Rich Harris Ron Johnston, Deborah Wilson, Simon Burgess

1
Rich HarrisRon Johnston, Deborah Wilson, Simon
Burgess

• YOU ARE WHERE YOU LIVE BUT YOUR SCHOOL AINT?
• EXPLORING COMPOSITIONAL, CONTEXTUAL AND SPATIAL
  EFFECTS OF ETHNICITY UPON SCHOOL CATCHMENTS
  WITHIN UK CITIES

2
Research Questions

• In a system where parents have constrained choice
  as to which schools their children attend
• What factors are associated with a pupil
  attending a near school?
• Is a consequence of pupils not attending a near
  school to increase observed ethnic segregation at
  the school level over and above that expected
  from the localities from which the pupils are
  drawn?
• Does increased segregation (if it occurs) affect
  school performance?

3
What is near?

• Nearest or near?
• If yes
• NR4 1
• Else
• NR4 0

4
What is segregation

• For any given pupil within a specific ethnic
  class (pre-given)
• (Proportion of pupils in the school they attend
  of that ethnicity)jk
• minus
• (Proportion of pupils in their residential
  locality of the same ethnicity)k
• Measure of concentration
• Locality is defined as the Lower Layer Standard
  Output Area in which the pupil resides
• A census zone

5
Initial restrictions

• A single unusual city
• A single ethnic group (white)
• Secondary schools only

6
Data sources

• Pupil Level Annual School Census returns (PLASC),
  2001 cohort
• OS CodePoint (Digimap)
• Postcode to OA LUTs (UK Borders)
• OA to Super OA LUTs (UK Borders)
• UK MosaicTM (courtesy of Experian)
• 18,495 from 20,436 records (91) for two
  settlements, fully postcode/Census geocoded,
  majority Mosaic coded

7
Geodemographic analysis !NR4 Vs eth diff(white
pupils, settlement A)
But Spearmans rank correlation 0.17 i.e. the
more white pupils dont attend a near school the
more white pupils are mixing with non-white
pupils at the school level vis-à-vis the locality
from which they come
8
Initial assessment of the geodemographic approach

• Advantages
• Clear and straightforward
• Quick to calculate
• Useful as an inductive approach for knowledge
  discovery
• Disadvantages
• Are the apparent differences between Mosaic Types
  statistically significant?
• Are the differences compositional/contextual
  i.e. would be explained away by explanatory
  variables or genuine neighbourhood effects
  (collective effects due to spatial interactions
  within neighbourhood types)?
• Would like to disentangle the effects of !NR4 on
  the proportion of white pupils in a school

9
Multilevel approachGeneral model structure

• Cross-classified, multilevel model
• Ignore geodemographics (Mosaic) for the time
  being and explore the compositional/contextual
  element

Mosaic Type
10
Multilevel probit model(random intercepts)
1
Proportion of pupils in school white
? (!NR4, )
0
Grand mean
Departure from mean for given census zone
11
Multilevel probit model(random intercepts
slope)
1
Proportion of pupils in school white
? (!NR4, )
0
Grand mean
Departure from mean for given census zone
12
(No Transcript)
13
(No Transcript)
14
(No Transcript)
15
Back to Geodemographics

• Cross-classified, multilevel model
• Re-introduce Mosaic as an additional level

Mosaic Type(level l)
16
Mosaic doesnt add anything
17
but it doesnt take much away, either
18
What is happening?

• The strong effects of school type (particularly
  faith schools) on the proportion of white pupils
  in the school.
• Mosaic is capturing well the spatial patterns
  of ethnic differentiation and concentration
  within settlement A.
• Space (specifically the geodemographic classes)
  therefore become a proxy for the missing
  variable loc_white
• This is useful because it means we can examine
  the regression residuals (the effects of !NR4) by
  Mosaic Type

19
(No Transcript)
20
South Asian Industry
21
(No Transcript)
22
Conclusions

• The effect of not attending a near school on
  increasing (or decreasing) the concentration of
  white pupils within a school relative to their
  residential localities from which they are drawn
  is strongly to the ethnic (white) composition of
  the locality.
• (As is the likelihood that pupils attend a near
  school)
• Mosaic offers no additional explanatory power
  over and above the predictor variables, though it
  does provide, in this case, a means to make sense
  of and group the data.
• (There is some evidence that not attending a near
  school is associated with a decreased GCSE
  performance in settlement A for white pupils no
  effect for Pakistani pupils but a positive
  effect for white pupils in a second settlement,
  B.
• Extremely complex!)