Centre for Market and Public Organisation An Economic Analysis of Parental Choice of Primary School in England Burgess, Greaves, Vignoles, Wilson June 2009
Mar 28, 2015
Centre for Market and Public Organisation
An Economic Analysis of Parental Choice of Primary
School in England
Burgess, Greaves, Vignoles, WilsonJune 2009
Introduction: School Choice in England
• Education Reform Act of 1988 – school choice mechanism by which parents can
choose the school their child attends.
• Funding follows the pupil. – Competitive pressure for schools to exert greater
effort to improve their academic achievement levels.
• Limited market– No indefinite expansion of good schools– Failing schools supported with additional resources– Not necessarily the case that academic standards are
key determinant of school choice by parents
Introduction: School Choice in England
• Parents’ preferences for schools matter for outcomes of “school choice”• In theory, schools compete according to parents’
preferences• This may lead to social stratification under some
conditions
• What constraints do parents face in school choice? • Small catchment areas for the best schools?• Transport?• Information?
Introduction: School Choice in England
• We look at parents’ stated and revealed preferences for schools
• Are stated and revealed preferences consistent?• What constraints matter in parents’ decisions?
Literature
• Markets in education and the role of school choice
• Rothstein, 2005, Hoxby, 2005
• Impact of competition minimal in England• Lavy, 2006, Gibbons et al., 2006, Burgess and Slater, 2006;
Allen and Vignoles, 2009• For contrary early evidence see Bradley, Johnes and
Millington, 2001
• Competition potentially leads to greater sorting but no evidence it increased in UK post 1988
• Söderström and Uusitalo, 2004, Burgess et al, 2006; Allen and Vignoles, 2007
Literature
• Stated parental preferences vary by socio-economic background and ethnicity
• Ball 2003; Gerwitz et al 1995; Hastings et al., 2005; Weekes-Bernard 2007; Reay, 2004; Butler and Robson 2003; West and Pennell 1999 and Coldon and Boulton 1991
• BUT Stated preferences may differ from their true preferences
Data
• Combine survey and administrative data• Millennium Cohort Study (MCS)• Pupil Level Annual Schools Census (PLASC)• EduBase
• This is an excellent combination. We have:• Detailed family level survey responses and
background controls• Detailed administrative information on all primary
schools in England• We essentially have the local market/choice set
Data
• MCS provides information on:– Up to 3 nominated schools on preference form (LA)– Other “truly preferred” schools not on form– Non-nominated schools that are feasible (more on
this later)– Stated reasons for preferences (all; most important)– Rich set of controls for families– Rich set of data on all schools– Actual school attended
Data
• MCS: Sample longitudinal survey • Random sample of electoral wards • Born 1st September 2000 – 31st August 2001 • Over-sampled from deprived areas and areas
with over 30% black or Asian families
• Wave 3 – children are aged 5, primary school age
• We look at England only• Final sample is 9,468 children
Stated preferencesTable 1: Most important reasons (grouped) for application for 1st choice school
Reason N %Proximity/ease of travel 2,567 27.5Sibling Rule 2,350 25.1Other family/friends 467 5.0Academic standards 1,521 16.3General good impression 1,439 15.4School characteristics and facilities 485 5.2
Strategic 62 0.7Religion 315 3.4School composition 44 0.5Pre-school & childcare 97 1.0Total 9,347 100Note: Table gives unweighted observations
Variation by family type
Table 2: Percentage giving each (grouped) most important reason, by the main respondent’s highest level of education
None 40.02 28.62 7.68 8.81 3.66 1.55‘Other’ or voc. qual. 31.81 26.01 13.22 13.86 4.92 3.79GCSE, grades D-G 29.51 24.46 12.14 13.77 6.67 2.75GCSE, grades A*-C 23.82 25.99 16.67 16.38 5.58 2.79AS/A Level 20.03 24.26 20.76 17.44 5.28 4.22Degree + 20.3 20.94 20.15 22.65 4.61 4.3Note: Table gives weighted %, using non response weight 2.
Good impression
School characteristics
ReligionEducation Proximity Sibling Academic
Stated Preferences: Problems
• Actual behaviour (or revealed preference) is not observed
• Revealed and stated preferences may diverge:– Only “socially desirable” responses may be given
(Jacob and Lefgren, 2007)– Stated preferences do not require parents to make
realistic trade-offs
– Parents may conflate preferences: • Proximity (did they move to a desirable catchment area
first?)• Older siblings (what was the initial choice based on?)
Revealed Preferences
• Use information from MCS wave 3• What school was put as the ‘first preference’ on
the LA application form?• Look at characteristics of this school, in relation
to other schools in the ‘feasible choice set’• What ‘type’ of school is chosen?
→ need to define feasible choice set
Feasible choice set
• All schools for which:• The pupil lives within 3km of the school• The pupil lives in the same LA as the school• Ignores geography within this boundary
Feasible choice set
• All schools for which – The pupil lives within the schools’ catchment
area, defined by the straight line distance in which 80% of pupils live
• The pupil lives within 20km of the school• The pupil lives in the same LA as the school
• Useful to compare results from each
‘Type’ of school
• 8 ‘types’ of school• Defined relative to the median in the feasible
choice set• Above/below median %FSM• Above/below median average KS2 score• Faith/non-faith
• So we have:• ‘Low FSM, high scoring, non-faith’ schools• ‘High FSM, low scoring non-faith’ schools….
• Not all pupils have each type of school in their feasible choice set but most have common types
Table 4: The % of pupils that have each school 'type' in their feasible choice setGroup % of pupils with group in choice set
1. Rich, low scoring non-faith school 59.59
2. Rich, high scoring non-faith school 89.53
3. Poor, low scoring non-faith school 93.04
4. Poor, high scoring non-faith school 75.85
5. Rich, low scoring faith school 31.23
6. Rich, high scoring faith school 85.76
7. Poor, low scoring faith school 51.99
8. Poor, high scoring faith school 41.47
Stated vs. Revealed• But different proportion of schools chosen…
Grouped 'type of school' indicator Proximity Academic Religion Strategic% % % %
Rich, low scoring non-faith school 8.14 4.97 0.52 4.44Rich, high scoring non-faith school 23.15 40.94 1.57 31.11Poor, low scoring non-faith school 37.88 13.16 3.14 20Poor, high scoring non-faith school 15.01 11.32 0 20Rich, low scoring faith school 1.78 1.75 7.85 2.22Rich, high scoring faith school 5.79 21.16 69.11 11.11Poor, low scoring faith school 5.62 2.76 8.38 8.89Poor, high scoring faith school 2.64 3.96 9.42 2.22Total 100 100 100 100Total sample size 1,783 1,092 192 46
Table 6: The prop. choosing each type of school, by the parent’s most important reason for school choice
Stated vs. Revealed
• Interesting similarities/differences• Parents that state academic standards are more likely
to choose the ‘rich, high scoring non-faith’ school• Parents that state proximity are more likely to choose
the ‘poor, low scoring non-faith’ school• Parents who claim to want high academic standards
are much more likely to choose rich high scoring schools than poor high scoring schools.
• Parents that state religious grounds are much more likely to choose the ‘rich, high scoring faith’ school but much less likely to choose the ‘poor, high scoring faith’ school than the ‘rich, high scoring faith’ school
• So more than religious considerations
Revealed preferences: Model
• What school ‘type’ is chosen?– Discrete choice modelling– Random utility framework
• How do school characteristics affect this choice?• How do parental characteristics affect this
choice?
Revealed preferences: Model
• We use a conditional/multinomial logit:
• Where schools indexed s=1,…,n , x varying characteristics of the schools, w represent the alternative invariant characteristics of the parent.
nse
eP
n
l
wx
wx
isliil
siis
,...,11
isX
Revealed preferences: specification
• What family characteristics affect the ‘type’ of school chosen?• Parents’ SES• Parents’ education• Parents’ religion• Index of Multiple Deprivation (IMD) of area• Child characteristics
Revealed preferences: specification
• What school characteristics affect the ‘type’ of school chosen?• % of pupils with FSM• % of pupils with SEN• % of pupils with EAL• % of pupils that are White British• Proportion of school that achieves all level 5 (highest
level) at KS2• Rank of distance from the home (closest, 2nd
closest…, furthest)
Revealed preferences: Role of School Characteristics
Table 8a: Results of 'mixed' logit regressionSchool char Odds-ratio t stat
% FSM 0.027***8.86
% Level 5 1.659 1.68
% EAL 0.922 0.18
% White 1.482 0.98
% some SEN 0.676 1.12
Distance rank 0.690***41.16
School characteristics are reported only. The schools are ranked closest, 2nd closest… in the feasible choice set
Revealed preferences: Role of Parental Characteristics
Table 8b: Results of 'mixed' logit regression, SES
Parental characteristics Group 1 Group 2 Group 4 Group 6SES: 2nd quinitle 1.571* 1.400* 1.584** 1.505*
SES: 3rd quinitle 1.406 1.546** 1.661** 2.215***
SES: 4th quinitle 1.472 1.843*** 1.604* 2.745***
SES: 5th quinitle 1.304 1.904*** 1.592* 2.521***
Note: Group 3 is the base category
Odds-ratio
1. Rich, low scoring non-faith school2. Rich, high scoring non-faith school3. Poor, low scoring non-faith school4. Poor, high scoring non-faith school5. Rich, low scoring faith school6. Rich, high scoring faith school
Revealed preferences: Role of Parental Characteristics
Table 8e: Results of 'mixed' logit regression, IMD
Parental characteristics Group 1 Group 2 Group 4 Group 6IMD: 2nd decile 1.061 1.019 0.903 0.99IMD: 5th decile 1.957* 1.715** 0.984 1.627*
IMD: 10th decile 2.220** 1.204 0.632 1.52Note: Group 3 is the base category. IMD is the Index of Multiple Deprivation A higher IMD indicates a more affluent area
Odds-ratio
1. Rich, low scoring non-faith school2. Rich, high scoring non-faith school3. Poor, low scoring non-faith school4. Poor, high scoring non-faith school5. Rich, low scoring faith school6. Rich, high scoring faith school
Importance of distance/feasible choice
Table 9b: The effect of distance rank on IMD coefficients
Parental characteristics Excluding Distance Including DistanceIMD: 2nd decile 1.486* 0.99IMD: 5th decile 3.388*** 1.627*
IMD: 10th decile 2.988*** 1.52Note: Group 3 is the base category
Odds-ratio: Group 2 only
1. Rich, low scoring non-faith school2. Rich, high scoring non-faith school3. Poor, low scoring non-faith school4. Poor, high scoring non-faith school5. Rich, low scoring faith school6. Rich, high scoring faith school
Ongoing work
• A more accurate definition of catchment areas
• Catchment area in which 80% of pupils live
• Define the feasible choice set as all schools for which the pupil lives inside the catchment area
Any ‘good’ schools left?
Table 11a: The % of each type of school exlcuded in '80%'Lowest IMD decile (most deprived)School type % of schools availableRich, low scoring non-faith school 15.19Rich, high scoring non-faith school 15.25Poor, low scoring non-faith school 21.88Poor, high scoring non-faith school 20.77Rich, low scoring faith school 44.47Rich, high scoring faith school 37.38Poor, low scoring faith school 32.09Poor, high scoring faith school 33.51
Any ‘good’ schools left?
Table 11b: The % of each type of school exlcuded in '80%'Highest IMD decile (least deprived)School type % of schools availableRich, low scoring non-faith school 56.86Rich, high scoring non-faith school 56.09Poor, low scoring non-faith school 33.36Poor, high scoring non-faith school 45.03Rich, low scoring faith school 57.14Rich, high scoring faith school 76.30Poor, low scoring faith school 60.92Poor, high scoring faith school 67.00
Conclusions
• Stated and revealed preferences vary• Parents’ socio-economic status and education
do play a role in their preferences– rich and poor do not have same preferences for
school factors
• High scoring advantaged schools are more likely to be ‘chosen’ by high SES individuals– Limit market forces in some areas– Increase social sorting
Conclusions
• Geography is crucial – are we really capturing genuine choice or
constrained choice
• We know that school de facto catchment areas have a big effect on the feasible choice set• Disproportionately for low SES families
more work needed