Centre for Market and Public Organisation

1
Centre for Market and Public Organisation
An application of geographical data inequalities
in school access Paul Gregg, and Neil Davies,
University of Bristol, CMPO
2
Overview

• What are the uses of geographic data?
• Geographic proximity Unique to ALSPAC
• How can it be applied?
• Description of the data
• Method
• Application SES gradients in school access
• Results
• Conclusion

3
Uses of geographic data

• Location has an effect on many processes, e.g.
• Access to services
• Exposure to pollutants
• Peer group effects
• Segregation
• It is useful to include neighbourhood in our
  models.
• Postcode fixed effects
• Spatial estimators

4
Constructing geographic data

• Postcodes
• are available
• ALSPAC records postcodes when sending out
  questionnaires
• Date of change is recorded
• Can be matched to longitude and latitude
• Problem - confidentiality
• Possible to identify individuals using postcodes

5
Constructing geographic data

• Solution
• Release postcodes attached to scrambled IDs
• Match IDs to a window of their peers within
  100m
• Remove postcodes
• Unscramble IDs to leave a dataset of linked IDs
• We have matched at 100m 200m and 500m
• For the years 1991-2005

6
An example, Clifton, Bristol
7
Clifton, Bristol Postcodes
8
Clifton, Bristol Postcodes
9
Clifton, Bristol One window
10
Clifton, Bristol Two windows
11
Number of peers within 100m for each child
12
Application School Access

• Work in progress!
• Questions/comments welcome

13
Motivation

• Schools matter
• Peer effects
• Teacher effects
• Previous studies have shown that access to good
  schools is not evenly distributed across
  neighbourhoods.
• Individuals sort across neighbourhoods to gain
  access.
• Individual students within a neighbourhood attend
  schools of differing quality,
• What individual level factors are these
  differences in school quality correlated with?
• What are the mechanisms are used to obtain high
  quality schooling?
• This paper seeks to describe these differences in
  school quality.
• Do these individuals have different preferences
  or is the assignment mechanism biased?
• Is there greater sorting across variables
  observable to schools?

14
Background School access (1)

• Allocation to schools by
• Location
• Academic Ability
• Prices
• Preferences
• Religion
• The English system is a hybrid of all them.
• Once we control for location how much of the
  variation in gradients of school quality remain?

15
Background School access (2)

• Location
• Large socio-economic gradients in access to
  school quality
• Individuals sort across neighbourhoods to gain
  access
• Largest determinant of school quality gradient is
  location,
• Poor children are 14 pp less likely to attend a
  good school than non-poor.
• Controlling for postcodes this difference falls
  to 2 pp.
• see Burgess and Briggs (2006)

16
Background School access (3)

• Individual students within a neighbourhood attend
  schools of differing quality,
• Why?
• What individual level factors are these
  differences correlated with?
• What are the outcomes of these allocation
  mechanisms?
• This paper uses the richness and geographic
  proximity of the ALSPAC observations to describe
  these differences.
• Conditional on location what determines the
  quality of school a child attends?

17
Defining school quality

• Our dependent variable is school quality,
  specifically
• Exam results of prior cohorts1,
• KS1, KS2, KS3, and KS4 (GCSE)
• of students who have
• Free school meals
• Statement of special educational needs
• Whether the school is oversubscribed
• 1 School quality Variables are lagged in time to
  obtain quality of school when child applied to
  school.

18
Method 1 Raw gradients

• Raw gradients
• This regression links the quality of school,
• an individual attends to there individual
  characteristics,
• One of the variables commonly used is whether the
  child takes free school meals,
• We wish to control for location

19
Method 2Spatial weighting

• Within neighbourhood estimate
• Differencing variables
• Where the mean of child is neighbours
  within 100m who attend a different
  school.
• is the difference in school quality
• We want to know how differences in the X
  variables are correlated with differences in
  school quality.

20
Spatial weighting (3)

• Bandwidth the window
• Postcodes, 100m, 200m, 500m
• Allows within neighbourhood estimates
• Sample selection
• Who is included?
• Same school?
• State/Private schools?
• Sample splits?

21
Results

• Results for secondary schools
• Average GCSE points
• Average KS2 of intake
• Whether the school was oversubscribed
• Further independent variables

22
Results (1) Avg GSCE
23
Results (1) Avg GSCE
24
Results (1) Avg GSCE
25
Results (2) Avg KS2
26
Results (2) Avg KS2
27
Results (2) Avg KS2
28
Results (3) - Oversubscription
29
Results (3) - Oversubscription
30
Results (3) - Oversubscription
31
Results (4)
32
Overview of complete results

• Secondary
• Variables observable to schools variables highly
  significant
• KS1, FSM, and location.
• Primary school quality
• Similar in magnitude to previous results
• Strongest sorting by religion, particularly
  through Catholic schools
• Primary
• Much smaller coefficients
• Evidence of sorting by FSM and KS1
• Evidence of school choosing?
• Some evidence of sorting by religion, again due
  to Catholic schools.

33
Conclusions for school markets

• There are socio-economic gradients in access to
  school quality
• These remains when controlling for location.
• Even within neighbourhoods school quality is
  correlated with measures of income.
• Most strongly with FSM, also KS1 evidence of
  schools choosing?
• Strongest correlations with religion
• School quality is highly persistent,
• primary school quality significant determinant of
  secondary school quality
• Some evidence that ethnic minorities attend
  better schools
• Would lotteries be fairer?

34
Uses of geographic data

• Location has an effect on many processes, e.g.
• Access to services
• Exposure to pollutants
• Peer group effects
• Segregation
• It is useful to control for neighbourhood.

35
Questions, Comments?