1

School Choice in Primary Schools:

Evidence from Wuppertal†

Andrea Riedel

Schumpeter School of Business and Economics, University of Wuppertal

Kerstin Schneider‡

Schumpeter School of Business and Economics, University of Wuppertal and CESifo

Claudia Schuchart

ZLB, University of Wuppertal

Horst Weishaupt

DIPF, Frankfurt

Preliminary version

February 2009

† Funding by the Deutsche Forschungsgemeinschaft (DFG) (SCHN 632/3-1) is gratefully acknowledged. The

authors like to thank the town council, the statistics department, the department for geo-data, the supervisory

school authority, and the principals of the primary schools in Wuppertal for their advice, encouragement,

patience, and the provision of the data for this research. ‡ Email: schneider@wiwi.uni-wuppertal.de

2

Abstract. In this paper we look at school choice in primary schools in Germany. The data

used is from Wuppertal, a major city in North-Rhine Westphalia (NRW), where school

districts were abolished in 2008 to allow for free school choice. Here we look at the situation

before 2008 to learn more about choice in the presence of school districts. It is not uncommon

to visit a primary school that is not the assigned public school. Moreover, parents choose

schools taking into account the distance to school, the quality of the school, and the

socioeconomic composition of the school. Families from disadvantaged neighborhoods tend

to send their children to the assigned school. A high percentage of migrants and/or

economically disadvantaged families in the school district, however, leads parents to choose

another school. Advantaged families make segregating choices, whereas disadvantaged

families tend to make integrating choices. The negative external effect of choice on the

composition of the not chosen school is significant and the level of segregation in the primary

schools is high and exceeds the level of residential segregation.

Keywords: Education system, segregation, school choice; migration; socioeconomic status.

JEL classification: I20, H75, J15.

3

1. Introduction

Unlike in other countries, school choice is not a prominent topic in education research

in Germany. It is commonly thought that at the primary school level there is no choice. At the

level of secondary schooling sorting is primarily driven by the rigid tracking system, where

students are allocated to the various tracks according to past performance or academic ability.

But school choice and ethnic school segregation exist in Germany, as was shown by Kristen

(2005), and clearly deserve more attention.

Past research and also the political discussion of the tracking system have pointed out

that the German school system leads to social and ethnic disintegration with strong effects of

the socio-economic background on performance in school (Entorf & Minoiu, 2005).

Interestingly, the focus of research and also the political debate has been more on tracking and

selection into secondary schools (Dustmann, 2004) or tertiary education than on earlier and

maybe even more important levels of schooling. This might be a shortcoming, as recent

literature points out the importance of early education and the associated high returns (Cunha,

Heckman, Lochner & Masterov, 2006). Pre-primary and primary education is important for

educational success and in particular for the opportunities of disadvantaged groups. In this

paper the focus is on primary education and in particular on the determinants and effects of

primary school choice in Germany.

School choice has been analyzed in numerous international studies. Choice is thought

to have a positive impact on competition between schools and therefore might increase the

quality of schooling (Hoxby, 2003). And also, choice can give parents a chance to find the

school that fits their preferences for education best. However, school choice has potential

negative effects as well. Choice might increase ethnic and social segregation (Burgess &

Briggs, 2006). Moreover, allowing parents to apply at a school of their choice, does not

mean that the child is accepted at the chosen school. The capacity of the school and the

distance to school remain the most important restrictions to choice. Furthermore, school

choice tends to be not practiced in disadvantaged families.

In particular in the US the issue of school choice has drawn considerable attention.

The intention of increased school choice by means of charter school programs was to reduce

racial and social segregation and to improve the chances for education of the more

disadvantaged groups (Hanushek, Kain, & Rivkin, 2002, Fryer & Levitt, 2004). The results

of many studies suggest the opposite (Lankfort & Wyckoff, 2001; Bifulco, Ladd & Ross,

2008). School choice tends to increase rather than to decrease segregation. Walsh (2008) does

4

not argue against this finding but claims that even without choice the within-school

heterogeneity is so low, that cream-skimming of the remaining high ability kids would not

have a sizable effect on those left behind. Echenique and Fryer (2006) and Echenique, Fryer,

& Kaufman (2006) also point to the fact that not only between school segregation but also

within school segregation matters. At percentage of more than 25 percent black students in a

school results in complete segregation, when looking at social interactions of the different

ethnic groups.

In this paper we look at school choice in a German context. The German education

system is strongly influenced by the federal structure of Germany. Each federal state can

decide on its schooling system. This has led to 16 education systems in Germany, with a lot of

variation between the federal states. For instance, each state decides on whether there should

be central exit exams (Jürges & Schneider, 2009) or how many tracks there are in secondary

education. Even the number of years in elementary school differs between the federal states

and there is no trend of convergence to a common education system in Germany. While the

federal system imposes substantial cost, the diversity within Germany leads to numerous

quasi-experimental situations that can be exploited for research.

One interesting feature of German federalism in schooling is that two German states

allow for public denomination schools, i.e., schools that are fully publically funded – just as

other public schools – but are (mostly) Catholic or Protestant schools. This gives parents a

choice not to send their children to the assigned public school even though there are school

districts. The two states are Lower Saxony (but only in some smaller regions) and North-

Rhine Westphalia (NRW). NRW is with a population of 18 Mio the largest German state and

it is very densely populated. In 2005 the NRW government decided to allow for more choice

by abolishing the school districts. The school year 2008/2009 was the first in which every

community had to enforce the new rule. The present paper is the first in a larger project in

which we evaluate the effects of school choice before and after abolishing the school districts.

The data used in this paper is from Wuppertal, a city with a population of about

350.000 that is located in NRW, south of the Ruhr area and east of the Rhineland. We have

information on the individual level that is enriched by data on the city block level, data on the

school district level, and data from the school statistics. To get a better understanding of

school choice under the system with school districts, this paper analyses the situation before

2008. It turns out that even with school districts, it is not uncommon to visit a primary school

that is not the assigned public school in the school district. Moreover, parents choose schools

5

taking into account the distance to school, the quality of the school, and the socio-economic

composition of the school.

In Section 2 give some more information on the institutional details of school choice

in NWR and also on Wuppertal. The data is described in Section 3 and in Section 4 we

explain our analytical strategy. Section 5 presents the results and we conclude and give an

outlook for future research in Section 6.

2. School choice in North-Rhine Westphalia: The situation before and after 2008

At first sight, there is rather limited school choice in German primary schools.

Students are assigned to a public school (Gemeinschaftsgrundschule) in a school district.

However, choice is not as limited as it appears to be. First, parents can apply for permission to

send their child to a different school (§39 SchulG-NRW (school law NRW)). They have to

come up with a convincing argument like a child care person in another school district. The

quality or the social composition of the school are not accepted reasons. The decision to

accept the application is made by the principals of the chosen school and often discussed with

the principal of the assigned school. To our knowledge, there is no research that analyzes the

permissions to visit a not assigned public primary school in Germany.

Second, there are public denomination schools (Bekenntnisschule) to which parents

can send their children. Public schools and public denomination schools do not charge school

fees. They are fully publically funded. In the following, we simply label them public schools

and denomination schools.

In addition to the public and the denomination schools, there are few private primary

schools, which however will be disregarded in this study. Private schools charge a school fee

and are often Waldorf schools, Montessori schools or private denomination schools with a

strong focus on the religious education. This is not the case with the public denomination

schools in NRW. Children in NRW have the right to attend a denomination school in their

community or a neighboring community, if the child belongs to that denomination (§26

SchulG-NRW). They might also be admitted to the denomination school if they don‟t belong

to the school‟s denomination, but the parents wish their child to be educated according to that

denomination. This is clearly a soft condition, which is not verifiable. Moreover, children of a

different denomination might be admitted if there is no school of the child‟s denomination

that can be reached. One commonly unknown feature of the public denomination school

system in NRW is that parents can vote to turn a public school into a public denomination

6

school (§27 SchulG-NRW). Hence, denomination schools exist in some but not all

municipalities. For instance the cities that are evaluated in the context of our larger school

choice project are the two neighboring cities Wuppertal and Solingen. Wuppertal has 356.000

inhabitants and runs 48 public primary schools, 11 Catholic schools, and 2 Protestant schools.

Solingen on the other hand has a population of 161.416 inhabitants and runs 24 primary

schools all of which are public schools. Denomination schools were disestablished in the

1965s by the local government in Solingen and were not reintroduced again.

Since the school year 2008/09 school districts for primary schools have been abolished

in NRW. Theoretically this should give parents free choice. However, the schools are given

fairly close guidelines on how to decide on admission. The distance to the chosen school is

the most important restriction. It is explicitly formulated in the law that students have the right

to be admitted to the closest school of the chosen school type (public or denomination), if the

capacity of the school permits (§ 46 SchulG). Interestingly – and this is not a result of the

school reform – it is stated in the NRW constitution that the economic or social background

must not be a criterion for admission to a school (Art. 10 LV NRW). But still, with the

demographic change, the number of school children is sinking, leaving more room for choice

and clearly also for increased competition between schools. Moreover, the new law might

affect parents in the two cities differently. Families in Wuppertal were always exposed to

choice, whereas choice in Solingen was much more limited.

The present paper is the first one to analyze the data collected and focuses on the

situation in Wuppertal before 2008. We use data from the official statistics on the level of the

school districts, data on the city block level, data from the school statistics and individual

level data from the schools, to get a first idea on the determinants of school choice.

Wuppertal is a city with a heterogeneous population. The unemployment rate in 2007 stood at

12.6 percent and the welfare dependence rate was 16.5 percent, which is above the German as

well as the NRW level. In this paper, we work with a slightly different definition of the

unemployment rate. We define the unemployment rate as the number of

unemployed/population between 16-65 years, since we have no information on dependent

employment in Wuppertal. According to that definition, unemployment was at 9 percent.

Figure 1 illustrates that Wuppertal is not only a city with economic problems, but also

a city with a lot of socioeconomic diversity. Just looking at the unemployment rates in the

school districts in Figure 1a, it shows that unemployment is in particular high in the middle

axis of the city around the famous Wuppertaler Schwebebahn (suspension line). The

7

suspension line is not only the city‟s landmark but also the most important element of the

public transportation system.

Unemployment drops considerably if one moves away from the suspension line, which

is in the Wupper valley, to the outer parts, the mountains of Wuppertal. A rather similar

picture is obtained, when looking at welfare dependency rates or the distribution of

immigrants in Wuppertal in Figure 1b. Furthermore, the parts of Wuppertal close to the axis

are more densely populated, when compared to the outer city regions.

Wuppertal has 61 primary public schools, of which there are 11 Catholic and 2

Protestant schools, with a total of about 13,000 students in 2007. As Figure 2 shows, the

public schools are allocated over the city area, while the public denomination schools are

more concentrated around the suspension line axis.

3. The data description and summary statistics

The data used in this analysis is from different sources and different years. Most of the

data is from 2007. The exceptions are the unemployment rate and the welfare dependency rate

on the school district level that were only available for 2006. Table 1 summarizes the data and

supplies the descriptive statistics for the total sample in column (1), the Catholic and

Protestant denomination schools in (2) and (3), and the public schools in column (4). Note

that three schools did not provide the student data and one school was excluded because that

school is about to be closed. The total number of students in the sample is 12,057. 9575

students visit a public school, 2124 students attend a Catholic school and 358 students visit a

Protestant school. The first interesting result of the sample statistics is that despite the still

existing school districts 32 percent of the students do not visit the assigned public schools,

either because they attend a denomination school or they attend a different public school,

which is true for 15 percent of the sample. Note that for each student there is, besides an

assigned public school, also an assigned Catholic school and an assigned Protestant school.

Since we have the students‟ address and know which school they visit, we are able to

assign each student to a public school in the school district and hence to determine whether

the student visits the assigned public school. The calculated percentage of students who visit a

different public school seems high, since parents have to apply to be admitted to another

public school and, as mentioned earlier, they have to present persuasive arguments.

On the student level we have addresses of 94 percent of the primary students in

Wuppertal. This enables us to calculate the distance (shortest distance) from the student‟s

8

home to school. We report four different values: the distance to the chosen school, the

distance to the assigned public school, the distance to the (assigned) Catholic school, and the

distance to the (assigned) Protestant school. The distance to the assigned public school is on

average the shortest. Choice leads to longer distances between the child‟s home and the

visited school. The assigned Catholic school is on average 1.5 km from the child‟s home and

the distance increases to 3.2 km for the assigned Protestant schools. Children who attend a

Catholic school can either attend the assigned Catholic school or another Catholic school. It

turns out that children, who live close to a Catholic school, tend to attend a Catholic school.

The same applies to children who visit a Protestant school. Children, who visit a public

school, tend to live further away from denomination schools and if they chose a not assigned

public school, they encounter a longer way to school. Figure 3 shows the distribution of

distances in more detail.

In this study we can only work with limited socioeconomic data on the student level.

The information on the student level we have is the students address, the nationality and – for

most students – the religious denomination. However, we can enrich the data with data from

the official statistics on 1966 city blocks in Wuppertal. For instance we have information on

nationality and welfare dependency and unemployment rates, which can be aggregated to the

school district level. Since we have the information of the student address, we can assign

socioeconomic information on the city block level to each student. This gives yields a fairly

detailed description of the students‟ neighborhood, which is an important predictor for the

socioeconomic performance of children (Borjas, 1995). For instance, if a student lives in a

city block with 20 percent migrants, which in the official statistics is the proportion of the

population without a German citizenship, she will be non-German with a chance of 20

percent. In our sample, the average percentage of non-Germans is 15 percent. Interestingly the

value is higher, when looking at the students in Catholic schools. This might be explained by

the fact that migrants from EU countries like Italy and Spain tend to keep their original

citizenship, even though the families have been living in Germany for generations, and are

typically well integrated. Since they are often Catholics they tend to visit Catholic schools.

When it comes to issues of integration, the citizenship has become a less reliable

indicator of integration. As has been recently shown in an integration report, the Turks are the

least integrated group (Berlin Institut für Bevölkerung und Entwicklung, 2009), but since the

children in families of Turkish descent are mostly already born in Germany, they have the

German citizenship and are not counted as migrants in the official statistics. The percentage

of Turks among the children aged 6-10 in a city block is only 5 percent, which underestimates

9

the size of the Turkish group in Wuppertal. This is supported by looking at the school

statistics. From the school statistics 2005 we know that 19 percent of the students in primary

schools are Muslims, which is a strong indicator that the percentage of Turks in the official

statistics underestimates the size of the population of Turkish descent.

According to the school statistics the proportion of Non-Germans in Wuppertal

primary schools is 22 percent. The number increases to 24 percent if resettlers from Eastern

Europe are added to the group of migrants. At this point, it is not clear, why we observe

substantial differences between the school statistic and the official statistics of Wuppertal. In

addition to the variables already described, we have information on whether the school is an-

all-day school or a (for Germany traditional) half-day school.

After primary school, German students get a recommendation for a secondary school.

In NRW, the choice is between a basic track school, an intermediate track school, an

academic track school, and a comprehensive school, which however has an internal tracking

system as well. The recommendation allows the student to choose between a comprehensive

school and one of the three other tracks. The most prestigious is the academic track school.

Students can always transfer to a lower track school than was recommended by the primary

school teacher. The schools in Wuppertal vary widely with respect to the percentage of

students transferring to the academic track. The average transfer rate is 34 percent, but ranges

from the lowest value of 6 percent to a maximum of 66 percent.

To add as much individual information as possible to the sample, we asked the schools

to give us in addition to the students‟ address the students‟ denomination, as an indicator for

migration status and the students‟ citizenship. 3 schools (2 public, 1 Catholic) did not give us

any student data. Moreover, not every school provided the denomination of the children,

therefore, we have information from 42 schools (34 public, 6 Catholic, 2 Protestant) on the

students denomination from 70 percent of the students and we know the citizenship of 87

percent of the students from 52 schools (41 public, 9 Catholic, 2 Protestant).

4. The analytical strategy

Our analysis proceeds in three steps and follows to some extent the analysis in

Bifulco, Ladd & Ross, 2008. First we model the decision to opt out of the assigned school.

We model the likelihood that student i will not visit the assigned primary school as a function

of the distance between the students residence and the assigned public school, iZD , students

neighborhood (city block) characteristics, iZS , the school district characteristics,

ZC , and the

10

characteristics of the schools, ZS , such as the percentage of students that transfer to the

academic track school after grade 4, and the random error term e.

( , , , , )iz iZ iZ Z ZY f S D C S e (1)

Equation (1) is estimated as a probit regression. We expect the distance to the assigned

school to have a positive impact on the decision to opt out, as it is a good indicator for

travelling costs but also for the costs a student has to incur if her classmates do not live in the

student‟s neighborhood.

Since we use the neighborhood characteristics as information to describe the

socioeconomic background of the student, students from economically disadvantaged

neighborhoods should be less likely to choose a school than students from more advantaged

backgrounds. Thus a high percentage of immigrants in the neighborhood reduces the

probability that the student chooses another than the assigned public school. The school

district variables, are not as much an indicator of the students‟ neighborhood, but reflect the

socio-economic composition of the schools. A school district with a high percentage of

families that are affected by unemployment or receive welfare payments suggest an

unfavorable socio-economic composition of the schools and should therefore lead families to

send their children to schools with a more favorable composition.

Finally, we have information about the chosen school, zS . The better the situation at

the school, the more likely it is that the school is chosen. One indicator of quality is the

transfer rate to the academic track school. The higher the transfer rate, the better the school

and the chances for the child to be in a high quality peer group. Another variable that could be

an indicator of high quality is the existence of all-day-schooling. Since the typical primary

school does not have all-day-schooling, but school is out at noon or 1:30pm, having access

better child care could be an argument to choose the school for working parents.

Next we consider the decision to choose a denomination school. Even though the

Catholics are a smaller group of the population in Wuppertal – 23 percent of the population

are Catholics and 35 percent are Protestants – the Catholic schools clearly outnumber the

Protestant school.1 Since we have only two Protestant schools in the sample, they are not

considered separately, but jointly with the Catholic schools. Since the family‟s denomination,

and in particular being Catholic might be of importance when deciding to send the child to a

public denomination school and in particular a Catholic School, we include in (1) also

variables that indicate the percentage of German and non-German Catholics in the

1

See West & Wößmann (2008) for a study with historic data.

11

neighborhood, iZRK . The effect of Catholic neighborhoods – German and non-German –

should therefore be positive. Also we include the distance between the student‟s residence and

the denomination school, iDD . The equation to be estimated is thus slightly modified to

( , , , , , , )iz iZ iZ iZ iD Z ZR f S RK D D C S e . (2)

Cleary, distance to the assigned school is an important factor when deciding on

whether to opt out of the assigned school or not, as has been pointed out in the literature

(Bifulco, Ladd, & Ross, 2008; Burgess & Briggs, 2006; Cullen et al 2005; Hastings et al.,

2006). The distance between the assigned and the chosen school can serve as a good

indicator of the cost of choice. How much extra travelling time and loss of neighborhood

classmates is a family willing to incur?

We therefore model the distance between the chosen school, chosenD , and the assigned

public school, assignedD , as a function of the variables used in (1) and the characteristics of the

chosen school. Thus the third equation to be estimated is

( , , , )iC iZ ZD f S C S e , with ln( ) ln( )iC chosen assignedD D D . (3)

In a second step, we look at the average composition of the schools that are chosen by

the students, cS , and the characteristics of the assigned public school, ,a actualS , and compare

them. Moreover, we compute the hypothetical composition of each school, had every student

chosen the assigned school, ,a hypoS . In addition, we differentiate between groups of students

according to denomination, nationality, and economic status. Use the variable „student is

Muslim‟ as an example. We first look at the average percentage of Muslims in the schools

that are visited by those who attend the assigned school. Within that group, we also

differentiate according to the denomination of the students. Second, we look at the average

percentage of Muslims in schools that are visited by students who do not attend the assigned

school. Again we differentiate by the denomination of the student and consider both

denomination and public schools. For each of the subgroups, we can calculate the average

percentage of Muslims at the chosen school, the actual and the hypothetical composition at

the assigned school. We compare those three measures and get a first insight on the effect of

choice on the composition of schools. Choice can have an integrating or a segregating effect.

For Muslims an integrating (segregating) choice is a choice of a school that has fewer (more)

Muslims than at the assigned public school. In contrast for non-Muslim students, an

12

integrating (segregating) choice is defined as the choice of a school that has more (fewer)

Muslims than the student´s assigned public school.

A final measure considered in this paper, is the extent of the observed segregation,

according to nationality, denomination and unemployment in the neighborhood. In particular

we consider the level of residential segregation and the level of segregation in the schools that

we actually observe. The segregation index between group a and b is computed as

1

1

2

Ni i

i

a bS

A B, (4)

where and i ia b are the number of individuals in group a and b in school district i and A and

B are the total number of individuals in group a and b.

5. Results

Determinants of choice

Tables 2-4 summarize the results of the regression analysis. Note that all t-values

reported are adjusted for clustering within school districts, and the reported coefficients in

Tables 2 and 3 are marginal effects.

In Table 2 the dependent variable is a binary variable indicating whether the student

does not attend the assigned public school. In model (1) we only include the distance to the

assigned school, the percentages of immigrants in the city block and in the school district, and

the percentage of students, which transferred to the academic track school in 2004. In

particular we use the difference between the transfer rates of the chosen school and the

assigned public school. Since the transfer rates are the only “objective” measure of school

quality, we expect parents to choose a better school, i.e., a school with a higher academic

track transfer rate.

As expected and in line with the literature, the distance to the assigned school plays an

important role. If the assigned school is far away from the student‟s home, she will be less

likely to choose that school. The percentage of immigrants in the city block is not significant,

the percentage of immigrants in the school district, however, comes in positive and

significant. While the significant effect is in line with our expectations, the size of the

coefficient is rather large and probably implausible. The academic track variable is positive

and significant, supporting our hypothesis that parents choose a higher quality school if they

choose. In model (2) the immigrant variables are replaced by the percentages of Turks. The

coefficients on the distance and the academic track variable remain almost unchanged. The

13

percentage of Turks in the city block has a significant negative effect and the percentage on

the school district level is smaller in magnitude and positive. Since we interpret the city block

variables as characterizing the students by their neighborhood and the school district variables

as characterizing the school environment, the results are quite intuitive. If a student has a

higher probability of having a Turkish background, the parents will not choose another than

the assigned school. If however, the school district has a large Turkish population, the

probability of attending another school rises.

In model (3) we use the student information on the religious denomination. Note that

due to missing observations the number of observations drops to 8420. As expected, the

variable has a negative impact. Students from disadvantaged families tend to visit the

assigned school. Since being disadvantaged might be reflected by the migration status but also

by the economic status of the student, we included the welfare dependency rates in (4). The

welfare dependency rate on the city block level lowers the probability to choose a school, but

the effect is not significant. The welfare dependency rate on the school district level, however,

shows the expected positive and significant effect. Combining migration background and

economic background in model (5) leaves the results fairly unchanged, even though the

significance levels drop and the percentage of Turks on the school district level is no longer

significant. This effect is possibly explained by the correlation between migration and

economic status. Finally we include the variable all day school, because having longer child

care at school might be considered as an indicator of school quality and in a country like

Germany with comparably poor day care facilities could be an argument for parents to choose

that school. Model (7) shows that this is not the case. All-day school enters with a negative

but insignificant coefficient. At first glance this is not intuitive. However, knowing that cities

in Germany introduced all-day schooling primarily in disadvantaged school district, the result

does not surprise. The socio-economic composition of the school is more important for school

choice than the availability of child care.

In Table 3 we analyze the decision to choose a public denomination school. In models

(1) to (4) the dependent variable is a binary indicator for whether the student attends a

Catholic school or not. In model (5) Catholic and Protestant schools are analyzed jointly.

As before, the distance to the school is of great importance. If the assigned public

school is further away from the student‟s home, it raises the probability that a Catholic school

is chosen. Increasing the distance to the denomination school lowers the probability to choose

a Catholic school. In column (1) the percentage of Catholic population in the city block is

included. Being from a German and also a non-German Catholic city block increases the

14

possibility to attend a Catholic school. The percentage of Turks turns out to be insignificant

and the academic track variable enters significantly. Thus parents appear to choose Catholic

schools because they are either from a Catholic background or because the school is better. In

column (2) we replace the percentage of Turks in the city block by the individual information

about the students‟ religion, i.e. whether the student is a Muslim. Not surprisingly, the

Turkish/Muslim variables enter with a negative and significant effect. Column (3) shows the

results, when the welfare dependency rates are included instead of the nationality /

denomination variables. The effect is insignificant. However, when including the

nationality/denomination variables together with the welfare variables, we find that lower

economic status, as measured by the welfare dependency rates, increases the probability to

visit a Catholic school and the effect on the non-German Catholics vanishes, which might be

due to some collinearity. When adding the Protestant schools and the Catholic schools the

socio-economic variables seize to be significant. The only variables that remain significant are

the distance variables except for the distance to the Protestant school and the academic track

variable that is almost doubled in size. A possible explanation for this result is that the choice

of a denomination school is not in the first place a choice against the assigned public school

but the school is chosen because it is more convenient to reach and the school appears to be of

better quality.

In Table 4 we report the results from a linear regression model to explain the

difference in the distance between the chosen school and the assigned school. The dependent

variable is the difference between ln(distance to chosen school) and ln(distance to assigned

school). The difference can be positive, when the chosen school is further away from the

child‟s home, it can be zero, if the chosen school is the assigned public school but also if the

chosen school, that is not the assigned school and the assigned school have the same address,

which is true for 2 schools (one public school shares the building with a Catholic school). If

the chosen school is closer than the assigned school, the difference might also be negative.

In column (1) all students are included in the regression. The percentage of immigrants

in the city block has a negative effect on the difference in distance. The interpretation is that

immigrants are less likely to incur additional travel costs by choosing a school. Immigrants at

the school district level have a positive but only marginally significant coefficient and the

academic track variable turns out to be insignificant. In model (2) we use the Muslim variable

instead of the city block immigrant variable and get a negative and significant effect. Note

that due to the missing observations are left with only 8409 observations. The school statistic

variable non-Germans and resettlers turns out to be insignificant.

15

In column (3) only public schools are included. The results now shows a negative and

significant effect of the city block migration variable and a positive and also significant effect

of the school district variable. Analyzing the public schools only, the academic track variable

is positive and significant. Parents are willing to pay the cost of travelling to school when the

school is a “better” school. Compared to the other models in Table 4 that show a rather poor

fit, the R2 increases. In column (4) we also include school size, average class size, and

whether a school is an all-day school. The school characteristics are all insignificant.

Choice and the composition of schools

Next we look at the effect of choice on the composition of the schools. Figure 4

summarizes the analysis. In Panel 1 of Figure 4 we calculate the average percentage of

Muslim students under different scenarios. We differentiate between percentage Muslims at

the chosen school, the percentage that would prevail had every student chosen the assigned

public school, and finally the actual percentage of Muslims in the assigned school. In addition

we differentiate between students who visit the assigned public school (left hand side) and

students who visit another school (right hand side). Within these groups we further distinguish

between Muslim students and non-Muslim students and whether the school is a public or a

denomination school.

The interpretation of the figures is straight forward. First, we get insight into the

effects of choice on the composition of the assigned school, because students who choose

another school change the composition of the assigned school. If a non-Muslim does not

attend the assigned school, the percentage of Muslims will ceteris paribus increase. We label

this externality the composition effect. The composition effect is negative, whenever as a

result of choice, the percentage of students from disadvantaged groups increases. It is positive

if the percentage of disadvantaged students decreases. And second, we can see whether choice

is integrating or segregating. Choice is integrating if students from disadvantaged groups

choose a school with a higher percentage of advantaged students.

Students, who do not attend the assigned public school, and who are not Muslims,

have a negative effect on the composition of the assigned school. The chosen school has on

average a lower percentage of Muslim students than the counterfactual, i.e. the percentage of

Muslims had all students visited the assigned public school. And because choice worsens the

composition compared with the composition that results from residential segregation, the

actual percentage of Muslim students at the assigned public school is even higher. This is true

16

regardless of whether the non-Muslim student chooses a denomination or another public

school. Muslim students, who choose a denomination school, attend a school with a

significantly lower percentage of Muslims, than they can expect at the assigned public school.

Thus choice is integrating, but also increases the percentage of Muslims that actually visit the

assigned schools even further and significantly.

The choice of Muslim students, who attend a not-assigned school, however, is

segregating and does not worsen the composition of the assigned school if they choose a

public school. Muslims who choose a not-assigned public school, attend schools with a

significantly higher percentage of Muslim students. Panel 1 also includes the category

missing, as for 30 percent of the students we have no information on the denomination.

The results are less pronounced for the percentage of German students. But German

students tend to choose schools with a higher percentage of Germans, thereby also increasing

tendencies of segregation. Particularly strong and significant effects occur in the group of

German students, who choose a denomination school. The adverse effects on the assigned

schools are significant.

Panels 3 and 4 repeat the analysis by using city block data. Consider Panel 3 first. A

student is labeled to be from a city block with low unemployment if the unemployment rate is

below the 10th

percentile of the distribution in Wuppertal. Students enter the calculations with

a probability of being affected by unemployment in the city block. If a student is from a city

block with a high unemployment rate, chances are high that the student is either directly

affected because the parents are unemployed or – since the risk of unemployment is highly

correlated with poor education – are from an economically disadvantaged background. This

assumes that city blocks have a fairly homogenous socio-economic background. As before,

we differentiate between denomination schools and public schools.

First look at the difference between the unemployment rates for students from high

and low unemployment neighborhoods, who visit the assigned public school. The difference

is 4.3 percentage points. The difference is with 3.8 percentage points a bit smaller, when

students choose another school. Interestingly, the unemployment rate is higher or almost as

high in the schools for students who do not attend the assigned school. One explanation might

be that students from low unemployment neighborhoods and school districts choose the

assigned school. Figure 1a shows that unemployment is low in the outer regions of Wuppertal

and from Figure we know that denomination school cannot be found in the outer regions of

Wuppertal. Students from school districts with high unemployment, that are close to the

17

suspension line axis will not choose schools with low levels of unemployment, as they are

further away, but choose schools that are fairly close and have only a slightly lower level of

unemployment than the assigned school.

The levels of unemployment at the not assigned schools are higher in denomination

schools than in the public schools, regardless of whether the student is from city blocks with

high or low unemployment. One explanation for this rather unexpected effect is that the

denomination schools are closer to the suspension train axis, with districts that are

characterized by unemployment and large proportions of migrants (cf. Figures 1a and b and

2). The wealthy parts of Wuppertal are on the mountains at the outer regions and not as easy

to reach by public transportation and have no denomination schools. And as we know from

the regression analysis, the distance to school has an important impact on choice. Moreover,

in particular Catholic schools might be chosen for religious reasons. The social composition

of the school might play a minor role. But again, choice worsens the composition of the

assigned schools.

Panel 4 shows a very similar pattern. We find the percentage of migrants in the

denomination schools to be substantially higher than in the public schools, which again might

be explained by the fact that Catholic migrants are often from EU member states and keep

their original nationality even though they are well integrated in the German society. As

before, the migrants„choices are integrating, the choice of the non-migrants tends to be

segregating.

Summarizing, we state that choice of the advantaged students has a segregating effect,

whereas the choice of the disadvantaged students tends to have integrating effects. This

confirms the results in Bifulco et al. (2008). Moreover, choice has a significant effect on the

social composition of the schools. The socio-economic composition of the assigned schools

becomes less balanced. It turns out that students, who choose, show a fairly similar pattern of

choice. The composition of those who choose is more balanced than the composition of the

students who do not choose, but attend the assigned public school.

Segregation

To conclude the analysis we calculate the net effect of school choice, to learn more

about the level of segregation. We compute the segregation indices in the chosen school and

the assigned public school and report them in Table 5. The first row shows the segregation

index, had every student decided to attend the assigned public school. The value of 0.296 in

column (1) indicates that 30 percent of the students would have to move to a different school

18

to result in an equal percentage of Germans at each public school, given that every student has

to attend a public school. Thus even in the absence of school choice, the level of segregation

would be substantial because of the socio-economic mix of the school districts is

heterogeneous, i.e., there is residential segregation.

In the first row of Table 5 we look at the actual segregation index, which differs from

the second row, because choice can alter the degree of segregation. We find that the

segregation index for Germans jumps by 8 percentage points. The segregation index for

denomination schools is even higher. The values calculated by using the school statistics

show a similar pattern, the segregation index, however, is smaller than the value obtained by

using individual data. The difference can probably be explained by the schools that did not

provide us with student data on nationality and the religious denomination. Moreover, when

using the school statistics we added the resettlers to the group of migrants. In the official

statistics we cannot analyze the group of resettlers, because they are German citizens.

However, if the migration variable should serve as a proxy for disintegration, it makes perfect

sense to include them. Not including resettlers in the group of migrants raises the difference in

the segregation index of denomination and public schools.

The degree of segregation is even higher when looking at the denomination variable in

(2) (using the school statistics). Moreover, the values differ again depending on the source of

data, since we do not have the information for 30 percent of the students. Another source of

variation might be that the data from the school statistics on the denomination variable is from

2005 and since then some primary schools have been shut down or are about to be shut down

in the next years (4 schools that still exist in 2005 have been shut down in 2007). As expected,

the level of denominational segregation is lowest at the denomination schools, but still

remains at a high level.

Finally we look at the extent of segregation with respect to unemployment. The degree

of segregation is clearly much lower than for the two other variables, although it might

however be misleading to compare the values, since the data is only available on the city

block level. Our results show that segregation increases when allowing for choice and within

the group of denomination schools the degree of segregation is lowest, which is mostly driven

by the two Protestant schools and not as much by the Catholic schools.

6. Conclusions and outlook

This study is one of the first to analyze primary school choice in Germany. Even

though school districts with an assigned public school exist and choosing a different school is

19

thought to be an exception, our data for Wuppertal – a major city in NRW close to the Ruhr-

Area – shows that this is not the case. Choice exists because in addition to the public schools

there are denomination schools the students can choose. Furthermore, there is the possibility

to apply for admittance at another than the assigned public school. 32 percent of primary

school students in Wuppertal do not visit the assigned public school in 2007/8. 20 percent

visit a denomination school and of the students who attend a public school, 15 percent go to a

public school that is not the assigned public school.

The paper gets some first insights on the determinants of choice. We find that the

distance to school and the perceived quality of the school influence the decision significantly.

Other important factors are the socioeconomic background of the students and the

composition of the school district. Families from disadvantaged neighborhoods tend to send

their children to the assigned school. A high percentage of migrants and/or economically

disadvantaged families in the school district lead parents to choose another school for their

children. Advantaged families make segregating choices, whereas disadvantaged families tend

more to make integrating choices. The external effect of choice on the composition of the not

chosen school is significant and the level of segregation in the Wuppertal primary schools is

high and exceeds the level of segregation of the population in the school districts.

In a next step, we will compare the situation in Wuppertal to the situation in the

neighboring city of Solingen, which is one of the few cities in NRW without denomination

schools and continue to analyze if and how choice has been altered by abolishing the school

districts in 2008.

References

Berlin Institut für Bevölkerung und Entwicklung (2009), Ungenutzte Potentiale. Zur Lage der

Integration in Deutschland, Berlin.

Bifulco, R. & Ladd, H. F. & S. L. Ross (2008). Public school choice and integration evidence

from Durham, North Carolina. Social Science Research,

doi:10.1016/j.ssresearch.2008.10.001.

Borjas, G. J. (1995). Ethnicity, neigborhoods, and human-capital externalities. The American

Economic Review, 85(3), 365-390.

Burgess, S. und Briggs, A. (2006). School assignment, school choice and social mobility.

CMPO Working Paper No. 06/157.

Chuna F, Heckman, J. J., Lochner, L. & Masterov D. V. (2006). Interpreting the evidence on

life cycle skill formation. In: Hanushek, E. A. and Welch, F. (eds.) Handbook of the

Economics of Education, Vol. 1, Chapter 12, 697-812.

20

Cullen, J.B., Jacob, B.A., & Levitt, S. D. (2005). The impact of school choice on student

outcomes: An analysis of the Chicago Public Schools. Journal of Public Economics, 89,

729-760.

Dustmann, C. (2004) Parental background, secondary school track choice, and wages. Oxford

Economic Papers, 56, 209-230.

Echenique, F. & Fryer, R. G., Jr. (2007). A measure of segregation based on social

interactions. Quarterly Journal of Economics, 122 (2), 441–485.

Echenique, F., Fryer, R. G., Jr. Kaufman (2006), Is school segregation good or bad?

American Economic Review, Papers and Proceedings, 96(2), 265-269.

Entorf, H. & Minoiu, N. (2005). What a difference immigration policy makes: A comparison

of PISA score in Europe and traditional countries of immigration. German Economic

Review 6(3), 355-376.

Fryer, R. G. & Levitt, S. D. (2004). Understanding the black-white test score gap in the first

two years of school. Review of Economics and Statistics, 86 (2), 447-464.

Hanushek, E. A., Kain, J. F., & Rivkin, S. G. (2002). New evidence about Brown v. Board of

Education: The complex effects of school racial composition on achievement. National

Bureau of Economic Research, Inc., NBER Working Papers: No. 8741.

Hastings, J., Kane, T. J. & Staiger, D. O. (2006). Preferences and heterogeneous treatment

effects in a public school choice lottery, NBER Working Paper 12145.

Hoxby, C. (2003). School choice and productivity: Should school choice be a tide that lifts all

boats? In C. Hoxby (Ed.), The Economics of School Choice. Chicago, IL: University of

Chicago Press.

Jürges, H. & K. Schneider (2009), Central exit exams increase performance…but take the fun

out of mathematics, forthcoming in Journal of Population Economics.

Kristen, C. (2005). School Choice and Ethnic School Segregation: Primary School Selection

in Germany. Münster: Waxmann.

Lankford, H. & Wyckoff, J. (2001). Who would be left behind by enhanced private school

choice, Journal of Urban Economics, 50, 288-312.

Walsh, P., (2008), Effects of school choice on the margin: The cream is already skimmed.

Economics of Education Review, doi:10.1016/j.econedurev.2007.11.005

West, M. R. & Wößmann, L. (2008). Every Catholic Child in a Catholic School: Historical

Resistance to State Schooling, Contemporary Private Competition, and Student

Achievement across Countries. CESifo Working Paper No. 2332.

21

Figure 1a. Unemployment in the Wuppertal school districts (2006)

22

Figure 1b. Allocation of immigrants in Wuppertal (2007)

23

Figure 2. Public and public denomination schools in Wuppertal

24

0

.00

05

.00

1.0

015

den

sity

0 1000 2000 3000 4000distance

Assigned School

0

.00

02

.00

04

.00

06

den

sity

0 2000 4000 6000 8000distance

Catholic School0

.00

01

.00

02

.00

03

den

sity

0 2000 4000 6000 8000 10000distance

Protestant School

0

.00

05

.00

1.0

015

den

sity

0 5000 10000 15000distance

Chosen School

Distance to school (in m)

Figure 3. Distance to school

25

0 .1 .2 .3 .4 0 .1 .2 .3 .4

Missing

Muslim

Other

Missing

Muslim

Other

Public

Denomination

Public

Denomination

Public

Denomination

Public

Denomination

Public

Denomination

Public

Denomination

Assigned Not Assigned

Chosen school Hypo. assigned Act. assigned

Student level data

Average % Muslim students

0 .2 .4 .6 .8 0 .2 .4 .6 .8

Missing

German

Other

Missing

German

Other

Public

Denomination

Public

Denomination

Public

Denomination

Public

Denomination

Public

Denomination

Public

Denomination

Assigned Not Assigned

Student level data

Average % German students

0 5 10 15 0 5 10 15

High

Low

High

Low

Public

Denomination

Public

Denomination

Public

Denomination

Public

Denomination

Assigned Not Assigned

City block data

Average Unemployment Rate

0 .05 .1 .15 .2 0 .05 .1 .15 .2

High

Low

High

Low

Public

Denomination

Public

Denomination

Public

Denomination

Public

Denomination

Assigned Not Assigned

City block data.

Average % Migrants

Figure 4. Effects of choice

26

Table 1. Sample description (1) (2) (3) (4)

all Catholic Protestant public school

Not assigned public school 0.323

(0.468)

0.147

(0.354)

Distance chosen school 742.5

(830.7)

682.8

(684.4)

1191.3

(1122.2)

738.9

(842.4)

Distance assigned public school 621.9

(469.8)

693.5

(424.0)

1023.1

(574.9)

591.0

(466.5)

Distance assigned Catholic school 1464.1

(1385.4)

637.7

(467.0)

1114.2

(944.7)

1660.5

(1463.0)

Distance assigned Protestant school 3201.0

(1581.5)

3023.6

(1338.5)

1164.2

(1012.5)

3316.5

(1593.3)

% Immigrants city block 0.153

(0.125)

0.189

(0.128)

0.125

(0.113)

0.147

(0.123)

% Turks city block 0.0453

(0.0627)

0.0527

(0.0555)

0.0272

(0.0441)

0.0444

(0.0646)

% Immigrants school district 0.139

(0.0710)

0.171

(0.0606)

0.136

(0.0508)

0.132

(0.0718)

% Welfare dep. rate city block 0.182

(0.140)

0.215

(0.132)

0.137

(0.122)

0.176

(0.141)

% Welfare dep. rate school district 0.166

(0.0770)

0.202

(0.0623)

0.156

(0.0530)

0.159

(0.0785)

All-day school in 2007 0.740

(0.439)

0.534

(0.499)

0.436

(0.497)

0.797

(0.402)

Observations 12057 2124 358 9575

27

Table 2. Decision to not attend assigned public school (1) (2) (3) (4) (5) (6)

Distance assigned public

school

0.000251***

(5.89)

0.000227***

(5.13)

0.000230***

(4.54)

0.000224***

(5.74)

0.000236***

(5.93)

0.000227***

(5.60)

% Immigrants city block -0.0329

(-0.27)

% Immigrants school district 1.548**

(3.16)

% Turks city block

-0.520*

(-2.55)

-0.546*

(-2.45)

-0.569*

(-2.51)

% Turks school district

0.0341**

(2.83)

0.0211

(1.94)

0.0197

(1.52)

0.0221

(1.68)

% Student is Muslim (d)

-0.0880**

(-3.07)

% Welfare dep. rate city

block

-0.0733

(-0.82)

-0.00164

(-0.02)

0.0131

(0.13)

% Welfare dep. rate school

district

1.318***

(3.67)

0.940*

(2.10)

1.018*

(2.31)

% Academic track: chosen -

assigned public school

0.0147**

(2.92)

0.0148**

(2.90)

0.0167**

(3.19)

0.0148**

(2.96)

0.0143**

(2.80)

0.0122*

(2.54)

All-day school in 2007 (d)

-0.123

(-1.22)

Observations 12057 12057 8420 12057 12057 12057

Pseudo R2 0.140 0.131 0.155 0.135 0.140 0.147

The dependent variable is the binary indicator for whether the student attends the assigned public school. Figures

reported are marginal effects from a probit estimation. In parentheses we report the t-values that are based on robust

standard errors adjusted for clustering within school districts.

(d) for discrete change of dummy variable from 0 to 1 * p < 0.05,

** p < 0.01,

*** p < 0.001

28

Table 3. Denomination School (1) (2) (3) (4) (5)

Catholic

school

Catholic

school

Catholic

school

Catholic

school

Catholic/protestant

school

Distance assigned public

school

0.0000908**

(3.12)

0.0000487

(1.74)

0.0000938**

(3.18)

0.0000425

(1.59)

0.000157***

(6.37)

Distance assigned

Catholic school

-0.000137***

(-10.51)

-0.0000994***

(-5.32)

-0.000132***

(-10.39)

-0.0000839***

(-4.99)

-0.000151***

(-11.73)

Distance assigned

Protestant school

-0.00000842

(-1.05)

% Non-German Catholics

city block

0.200

(1.52)

0.356*

(2.36)

0.144

(1.28)

% German Catholics city

block

0.194**

(2.90)

-0.0380

(-0.57)

0.0253

(0.36)

% Turks city block -0.0450

(-0.42)

-0.206

(-1.38)

% Turks school district 0.00333

(1.20)

-0.00839*

(-2.00)

-0.0182**

(-3.22)

0.00203

(0.43)

% Student is Muslim (d)

-0.0172**

(-2.65)

-0.0217**

(-3.15)

% Welfare dep. rate city

block

-0.00149

(-0.04)

0.112*

(2.34)

0.0152

(0.34)

% Welfare dep. rate

school district

0.194

(1.49)

0.453**

(2.92)

0.0753

(0.37)

% Academic track:

chosen - assigned public

school

0.00324*

(2.27)

0.00351**

(2.91)

0.00332*

(2.31)

0.00349**

(3.08)

0.00617**

(2.72)

Observations 12057 8420 12057 8420 12057

Pseudo R2 0.228 0.270 0.226 0.301 0.235

The dependent variable is in models (1) to (4) the binary indicator for whether the student attends a Catholic primary

school. In model (5) the dependent variable indicates whether the student attends a (Catholic or Protestant)

denominational school. Figures reported are marginal effects from a probit estimation. In parentheses we report the

t-values that are based on robust standard errors adjusted for clustering within school districts.

(d) for discrete change of dummy variable from 0 to 1 * p < 0.05,

** p < 0.01,

*** p < 0.001

29

Table 4. Difference in distance to chosen school versus assigned school (in logs) (1) (2) (3) (4)

% Immigrants city block -0.496*

(-2.53)

-0.236*

(-2.34)

-0.446*

(-2.27)

% Immigrants school district 0.857

(1.73)

1.069

(1.24)

0.702**

(3.39)

1.487*

(2.31)

% Academic track: chosen - assigned school 0.00382

(0.50)

-0.0000686

(-0.01)

0.0163***

(3.71)

0.00224

(0.32)

School size

-0.000177

(-0.49)

Average class size

-0.00673

(-0.57)

% Non-Germans and resettler in school

-0.373

(-1.23)

-0.456

(-1.88)

All-day school in 2007

0.0375

(0.57)

% Student is muslim

-0.0813*

(-2.41)

Observations 12057 8420 9574 12057

R2 0.013 0.009 0.068 0.019

The dependent variable is the difference between ln(distance to chosen school) and ln(distance

to assigned school). Figures reported are from OLS estimation. In parentheses we report the

t-values that are based on robust standard errors adjusted for clustering within school districts. * p < 0.05,

** p < 0.01,

*** p < 0.001

30

Table 5. Segregation Index

(1)

German

(2)

Muslim

(3)

Unemploy-

ment

N Schools

Assigned public school

(hypothetical values)

0.296 0.389 0.183 45

Chosen school 0.37 0.438 0.185 57

Chosen public school 0.365 0.410 0.203 45

Chosen denomination school 0.385 0.320 0.133 12

Chosen school

(school statistics)

0.344 0.409 57

Chosen public school

(school statistics)

0.343 0.423 45

Chosen denomination school

(school statistics)

0.349 0.361 12

Population in school district 0.241 0.000 0.183 45

The segregation index in column 1 uses the values from the school statistics (rows 5-7) that includes resettlers

in the group of non-Germans. In columns (1) and (2) rows 2-4 information is on the student level, whereas in

column (3) the city block data is used.