playful learning and Montessori education. American Journal ofPlay, 5(2), 157–186.

Manski, C. F. (1993). Identification of endogenous social effects: The reflection prob-lem. Review of Economic Studies, 60(3), 531–542.

Miller, L. B. & Bizzell, R. P. (1984). Long-term effects of four preschool programs:Ninth- and tenth-grade results. Child Development, 55(4), 1570–1587.

Miller, L. B., Dyer, J. L., Stevenson, H., & White, S. H. (1975). Four preschoolprograms: Their dimensions and effects. Monographs of the Society for Research inChild Development, 40(5-6), 1–170.

159

Ministerie van Onderwijs (2012a). OCW uitgaven voor onderwijs per deelnemer [Edu-cation expenditure per student from the Department of Education]. Retrieved fromhttp://www.trendsinbeeld.minocw.nl/grafieken/3_1_3_17.php.

Ministerie van Onderwijs (2012b). OCW uitgaven voor (voortgezet) speci-aal onderwijs per deelnemer [Education expenditure per special (secondary)education student from the Department of Education]. Retrieved fromhttp://www.trendsinbeeld.minocw.nl/grafieken/1_116.php.

Montessori, M. (1972). The discovery of the child. New York: Ballantine Books.

Montessori, M. (1973). From childhood to adolescence. Including "Erdkinder" and thefunctions of the university. Second Edition. New York: Shocken Books.

Montessori, M. (1989a). The child, society and the world. Unpublished speeches andwritings. The Clio Montessori series volume 7. Oxford: Clio Press.

Montessori, M. (1989b). What you should know about your child. Based on lecturesdelivered by Maria Montessori. The Clio Montessori series volume 4. Oxford: ClioPress.

Mooney, C. G. (2000). Theories of childhood: An introduction to Dewey, Montessori,Erikson, Piaget, and Vygotsky. St. Paul: Redleaf Press.

Murphy, R. &Weinhardt, F. (2013). The Importance of Rank Position. CEP DiscussionPaper 1241, Centre for Economic Performance.

Onderwijsinspectie (2007). Aansluiting voortgezet onderwijs op het basisonderwijs[Connection between primary and secondary education]. Utrecht: Onderwijsinspectie.

Onderwijsinspectie (2010). De staat van het onderwijs. Onderwijsverslag 2008/2009.[The state of education. Education Report 2008/2009]. Utrecht: Onderwijsinspectie.

Pop-Eleches, C. & Urquiola, M. (2013). Going to a better school: Effects and behavioralresponses. American Economic Review, 103(4), 1289–1324.

Rathunde, K. & Csikszentmihalyi, M. (2005a). Middle school students’ motivation andquality of experience: A comparison of Montessori and traditional school environ-ments. American Journal of Education, 111(3), 341–371.

Rathunde, K. & Csikszentmihalyi, M. (2005b). The social context of middle school:Teachers, friends, and activities in Montessori and traditional school environments.The elementary school journal, 106(1), 59–79.

160

Rothstein, J. (2007). Does competition among public schools benefit students andtaxpayers? Comment. American Economic Review, 97(5), 2026–2037.

Rubinstein, M. (2008). Karakteristieken van scholen voor voortgezet Montessori-onderwijs [Characteristics of schools for secondary Montessori education]. Apeldoorn:Garant.

Ruijs, N. M. & Oosterbeek, H. (2014). School choice in Amsterdam: Which schoolsare chosen when school choice is free? TIER Working Paper 14/21, TIER.

Ruijs, N. M., Van der Veen, I., & Peetsma, T. T. D. (2010). Inclusive education andstudents without special educational needs. Educational Research, 52(4), 351–390.

Saavedra, J. E. (2009). The learning and early labor market effects of college quality:A regression discontinuity analysis. Working paper.

Sacerdote, B. (2001). Peer effects with random assignment: Results for Dartmouthroommates. The Quarterly Journal of Economics, 116(2), 681–704.

Salend, S. J. & Garrick Duhaney, L. M. (1999). The impact of inclusion on studentswith and without disabilities and their educators. Remedial and Special Education,20(2), 114–126.

Seldin, T. & Epstein, P. (2003). The Montessori Way. Sarasota: The MontessoriFoundation Press.

Smits, J.A.E. & Vorst, H. (2008). Schoolvragenlijst: Herziene uitgave 2008, Voort-gezet Onderwijs [School questionnaire: revised edition 2008, secondary education].Amsterdam: Pearson.

Søgaard, A. J., Selmer, R., Bjertness, E., & Thelle, D. (2004). The Oslo Health Study:The impact of self-selection in a large, population-based survey. International journalfor equity in health, 3(1), 3.

Staatscourant (2011). Wijziging regeling Leerplusarrangement VO, Nieuwkomers VOen eerste opvang vreemdelingen. Regeling VO/FBI/259441 (7695). [Change regula-tion learnplusarrangement secondary education, newcomers secondary education andreception of immigrants]. Den Haag: Ministerie van Onderwijs, Cultuur en Weten-schap.

Staatscourant (2012). Regeling bekostiging personeel PO 2012-2013 en aanpassingbedragen leerlinggebonden budget VO 2012-2013. Regeling WJZ/373717 (10136).[Regulation funding personnel primary education 2012-2013 and adjustment fund-ing student bound budget secondary education 2012-2013]. Den Haag: Ministerie vanOnderwijs, Cultuur en Wetenschap.

161

Statistics Netherlands (2013). Bevolking en huishoudens; vier-cijferige postcode, 1 januari 2012 [Population and house-holds; 4-digit postcodes, 1st of january 2012]. Retrieved fromhttp://statline.cbs.nl/StatWeb/publication/default.aspx?DM=SLNL&PA=81922NED&D1=0&D2=a&HDR=T&STB=G1&VW=D.

Train, K. E. (2009). Discrete choice methods with simulation (second edition). NewYork: Cambridge University Press.

Van Loon, A. J. M., Tijhuis, M., Picavet, H. S. J., Surtees, P. G., & Ormel, J. (2003).Survey non-response in the Netherlands: Effects on prevalence estimates and associ-ations. Annals of Epidemiology, 13(2), 105 – 110.

Vorst, H. C. M. (1993). Leren en studeren: Vragenboekje voortgezet onderwijs enmiddelbaar beroepsonderwijs [Learning and studying: Questions for secondary andvocational education]. Nijmegen: Berkhout B.V.

Whitescarver, K. & Cossentino, J. (2008). Montessori and the mainstream: A centuryof reform on the margins. The Teachers College Record, 110(12), 2571–2600.

162

Samenvatting (Summary in Dutch)

Dit proefschrift bestaat uit vier studies naar vragen uit de onderwijseconomie. In dithoofdstuk vat ik de conclusies samen en beschrijf ik het belang van deze studies voorhet onderwijsveld. Opvallend genoeg vinden alle studies resultaten die in strijd zijnmet populaire opvattingen in het onderwijsdebat.

Hoofdstuk 2 gaat over de keuze voor middelbare scholen in Amsterdam. Nederlandheeft een lange traditie van vrije schoolkeuze, in tegenstelling tot veel andere landen.In landen zoals de VS is er een discussie gaande of ouders meer mogelijkheden moetenkrijgen om de school van hun kind te kiezen. Eén van de belangrijkste argumentenhiervoor is dat schoolkeuze competitie tussen scholen creëert. Het idee is dat oudersen leerlingen vaker kiezen voor goede scholen, zodat schoolkeuze scholen een prikkelgeeft om de kwaliteit van hun onderwijs te verbeteren. Omdat alle scholen deze prikkelkrijgen zou schoolkeuze de algemene kwaliteit van het onderwijs kunnen verbeteren.Dit mechanisme werkt alleen als leerlingen en ouders kwaliteit inderdaad belangrijkvinden bij de keuze voor een school.

De resultaten van hoofdstuk 2 laten echter zien dat schoolkwaliteit en veranderin-gen in schoolkwaliteit geen duidelijke invloed hebben op middelbare schoolkeuze. Dezebevindingen duiden erop dat het mechanisme van schoolkeuze als aanjager van kwa-liteitsverbetering misschien niet werkt, in ieder geval niet in de Amsterdamse contextmet de huidige maten voor schoolkwaliteit.

Klasgenootjes lijken daarentegen wel belangrijk te zijn bij de keuze voor een middel-bare school. Hoewel het niet kan worden geïnterpreteerd als een causaal effect, kiezenleerlingen vaker voor scholen die ook worden gekozen door hun klasgenootjes van debasisschool. Verder lijken leerlingen scholen te vermijden die in het voorgaande jaarhebben geloot door overaanmelding en kiezen leerlingen graag een school dichtbij hunhuis.

Eén van de gevolgen van vrije schoolkeuze is dat een deel van de Amsterdamsescholen te veel aanmeldingen krijgt. Scholen mogen niet zelf kiezen welke leerlingen zeplaatsen: ze moeten loten om de beschikbare plaatsen te verdelen. Het Amsterdamse

163

lotingssysteem krijgt ieder jaar veel kritiek van ouders. De leerlingen die zijn uitge-loot moeten een andere school kiezen, maar omdat alle scholen tegelijk loten hebbenpopulaire scholen meestal geen plaatsen meer.

In Hoofdstuk 3 onderzoek ik wat het effect is van het verliezen van een schoolloterijop de schoolprestaties van leerlingen. Ik volg hiervoor leerlingen die zijn in- en uitgelootvoor de school van hun eerste keuze. Leerlingen die zijn uitgeloot blijken niet slechter tepresteren dan leerlingen die zijn ingeloot. Na vier jaar middelbaar onderwijs presterende uitgelote leerlingen zelfs wat beter dan leerlingen die werden ingeloot. Omdat demeeste leerlingen nog op de middelbare school zitten is het nog onduidelijk of dezeresultaten ook opgaan voor het eindexamen. Voor nu kunnen de resultaten wordengezien als een geruststelling voor leerlingen die zijn uitgeloot.

Hoofdstuk 4 gaat over de effecten van voortgezet Montessorionderwijs. Het be-langrijkste verschil met reguliere scholen is dat leerlingen op Montessorischolen meerkeuzevrijheid hebben. Montessorionderwijs is erg populair, maar er is weinig bekendover de causale effecten van Montessorionderwijs op de ontwikkeling van leerlingen.De komt omdat leerlingen op Montessorischolen niet zomaar vergeleken kunnen wor-den met leerlingen op reguliere scholen. Montessorileerlingen hebben bijvoorbeeld vakerhoger opgeleide ouders.

Om de effecten van Montessorionderwijs te onderzoeken maak ik gebruik van school-loterijen. Lotende Montessorischolen zorgen voor een bijzondere situatie: ze creëreneen natuurlijk experiment. De resultaten laten zien dat Montessorionderwijs een andereroute naar vergelijkbare uitkomsten is. In tegenstelling tot claims van tegenstandersvind ik geen bewijs dat Montessorionderwijs negatieve effecten heeft op schoolpres-taties. Leerlingen van Montessorischolen halen hun eindexamen net zo vaak zondervertraging en met vergelijkbare cijfers als niet-Montessori leerlingen. Wel volgen zeeen andere weg naar het eindexamen. Aan de andere kant heeft Montessorionderwijsweinig effect op de sociaal-emotionele ontwikkeling van leerlingen. Het meest opvallendis dat Montessori leerlingen niet beter scoren op zelfstandigheid en motivatie.

Hoofdstuk 5 gaat over de aanwezigheid van zorgleerlingen in reguliere klassen, ookwel inclusief onderwijs genoemd. Inclusief onderwijs is een omstreden onderwerp inhet onderwijsveld. Een aantal belangrijke argumenten in de discussie richt zich op demogelijke effecten op niet-zorgleerlingen. Tegenstanders zijn bezorgd dat zorgleerlingenhun klasgenootjes afleiden en claimen dat zorgleerlingen meer aandacht van de leraarnodig hebben wat ten kostte gaat van de andere leerlingen in de klas. Voorstandersvan inclusief onderwijs stellen dat de extra ondersteuning in klassen met zorgleerlingenpositieve effecten kan hebben op de niet-zorgleerlingen in de klas.

164

In Hoofdstuk 5 onderzoek ik het effect van de aanwezigheid van zorgleerlingen opde schoolprestaties van niet-zorgleerlingen. Ik gebruik hiervoor op drie verschillendemanieren variatie in het aantal zorgleerlingen: ik gebruik variatie tussen vakken, binnenscholen en in buurten. Ik kijk naar zorgleerlingen met leerlinggebonden financiering(LGF), oftewel een ’rugzakje’. Dit zijn bijvoorbeeld leerlingen met aanzienlijke leer-of gedragsproblemen. Uit het onderzoek blijkt dat de aanwezigheid van zorgleerlingengeen positief of negatief effect heeft op de schoolprestaties van hun klasgenoten.

Deze resultaten komen niet overeen met eerder onderwijseconomisch onderzoek,waarin meestal wordt gevonden dat de aanwezigheid van zorgleerlingen een negatiefeffect heeft op de schoolprestaties van de andere leerlingen in de klas. Een mogelijkeverklaring voor dit verschil is dat zorgleerlingen onder het Nederlandse LGF-beleid meteen flinke aanvullende financiering naar reguliere scholen gingen. Het is niet ondenk-baar dat het effect van zorgleerlingen op niet-zorgleerlingen verandert wanneer scholenandere keuzes maken binnen het nieuwe beleid van Passend Onderwijs. Desondanksvoegt deze studie een belangrijke nuance toe aan het debat over inclusief onderwijs: ineen situatie met substantiële aanvullende financiering heeft de aanwezigheid van zorg-leerlingen geen positieve of negatieve effecten op de prestaties van reguliere leerlingen.

165

Top Institute for Evidence Based Education Research (TIER)

The Top Institute for Evidence Based Education Research (TIER) is an inter-universityinstitute that conducts research to develop evidence based education. The institute hasthree partners: the University of Amsterdam, Maastricht University and the Universityof Groningen and is located in Amsterdam, Maastricht and Groningen.

TIER contributes to the improvement of the quality of education in the Netherlandsby promoting an evidence based approach as a guiding principle in education policyand practice. It accomplishes this by developing (cost) effective education interventionsthat are grounded in sound scientific research. TIER research is funded by the Ministryof Education, Culture and Science and the participating universities through the NWOand complies with the quality standards and evaluation procedures used by NWO.

The following books recently appeared in the TIER Research Series:

I. C. Haelermans (2012), On the productivity and efficiency of education. The roleof innovations in Dutch secondary education

II. L. van Welie (2013), They Will Get There! Studies on Educational Performanceof Immigrant Youth in the Netherlands

III. S. Cabus (2013), An Economic Perspective on School Dropout Prevention usingMicroeconometric Techniques

IV. I. Cornelisz (2013), School Choice, Competition and Achievement: Dutch Com-pulsory Education

V. M. Heers (2014), The Effectiveness of Community School: Evidence from theNetherlands

VI. N. Ruijs (2014), Empirical Studies in the Economics of Education

166