Presentation on theme: "Are England’s schools segregating or integrating? And does it matter? Simon Burgess and Jack Worth January 2011."— Presentation transcript:
Are England’s schools segregating or integrating? And does it matter? Simon Burgess and Jack Worth January 2011
January 2011, UoBwww.bris.ac.uk/cmpo2 Introduction Dynamics of sorting, changes in school composition. Depending on the process: –Maybe any composition is an equilibrium, or –Maybe the only equilibrium is complete segregation? Or integration? –Are there “tipping points”?
Some questions from a quantitative perspective using large-scale datasets Different actors making decisions: –Families choosing where to live and which schools to apply to –Schools and LAs making decisions – subject to the Admissions code – about who goes to over-subscribed schools January 2011, UoBwww.bris.ac.uk/cmpo3
Does it matter? Role of context in ethnic inequalities Context: –School peer groups and neighbourhoods. –Ethnic composition of schools and neighbourhoods and ethnic segregation. –Friendship formation. –How, if at all, does context affect outcomes? January 2011, UoB4www.bris.ac.uk/cmpo
Plan Analyse changing school ethnic group composition –Segregation or integration? –Differences? Does school composition matter? –Educational attainment –Values and attitudes –Identity January 2011, UoBwww.bris.ac.uk/cmpo5
January 2011, UoBwww.bris.ac.uk/cmpo6 Change in pop’n of whites 0 Fraction of non-whites
January 2011, UoBwww.bris.ac.uk/cmpo7 Change in pop’n of whites 0 Fraction of non-whites City
January 2011, UoBwww.bris.ac.uk/cmpo8 Change in pop’n of whites 0 Fraction of non-whites Growth in white population in areas with already high white population Decline in white population in areas with high white population Decline in white population in areas with already low white population Growth in white population in areas with low white population
January 2011, UoBwww.bris.ac.uk/cmpo9 Dynamic processes 0 Integration
January 2011, UoBwww.bris.ac.uk/cmpo10 Dynamic processes 0 Segregation
January 2011, UoBwww.bris.ac.uk/cmpo12 Research questions We aim to characterise the dynamics of sorting Is it the same process in all cities? Or not? To be specific –is it the same process, but different phases? –Or different processes? Is the process towards segregation or integration? At what speed? Are any differences in the dynamics correlated with any city structural factors?
January 2011, UoBwww.bris.ac.uk/cmpo13 Data Administrative data set: PLASC/NPD, Pupil Level Annual Schools Census part of the National Pupil Database. All pupils in all state schools in England. Data on pupil demographics including gender, age, ethnic group, FSM eligibility, SEN status, school and residence of pupils Primary (Age 5 – 11) and Secondary (12 – 16) schools. Composition from aggregation to school-cohort level. School information: Admissions policy, VA/VC, religious denomination, school post code etc. Area information including population, density, location,...
January 2011, UoBwww.bris.ac.uk/cmpo14 Sample Focus on areas above a threshold of ethnic minority populations (10%+ non-white) Study (initially at least) change in proportion white pupils as function of initial proportion of non-whites. Will definitely need to look at different groups: –Look at the change in their populations too –Look at change in proportion of whites relative to the base proportion of different ethnic sub-groups
January 2011, UoBwww.bris.ac.uk/cmpo15 Space and Time Use two very different geographies: analysis by LEA & Travel-to-work area (TTWA): –LEA – affects educational policy and education market boundaries –TTWA – local labour markets, where people live ‘Short’ datasets – 2003 through 2007 PLASCs ‘Long’ datasets – extends the 2003 PLASC back to entry year, using technique of a “frozen” entry cohort. –Longer time frame, but assumes changes in school composition random (tested before; and will test again)
June 2010www.bris.ac.uk/CMPO16 Figure 1. Definition of Aggregate Areas This work is based on data provided through EDINA UKBORDERS with the support of the ESRC and JISC and uses boundary material which is copyright of the Crown, the Post Office and the ED-LINE Consortium. Local Education Authorities Travel-to-Work Areas
January 2011, UoBwww.bris.ac.uk/cmpo17 Short and Long Year Age ‘98‘99‘00‘01‘02‘03‘04‘05‘06‘07 11 10 9 8 7 6 Available Data Used Data Infant to Junior
January 2011, UoBwww.bris.ac.uk/cmpo18 ‘Frozen Entry’ Approach Infant & Junior – some Primary schools split into Infant (Age 5-7) & Junior (Age 5-7) schools –Match the same pupils using adjacent censuses –Match where 90% of Junior cohort came from the same Infant and the two schools are within 1km of each other = 1,000 more schools Middle schools – in some LEAs schools of 11 year olds do not match schools of 6 year olds –Excluded in Primary Long dataset (mostly rural, white areas) –Secondary Long extended back 2 years to age 14
PrimarySecondary ShortLongShortLong Cohorts2003 & 20071998 & 20072003 & 20072001 & 2007 Schools12,42110,8092,5382,514 Pupils (‘03)503,449416,687460,415452,748 Non-whites84,50175,90770,05170,180 Final Sample Balanced Panel Focussing on schools that remained open through that period. Compare opening and closing schools.
January 2011, UoBwww.bris.ac.uk/cmpo20 Where, Dependent Variable Change in Whites:
January 2011, UoBwww.bris.ac.uk/cmpo21 Results Graphical data exploration –Different patterns in different places –Different dependent variables –Data problems in some places Techniques for analysing these patterns : –Distribution dynamics (“twin peaks” dynamics) –Characterise areas with a non-parametric estimate of the average relationship.
January 2011, UoBwww.bris.ac.uk/cmpo22 Techniques for analysing the patterns 1 Graphical data exploration
Figure 6c: Growth vs Initial Level plot – Manchester January 2011, UoB23www.bris.ac.uk/cmpo
Figure 6b: Growth vs Initial Level plot – Birmingham January 2011, UoB24www.bris.ac.uk/cmpo
Figure 6a: Growth vs Initial Level plot – London January 2011, UoB25www.bris.ac.uk/cmpo
Figure 6d: Growth vs Initial Level plot – Oldham January 2011, UoB26www.bris.ac.uk/cmpo
Figure 6e: Growth vs Initial Level plot – Bradford January 2011, UoB27www.bris.ac.uk/cmpo
Figure 6f: Growth vs Initial Level plot – Kirklees January 2011, UoB28www.bris.ac.uk/cmpo
Figure 6g: Growth vs Initial Level plot – Leicester January 2011, UoB29www.bris.ac.uk/cmpo
Figure 6h: Growth vs Initial Level plot – Blackburn with Darwen January 2011, UoB30www.bris.ac.uk/cmpo
January 2011, UoBwww.bris.ac.uk/cmpo31 Techniques for analysing the patterns 2 Summarising these graphs using different statistical procedures
January 2011, UoBwww.bris.ac.uk/cmpo32 Initial levels plot as before
January 2011, UoBwww.bris.ac.uk/cmpo33 Split into deciles with top and bottom 5% clipped
January 2011, UoBwww.bris.ac.uk/cmpo34 Coefficient = 0.0521 Take the weighted differences between decile means
January 2011, UoBwww.bris.ac.uk/cmpo35 Coefficient = 0.0521 Take the differences between decile means
January 2011, UoBwww.bris.ac.uk/cmpo36 Coefficient = -0.0124
January 2011, UoBwww.bris.ac.uk/cmpo37 Coefficient = 0.1013
Table 2: 4 th Order Polynomial 4th order polynomial LEA10%15%20%25%30%p-value(F) Oxfordshire0.38-1.43-2.61-1.095.240.361 Solihull2.210.11-2.61-1.537.770.011 Brighton and Hove0.80-0.52-1.05-0.750.430.000 Thurrock3.111.06-1.040.158.000.008 Stoke-on-Trent0.07-0.54-0.77-0.71-0.440.509 Calderdale-0.41-0.66-0.76-0.74-0.610.000 Slough-2.68-1.51-0.67-0.110.220.374 Peterborough-0.44-0.56-0.58-0.51-0.370.004 Rochdale-0.42-0.54-0.57-0.53-0.430.000 Walsall-0.18-0.41-0.52-0.54-0.480.000 Oldham-0.91-0.65-0.44-0.28-0.140.000 Lancashire-0.21-0.37-0.44-0.43-0.360.000 Kirklees0.34-0.12-0.43-0.63-0.710.000 Bradford-0.16-0.32-0.41-0.42-0.390.000 Blackburn-0.68-0.52-0.38-0.24-0.120.000 Liverpool-0.19-0.33-0.37-0.33-0.220.000 Reading-0.22-0.39-0.36-0.190.060.370 Dudley0.19-0.07-0.23-0.29-0.260.012 Birmingham0.550.15-0.12-0.27-0.330.000 Buckinghamshire0.350.09-0.09-0.19-0.230.374 Sandwell-0.67-0.32-0.080.090.180.001 Sheffield0.620.23-0.06-0.26-0.360.000 Trafford-0.19-0.10-0.04-0.02 0.001 Wolverhampton-0.53-0.23-0.020.120.200.032 Hertfordshire0.100.170.01-0.25-0.510.573 Leeds-0.08-0.010.020.040.030.000 London-0.16-0.020.100.180.240.000 Tameside0.340.210.120.060.020.001 Bolton-0.64-0.170.160.380.500.000 Nottingham0.040.120.180.220.240.210 Manchester0.980.630.370.190.080.000 Middlesbrough-0.230.280.490.480.310.000 Leicester0.760.660.540.410.280.000 Bristol City of-0.090.360.620.710.670.027 Southampton0.520.710.670.480.230.126 Coventry-0.040.450.700.760.670.001 Milton Keynes0.210.500.730.750.430.243 Luton2.521.610.920.440.120.000 Derby0.770.880.950.960.930.000 Bury-0.161.241.992.191.970.000 Wokingham-2.2210.8669.38203.12441.860.023 Sorted by slope coefficient at 20% non-white January 2011, UoB39www.bris.ac.uk/cmpo
January 2011, UoBwww.bris.ac.uk/cmpo40 Techniques for analysing the patterns 3 Estimating transition matrices to model the dynamics of the distribution
Estimating transitions Take a school with low % non-white pupils; what happens to that school over the next few years? –Does it see a further fall in the number of non- white pupils? –Or a rise? … school with high % non-white pupils: –Does it evolve to an all non-white school? –Does it see a more mixed pupil population? January 2011, UoBwww.bris.ac.uk/cmpo41
Create groups of schools within an LA based on their initial % non-white pupils. Estimate how schools move between those bands over the next ten years. Use this to compute/extrapolate a “long- run” or ergodic distribution if the same process continued indefinitely January 2011, UoBwww.bris.ac.uk/cmpo42
Summary Bringing together all that evidence: January 2011, UoBwww.bris.ac.uk/cmpo51
Table 4: Characterisation of Areas PolarisingUnclearIntegrating Blackburn with DarwenBirminghamBristol BradfordBoltonBury BuckinghamshireBrighton and HoveCoventry DudleyCalderdaleDerby LancashireHertfordshireLeeds LiverpoolKirkleesLeicester OldhamReadingLondon OxfordshireSandwellLuton PeterboroughSheffieldManchester RochdaleThurrockMiddlesbrough SloughTraffordMilton Keynes SolihullWolverhamptonNottingham Stoke-on-TrentSouthampton WalsallTameside Wokingham January 2011, UoB52www.bris.ac.uk/cmpo
Table 5: Segregation Dynamics and Structural Factors NonparametricPolynomialPolarisingIntegrating Proportion Autonomous Faith Schools-0.106 * -0.6852.024 ** -0.580 (0.0545)(0.756)(0.488)(0.635) Proportion Non-Autonomous Faith Schools-0.254-3.3612.487 * -0.875 (0.200)(3.186)(1.327)(1.111) Proportion Autonomous Secular Schools-0.09190.7644.926 ** 0.446 (0.134)(3.076)(1.731)(2.459) Proportion Non-White-0.0688-0.359-0.348-0.386 (0.0546)(0.756)(0.649)(0.789) Population Density0.08801.235-0.2441.867 ** (0.0599)(0.842)(0.676)(0.722) Proportion Independent Schools-0.0297-1.180-1.603 * 1.895 * (0.109)(1.854)(0.831)(1.118) Constant0.0528 * 0.106-0.04640.0434 (0.0269)(0.437)(0.210)(0.304) Observations414041 Standard errors in parentheses * p < 0.10, ** p < 0.05 January 2011, UoB53www.bris.ac.uk/cmpo
January 2011, UoBwww.bris.ac.uk/cmpo54 Summary Considerable differences between places in terms of the dynamics of sorting Some appear to be consistent with an integrated equilibrium; others with a segregated equilibrium.
How does context affect outcomes? School and neighbourhood peer groups might affect: –Educational outcomes –Values and attitudes –Identities January 2011, UoB55www.bris.ac.uk/cmpo
January 2011, UoBwww.bris.ac.uk/cmpo56 Does ethnic segregation in schools have a causal effect on differential school attainment? Big differences in attainment between different ethnic groups in England: –Black Caribbean pupils score about 0.4 SDs lower than White pupils in GCSEs –Indian students score 0.3 SDs higher than White pupils. Potential explanations: poverty, school quality and/or resources, teacher quality, teacher bias and expectations, ethnic composition of schools, …
January 2011, UoBwww.bris.ac.uk/cmpo57 Black Caribbean pupils Indian pupils Pakistani pupils Test score gap and ethnic segregation School NeighbourhoodSchool Neighbourhood
January 2011, UoBwww.bris.ac.uk/cmpo58 Conclusions We find that segregation has no consistent and significant impact on the minority-White British test score gap. Comparing the performance of a particular minority group across cities with varying levels of segregation, we find no tendency for significant negative effects of school segregation. This is in strong contrast to findings for the US –Card and Rothstein (2007) show that comparing a highly segregated city to a nearly integrated city closes the Black – White test score gap by about a quarter. Why?
Levels of school segregation are much lower in England than in the US. The nature of the academic performance of the relevant minority groups is very different. Much smaller variation in school quality in England. –Our approach subsumes in the segregation effect any differences in quality between the schools differentially attended by the ethnic minority group and by White students. January 2011, UoBwww.bris.ac.uk/cmpo59
These differences are likely to be much larger in the US than in England because the greater centralisation of education funding in England actively attempts to equalise educational spending per head: –the great majority of school funding is determined by a centrally-set funding formula. –the system provides significantly higher funding per pupil to schools with more deprived intakes –there is a smaller, specific funding stream, the Ethnic Minority Achievement Grant, which channels further additional funding to schools with high minority populations. January 2011, UoBwww.bris.ac.uk/cmpo60
How does context affect outcomes? School and neighbourhood peer groups might affect: –Educational outcomes –Values and attitudes –Identities January 2011, UoB61www.bris.ac.uk/cmpo
Context and Attitudes What is the impact of school and neighbourhood ethnic composition context on students’ attitudes to other ethnic groups? Be great to know … Putnam: diverse communities associated with “hunkering down” One example from school twinning programmes in Bradford, Kirklees, Oldham … January 2011, UoB62www.bris.ac.uk/cmpo
“Some of our children could live their lives without meeting someone from another culture until they go to high school or even the workplace” “They can grow up with such a lot of misconceptions and prejudices” (Primary school headteacher, Huddersfield; 92% pupils of Pakistani heritage; reported in TES 27.06.08) January 2011, UoBwww.bris.ac.uk/cmpo63
“Our pupils think its amazing that they like pizza too” (Primary school headteacher, Huddersfield; 92% pupils of Pakistani heritage; reported in TES 27.06.08) January 2011, UoBwww.bris.ac.uk/cmpo64
Context and Identity Co-evolution of identity and social network, social capital. –Who you think you are or want to be influences the friends you (try to) make Eg. “Acting white” dilemma (Fryer) –The friends you have influence your view of who you are Eg. Evolution of important social “lines” (Putnam) January 2011, UoBwww.bris.ac.uk/cmpo65
One way of finding out … Questionnaire on ethnic identity, eg. as being developed by Lucinda Platt and others for ‘Understanding Society’ Intervention study – school twinning programme Re-issue questionnaire to later generations of students (in twinned and non-twinned schools). January 2011, UoBwww.bris.ac.uk/cmpo66
Conclusions Its hard to know whether context matters, but we suspect it matters for some outcomes. But it doesn’t appear to matter for educational attainment: we find no strong, systematic relationship of ethnic segregation and educational attainment. Attitudes? Identity? January 2011, UoBwww.bris.ac.uk/cmpo67
January 2011, UoBwww.bris.ac.uk/cmpo69 Plans 1 Finish this “macro” characterisation of the (differing) processes of segregation Characterising the dynamics in a compact way, allowing for heterogeneity, and analyse differences: –Panel econometrics, autoregression with fixed effects, heterogeneity in the slope function (but AR>1) –Distribution dynamics, create composition classes and estimate transitions (first order markov assumption?; discretisation ad hoc) –Relate the characterisation of the area dynamic process to measures of structural city factors that the underlying behavioural model suggests.
January 2011, UoBwww.bris.ac.uk/cmpo70 Plans 2 Pupil analysis: –Following Burgess and Briggs (2006), we can use variation within postcode to decompose changes in composition: Ethnic change in postcode population Change in school destinations of people in each postcode. Different geographies: –LEA, TTWA –Data-based catchment areas (Rich Harris) Mathematical modelling: –Schelling –Quah