Geographical Ways of

looking at segregation

Rich Harris, University of Bristol, UK

School of Geographical Sciences &
Centre for Market and Public Organisation
 Outline

• Focus on two opportunities

• Modelling micro data geographically
– Mapping school catchment areas to identify
polarization
• Building geographical models
– Example of Geographically Weighted Regression
• Common framework for analysis
– “R”
• Open source software for computing and statistics
• http://cran.r-project.org/
 Outline

• Focus on two opportunities

• Modelling micro data geographically
– Mapping school catchment areas to identify
polarization
• Building geographical models
– Example of Geographically Weighted Regression
• Common framework for analysis
School “choice” & Social segregation?
Ethnic polarization?
 Geographical perspective

• Economic theory and government policy suggest schools

operate within local markets to attract pupils and funding.
• However, there is a deficit of understanding about the
scales and configurations of those admission spaces.
• Whilst competition for pupils and for school places is
assumed to operate at some localised scale, the actual
geographies of the markets, where they overlap and where
they might be changing are generally unknown.
• Aim: To understand processes of polarization in
the context of the local markets within which
schools operate.
• Task: To use micro-data to model those markets
 The data

• PLASC
– Pupil Level Annual Census Returns
• Data on all pupils in primary (and secondary) schools in
England
• 2005/6 data
– Information on state educated primary school students (5-
11 years old)
– 'Self-identified' ethnic category collected from parents
when students enrol
– Also records postcode unit of pupils' homes
– Which they school they attend
– School type (selective? Faith school?)
– Measure of deprivation (take a free school meal)?
 Defining ‘core catchments’

• Imagine centring a polygon

at (mid-x, mid-y) based on
the residential postcodes
of pupils attending a
school
– Let the polygon grow
outwards
• The direction of growth is
determined as that which
returns highest n1 / n2
– where n1 is number of
pupils in area going to
the school
– n2 is all pupils in the
area (go to any school)
– Measuring prevalence
• Continues until a certain
proportion of all pupils who
attend the school are
enclosed…
• p = 0.30
• Continues until a certain
proportion of all pupils who
attend the school are
enclosed…
• p = 0.40
• Continues until a certain
proportion of all pupils who
attend the school are
enclosed…
• p = 0.50
• Catchment is then defined
as the convex hull for
pupils of school within the
search area.
London
 Evidence of polarisation

• Are particular social (ethnic) groups travelling

further to school ‘than they need to’?
• Are there (primary) schools with an intake not
representative of the local community?
• Are there (primary) schools with shared
admission spaces but where one has a very
different intake to the other?
• Study region: London
 Evidence of polarisation

• Are particular social (ethnic) groups travelling

further to school ‘than they need to’?
• Are there (primary) schools with an intake not
representative of the local community?
• Are there (primary) schools with shared
admission spaces but where one has a very
different intake to the other?
• Study region: London
 Defining ‘Near’

• Define as being near to a

pupil any primary school
that has a core catchment
that includes the pupil’s
residential postcode
• Here the pupil has three
near schools
Proportion attending any near school
(target catchment p=0.50) LONDON
 Evidence of polarisation

• Are particular social (ethnic) groups travelling

further to school ‘than they need to’?
• Are there (primary) schools with an intake not
representative of the local community?
• Are there (primary) schools with shared
admission spaces but where one has a very
different intake to the other?
• Study region: London
 Pairwise Comparisons

• Look inside the catchments

– Expected intake Vs
Actual ethnic profile of each school
• Compare the profiles of locally ‘competing
schools’
– ones that overlap (strongly) in terms of their core
catchment areas
 Visual Summary (LONDON)

• Consider those
schools with highest
expected % Black
Caribbean
 Visual Summary (LONDON)

• Consider those
schools with highest
expected %
Bangladeshi
 Outline

• Focus on two opportunities

• Modelling micro data geographically
– Mapping school catchment areas to identify
polarization
• Building geographical models
– Example of Geographically Weighted Regression
• Common framework for analysis
 Example
Data Numerator/Denominator Source
Y Higher education Successful entrants under 21 in UCAS data, for 2002–2005/ 2007 Index of Multiple
participation Census population 14–17 Deprivation
X1 No qualifications Adults aged 25–54 in the area with no qualifications or with 2007 Index of Multiple
qualifications below NVQ Level 2, for 2001 /All adults aged 25– Deprivation
54.
X2 No post 16 Those aged 17 still receiving Child Benefit in 2006/ Those aged 2007 Index of Multiple
qualifications 15 receiving Child Benefit in 2004. Deprivation

X3 Average KS4 Total score of pupils taking KS4 in 2004 and 2005 in 2007 Index of Multiple
Points maintained schools from the NPD / All pupils in their final year Deprivation
of compulsory schooling in maintained schools for 2004 and
2005 from PLASC.
X4 Four or more Four or more cars in household / total households 2001 Census
cars
X5 Asian Total Indian, Pakistani, Bangladeshi people / total people 2001 Census
 Example
Data Numerator/Denominator Source
Y Higher education Successful entrants under 21 in UCAS data, for 2002–2005/ 2007 Index of Multiple
participation Census population 14–17 Deprivation
X1 No qualifications Adults aged 25–54 in the area with no qualifications or with 2007 Index of Multiple
qualifications below NVQ Level 2, for 2001 /All adults aged 25– Deprivation
54.
X2 No post 16 Those aged 17 still receiving Child Benefit in 2006/ Those aged 2007 Index of Multiple
qualifications 15 receiving Child Benefit in 2004. Deprivation

X3 Average KS4 Total score of pupils taking KS4 in 2004 and 2005 in 2007 Index of Multiple
Points maintained schools from the NPD / All pupils in their final year Deprivation
of compulsory schooling in maintained schools for 2004 and
2005 from PLASC.
X4 Four or more Four or more cars in household / total households 2001 Census
cars
X5 Asian Total Indian, Pakistani, Bangladeshi people / total people 2001 Census
 Global regression model (n = 31 378 )

β Standard error t value Significan

t at α0.01?
(Intercept) 3.620 0.0213 170.2 Yes
X1: No Qualifications -0.027 0.0002 -152.5 Yes
X2: No Post 16 Qualifications -0.002 0.0001 -15.1 Yes
X3: Average KS4 attainment 0.003 0.0002 52.6 Yes
X4: Four or more cars 0.018 0.0005 35.9 Yes
X5: Asian 0.012 0.0002 68.1 Yes
But… Geographical variation in the
“Asian” coefficient
 Geographically Weighted Regression

• What is it?
– Extension of
regression model Regression Point
– Allows model to vary Data Points
over space
• How it works...
 Summary of GWR model

β β (u,v) β (u,v) β (u,v) β (u,v) β (u,v) β (u,v) β (u,v) β (u,v)

(global
Min 1st 3rd Median 7th 9th Max. IQR
value)
decile decile decile decile
(Intercept) 3.620
X1: No -0.027 -0.047 -0.036 -0.032 -0.030 -0.027 -0.023 -0.014 0.006
Qualifications
X2: No Post 16 -0.002 -0.008 -0.003 -0.002 -0.001 -0.001 0.000 0.005 0.002
Qualifications
X3: Average KS4 0.003 0.000 0.001 0.002 0.003 0.003 0.004 0.006 0.001
attainment
X4: Four or more 0.018 -0.013 0.011 0.016 0.021 0.027 0.040 0.101 0.014
cars
X5: Asian 0.012 -0.156 -0.006 0.009 0.012 0.015 0.020 0.217 0.008
Geographical variation in the
“Asian” coefficient
 Outline

• Focus on two opportunities

• Modelling micro data geographically
– Mapping school catchment areas to identify
polarization
• Building geographical models
– Example of Geographically Weighted Regression
• Common framework for analysis
 Framework for analysis

• “R”
– Open source software
for statistical
computing
– Available at CRAN
• http://cran.r-project.org/
– WUN GIS Academy
• eSeminars about
Spatial analysis in ‘R’
• http://www.wun.ac.uk/g
gisa/
Thank you!