Berlin school sorting

Bernd Beber, Moritz Matzner, Lennard Naumann, Macartan Humphreys

1 Overview

  • Questions and Answers
  • Data
  • Analysis
    • Descriptives
    • Causal Estimates
    • Explanation
  • Discussion

1.1 Schools sorting: Logics

Do parental school switching decisions worsen between group inequalities?

Background expectations: parental choice leads to increased segregation and increased inequality in school access

1. Schelling logic:

  • classic: even moderate micro homophily can lead to radical segregation

2. Moving to opportunity (Counter logic)

  • all parents wish to send students to better performing schools; identity used as a proxy

3. Combined Logic

  • segregation and inequality due to unequal ability to move

2 What we find

Post assignment shifts produce:

  1. reduced segregation and
  2. reduced inequality

Along the way:

  • Demographics unrelated to (official measures of) school quality
  • Flight from high migration and poor performing schools
  • Migrants are less likely than non-migrant peers in a given high migration school to request a change, but they are more likely overall
  • Migrant requests are less likely to be accepted, but still more requests accepted overall
  • Migrants put relatively more weight on school quality than nonmigrants

3 Setting and data

3.1 Parental choices

Berlin school system

  • students assigned based on small catchment zones
  • can request a change: select up to three alternatives, with explanation
  • requests approved or not
  • students stay in or exit system (e.g. private, or change Berzirk)

3.2 Parental choices: forms

3.3 Parental choices: data

We have data for every entering student with basic information from these forms for Tempelhof-Schöneberg for 2009 - 2018

  • Assigned school
  • Change request
  • Change preferences (1, 2, 3)
  • School ultimately attended
  • Street address (not house number or demographic information)

3.4 Parental choices: snapshot

Change requests are in fact remarkably common (and constant over time)

3.5 Schools in Tempelhof-Schöneberg

3.6 School quality

3.7 School quality snapshot

3.8 Official school data

  • As available to parents: includes school demographics, location, languages, absentees, …

3.9 Official school data snapshot

3.10 Official school data snapshot

Suggestive of increasing bimodality (though can also be explained by residential segregation)

3.11 Demographics data

From Berlin Kommunalstatistik department, imputed to street / plz level.

year strname plz likely_migrant
2013 Aachener Straße 10713 0.4835143
2013 Aalemannufer 13587 0.2924586
2013 Aarauer Straße 12205 0.1128205
2013 Aarberger Straße 12205 0.4786325
2013 Abbestraße 10587 0.6486486
2013 Abendrotweg 12307 0.0333333

3.12 Key data

No relationship between school demography and main school quality measure

3.13 Key data

Though related to some other measures

4 Major patterns

We are not seeing

  1. Increased access inequality
  2. Increased segregation

as a result of parental choices and bureaucratic responses to them

We seem to be seeing the opposite!

4.1 Outcomes 1: Increased segregation?

Estimated share migrants in class in assigned and actual schools, given own identity.

4.2 Outcomes 1: Increased segregation?

Outgroup exposure in class in assigned and actual schools, given own identity.

4.3 Outcomes 2

Inequality in access appears to go down (eliminated)

Relationship between demography and quality of assigned, requested, and actual school

Statistical models
  Assigned school School requested School attended
likely_migrant -0.41*** 0.06** -0.01
  (0.02) (0.02) (0.03)
R2 0.01 0.00 0.00
Adj. R2 0.01 0.00 0.00
Num. obs. 31610 29200 23258
RMSE 0.65 0.61 0.61
***p < 0.001; **p < 0.01; *p < 0.05

4.4 Outcomes 2

4.5 Outcomes 2

4.6 Joint shifts

Do actual movements produce more segregation?

5 Causal analysis? RDD results

  • RDD idea is to examine the differences in request behavior for parents who sit just on either side of a school zone border.

  • Is a parent more likely to seek a switch if they are very very close to a catchment zone with a more German school compared to a neighbor just inside that zone?

  • Causal effect: of having a school with given features (not: effect of these features)

  • Note: sorting works in our favor here

5.1 RDD results: All (demography)

5.2 RDD results: By subgroup

5.3 RDD results: All (quality)

5.4 RDD results: By subgroup

Migrants possibly more sensitive to moderate quality differences

5.5 RDD summary

  • Evidence that all types are, on average, likely to move to more German school; moderately stronger for natives
  • Evidence that all types are, on average, likely to move to a higher scoring school; moderately stronger for migrants
  • RDD gets teh effect of one assignment over another: but cannot infer that it is because of any specific characteristics

6 Core patterns

Lets make sense of all this

6.1 Identity or quality? 1

Requests given dyad demographies

6.2 Identity or quality? 2

Requests given dyad scores

6.3 Who moves away?

Migrants less sensitive to from demography

6.4 Who moves towards?

Migrants move more and less sensitive to to demography

6.5 Request out

Migrants leave nongerman schools at lower rates but in higher numbers

7 Explanation: Model based inference

7.1 Choices model

We implement a model to predict parental choice given all options in a given radius (2.5 km) as a flexible function of

  1. likely migrant background
  2. assigned school demography
  3. assigned school quality
  4. alternative school demography
  5. alternative school quality
  6. distance

The model allows for all threeway interactions involving 1 and two of 2 - 5, year and school fixed effects, with clustering by student.

7.2 Choices model: Predictions

In their choices, (likely) migrants weigh quality more than identity; non-migrants do the opposite.

7.3 Differential demands

8 Coda: Bureaucracy

8.1 Overall pattern

Migrant kids in nongerman schools are:

  • requesting transfers to German schools at lower rates
  • less successful in their applications
  • nevertheless transferring in higher numbers

8.2 Success differentials

Condition on the same from and to schools in a given year:

Statistical models
  Simple ‘From’ fixed effects ‘From/To’ fixed effects
(Intercept) 0.88***    
  (0.02)    
likely_migrant -0.42*** -0.39*** -0.43***
  (0.05) (0.06) (0.06)
R2 0.02 0.07 0.14
Adj. R2 0.02 0.06 0.12
Num. obs. 3273 3273 3273
RMSE 0.47 0.46 0.45
***p < 0.001; **p < 0.01; *p < 0.05

8.3 Success differentials

Condition on the same from and to schools in a given year:

9 Discussion

9.1 Limitations

  • Geographic scope small
  • Identity data is imputed
  • Migrant / Nonmigrant categories too coarse: “Homophily” among migrants stretches concept
  • Quality data cannot capture all dimensions of quality

9.2 Implications

  • Story is positive despite institutional inequalities: in initial assignments and responses to requests
  • For all that: still high levels of segregation
  • Possible that native discrimination is costly on its own terms
  • Mercator recommendations focus on accommodating segregation; better to focus on the decision making?

10 Extra slides

10.1 Key data

Though related to some other measures

10.2 Key data

Though related to some other measures

10.3 Joint shifts: Intended

Do actual movements produce more segregation?

10.4 Movements mapped

Most movements are to more German schools (turquoise)

10.5 Basic exit requests: Distance and Demography

Exit from migrant schools, dependent on distance

10.6 RDD models

Statistical models
  Model 1 Model 2 Model 3 Model 4 Model 5 Model 6
(Intercept) 0.44***   0.44***   0.44***  
  (0.02)   (0.02)   (0.02)  
treat 0.11* 0.06*** 0.12** 0.07*** 0.12** 0.05***
  (0.05) (0.01) (0.04) (0.01) (0.04) (0.01)
dist_signed 0.00· -0.00 0.00 -0.00 0.00 -0.00
  (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
treat:dist_signed -0.00 -0.00· -0.00 -0.00· -0.00 -0.00
  (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
abs_share_diff     0.00* 0.00** 0.00* 0.00*
      (0.00) (0.00) (0.00) (0.00)
treat:abs_share_diff     0.01** 0.00* 0.01** 0.00·
      (0.00) (0.00) (0.00) (0.00)
dist_signed:abs_share_diff     0.00 -0.00 0.00 -0.00
      (0.00) (0.00) (0.00) (0.00)
treat:dist_signed:abs_share_diff     -0.00 -0.00 -0.00 -0.00
      (0.00) (0.00) (0.00) (0.00)
likely_migrant         0.26** -0.21***
          (0.09) (0.05)
treat:likely_migrant         0.14 0.16·
          (0.18) (0.08)
dist_signed:likely_migrant         -0.00* -0.00*
          (0.00) (0.00)
abs_share_diff:likely_migrant         -0.00 0.01**
          (0.01) (0.00)
treat:dist_signed:likely_migrant         -0.00 0.00
          (0.00) (0.00)
treat:abs_share_diff:likely_migrant         -0.00 0.01
          (0.01) (0.01)
dist_signed:abs_share_diff:likely_migrant         -0.00* -0.00**
          (0.00) (0.00)
treat:dist_signed:abs_share_diff:likely_migrant         -0.00 -0.00*
          (0.00) (0.00)
R2 0.02 0.13 0.04 0.13 0.05 0.13
Adj. R2 0.02 0.13 0.04 0.13 0.05 0.13
Num. obs. 15163 15163 15163 15163 15093 15093
RMSE 0.49 0.46 0.48 0.46 0.48 0.46
N Clusters 220   220   220  
***p < 0.001; **p < 0.01; *p < 0.05; ·p < 0.1

11 Net changes in school size

More nongerman and low scoring schools have net losses (possibly with high targets in anticipation of losses)

11.1 Out movement, in movement: Demography

11.2 Out movement, in movement: Quality

11.3 Exiting migrants

11.4 School level segregation before and after

Reduced segregation comes at individual level, not school level

11.5 Expectations

Segregation at Primary Schools in Germany: The Effect of Parental Choice of School

The Berlin case study shows that many native German families take active steps to circumvent the official (binding) assignment of primary schools. As a result one in five primary schools enrols twice as many migrant students than are found in its catchment area.