‘The Price You Pay’: The Impact of State Funded Secondary School Performance on
Residential Property Values in England1
John Glen and Joseph Nellis
School of Management, Cranfield University.
Abstract:
This paper examines the relationship between state funded secondary school performance
and local residential property values in seven major English cities. When choosing which
secondary school they wish their sibling(s) to attend, parents will be aware of the school’s
performance in Key Stage 3, GCSE and A- level examinations. We suggest that GCSE
examination results will be the measure of school performance that parental choice will be
most closely correlated with. Therefore, secondary schools with good GCSE examination
results will be ‘oversubscribed’ in that more students will wish to attend these schools than
there are places available. Schools will then have to develop mechanisms for rationing the
available places; central to rationing strategies in English schools at the moment is
geographical proximity to the school of choice. Parents will thus have a strong incentive to
purchase houses in the ‘catchment’ area of high performing schools. Our results suggest that
this is the case, with high performing schools stimulating a price premium in local residential
property markets of between 1-3% for each additional 10% point improvement in the pass
rate in GCSE examinations.
1. Introduction
Gibbons and Machin (2008) state that anecdotal evidence, media reports, and even dinner
party discussions lend credence to the claim that good schools raise local house prices. The
capitalisation of school performance into local property values has been the subject of a
number of studies in the US and latterly in the UK. Typically, these studies have been crosssectional and have found a positive relationship between a chosen measure of standardised
student test scores performance and local residential property values. Taylor (2005) provides
a review of the literature on the capitalisation of school performance into property values,
citing studies by Cheshire and Shepherd (2004); Weimer and Wolkoff (2001); Hayes and
Taylor (1997); Bogart and Cromwell (1997); Haurin and Brasington (1996); Leach and
Campos (2003).
The 1988 Education Reform Act introduced the national curriculum into English schools and
legislated for the provision of testing of schoolchildren at ages 7, 11 and 14 at the end of Key
Stages 1, 2 and 3 respectively. Students would then typically complete GCSE (General
Certificate of Secondary Education) examinations at the end of Key Stage 4, aged 16. The
1
The authors are grateful to HSoS plc for providing the database used in this study and for research funding
support.
1
Act placed a focus on measuring the outputs of state funded education in England at the end
of all four Key Stages. This reflected a general concern on the part of those legislating to
provide outputs from the state education system which would equip the UK labour force with
the skills, knowledge and competencies necessary to drive economic growth in an
increasingly competitive global economy (Elkins and Elliott, 2004).
The publication of school performance data in relation to public examinations at GCSE level
has led to predictable outcomes according to Elkins and Elliott (2004) namely that those
schools which perform best are oversubscribed while underperforming schools fail to achieve
their target intake of pupils. The argument in favour of publishing examination performance
data is that it will encourage underperforming schools to do better and high performing
schools to maintain or further improve their academic standards.
The introduction of testing of school pupils at ages 7, 11 and 14, as well as public
examinations at GCSE and A-level2 has provided a wealth of information on the performance
of schoolchildren at various stages of their school lives. In addition, the formal inspection of
schools by OFSTED (Office for Standards in Education) has provided additional detailed
information on the performance of children, teaching staff and management in state funded
schools. This has given parents a plethora of data upon which to assess the performance of
schools. As a result, teachers are now deemed to be more accountable to head teachers,
governors, parents and central government for their performance.
This paper is concerned with the impact of state funded secondary school performance on
house prices in seven3 major cities/urban conurbations in England. The significance of this
study stems from the fact that housing and house prices are critically important to the local
and national economies. Housing is the most important asset purchased by most UK
households and a large proportion of personal sector wealth is tied up in residential property.
Furthermore, any capitalisation of school performance into local property prices will have
implications concerning the ability of low income families to access high performing, state
funded schools.
2
A- Level examinations are typically taken by students after a further two years of post-compulsory education,
typically age 18.
3
Greater London, Birmingham, Bristol, Leeds, Greater Manchester, Liverpool and Newcastle.
2
The structure of the paper is as follows: the next section briefly summarises the details of the
schools’ performance regime which has existed in England since 1988 and the nature of
capitalisation of school performance into residential property prices. Section 3 describes the
dataset employed in this study and the main procedures adopted to render it suitable for
statistical analysis. Section 4 provides an abridged outline of the modelling framework
adopted in this paper (a more detailed description is available from the authors on request),
while Section 5 discusses the main empirical results. Finally, Section 6 draws together the
main conclusions and sets out the key findings.
2. The Capitalisation of Secondary School Performance in England into residential
property prices
The value which parents place on good school performance would be easy to assess if
education services were sold in a free and competitive market. However, since there is no
observable market price for state-provided schooling, an indirect method must be found to
place a monetary value on the services provided by state-funded schools.
The hypothesis we are proposing in this paper is that, all other things being equal, a higher
level of school performance will be capitalised into higher property values. In other words,
parents wishing to gain access to schools with a superior academic performance will bid up
the price of properties which are most geographically proximate to a high performing school.
They will do this in order to ensure that their children are in the catchment area of the high
performing/oversubscribed school and are therefore eligible for entry into that school.
This theoretical argument follows Starrett (1981), Wildasin (1987) and Hoyt (1999) who all
argue that where there is a relatively inelastic supply of housing, land and house prices will
rise in those areas which provide relatively more attractive amenities, to the point where the
additional price paid reflects the perceived additional value of high-quality local amenities.
The capitalisation of local amenities and neighbourhood effects into residential property
values has been widely researched in the US.
Brasington (2002) provides a thorough
overview of the capitalisation debate, the key point being that capitalisation of any
differential in local amenity is contingent upon a supply of housing which is inelastic with
respect to price. In extremis where this is not the case, any differential in local amenity
which produces an increase in housing demand would be met instantaneously by an increase
3
in housing supply and, therefore, there would be no measurable impact on residential house
prices.
Furthermore, there could be differing levels of capitalisation across a given urban space due
to different elasticises of supply of housing within it. For example, we may expect the
elasticity of supply of housing to be more responsive to demand on the edge of urban areas
where more land for development may be available and planning restrictions may be less
restrictive. However, recent inner-city developments in some of the major conurbations
within England may have produced a more responsive supply of housing in those areas. The
recent debate about the lack of supply of housing in the UK and the response emanating from
the Office of the Deputy Prime Minister would suggest that it is reasonable to assume that the
elasticity of supply of housing in the UK is such that differences in local amenities are likely
to be capitalised into local property prices. Although, in so far as there has been a more
elastic response from housing supply, we might expect the impact of school performance on
property prices to have been dampened.
Bartik (1988) provides an informative taxonomy of the impact of local amenities on house
prices, location choices and housing supply. Firstly, he argues that an increase in the supply
of high amenity locations would reduce the house price premium which people would be
willing to pay for the high amenity. This is interesting in the context of school performance
because the introduction and publication of school performance measures has resulted in an
increase in the average performance of schools.
On the basis of Bartik’s assertion we might expect this to reduce the premium paid for good
school performance, as the average level of school performance across all schools has
increased.
Table 1 below shows that there has been a step increase in the average performance of
schools in England as a whole over the period 1988 to date. As such, this means that there
are more schools offering enhanced performance in terms of student examination results at
GCSE level. In a competitive market this effectively means that there is a greater supply of
good schools, as measured by GCSE performance of students. A greater supply of good
schools should lead to a reduction in the “price” which parents have to pay in order for their
siblings to gain entry to a good school.
4
Table 1: Percentage of students achieving 5 or more A*-C GCSE pass grades in
English Schools.4
1988/89
1989/90
1990/91
1991/92
1992/93
1993/94
1994/95
1995/96
1996/97
1997/98
1998/99
1999/00
2000/01
2001/02
2002/03
2003/04
2004/05
2005/06
2006/07
Boys
29.8
30.8
33.3
34.1
36.8
39.1
39.0
39.9
40.5
41.3
42.8
44.0
44.8
46.4
47.9
48.8
51.4
53.8
56.4
Girls
35.8
38.4
40.3
42.7
45.8
47.8
48.1
49.4
50.0
51.5
53.4
54.6
55.4
57.0
58.2
58.8
61.4
63.5
65.5
Total
32.8
34.5
36.8
38.3
41.2
43.3
43.5
44.5
45.1
46.3
47.9
49.2
50.0
51.6
52.9
53.7
56.3
58.5
60.8
Source: Department for Education and Science.
However, it could well be the case that parents are not only assessing the absolute level of
school performance but also its relative level. If this is the case then it is likely that parents
will be willing to “pay” not only for high performing schools but for schools which are high
performing and out performing the other providers in any particular geographical location.
This introduces the concept of school performance league tables. These tables are a creation
of the media in the UK and it should be noted that the UK Department for Education and
Science does not produce official league tables.
Second, Bartik (1988) suggests that where amenity levels are low this will tend to attract low
income households. Locations that improve more than the average will tend to attract high
income households. A number of studies in the US have shown that school performance is
positively correlated with the composition of the student body. Crone (1998) argues that
empirical evidence is available to show that academic achievement can be improved by the
peer group effect.
4
The GCSE A* grade was introduced in 1994 to recognise exceptional performance at the top end of the scale.
5
Third, any increase in the quantity of local amenities will decrease demand for housing
characteristics that are amenity substitutes. Conversely, there will be an increase in demand
for housing characteristics that are amenity compliments. This last point may mean that
improved performance of a school which is located in a low income location may attract
higher income households into that area with the result that not only is the performance of the
local school improved but a process of gentrification may occur in the local housing stock.
Finally, Bartik (1988) focuses upon the inter-temporal nature of the capitalisation of
improved local amenity. He identifies a three stage process. In the first instance there is an
improvement in local amenity. This is followed by a second stage whereby the improvement
in local amenity is recognized by those agents who are active in the local property market.
The third stage involves these agents reflecting improvements in local amenity in their
willingness to pay a premium for properties that embody the improved local amenity.
This process is important in the context of this study. Schools are incentivised to improve the
academic performance of their students in public examinations; this improved performance is
then reported in official statistics released by the Department for Education and Science and,
finally, house buyers can be expected to react to the improved school performance in their
willingness to pay a premium for properties which are proximate to the high performing
schools.
The publicity in both local and national media that has surrounded the performance of state
schools has ensured that parental awareness of the performance of state secondary schools
has been greatly enhanced. It would, therefore, seem a reasonable expectation, a priori, that
parents have the requisite knowledge available to them in order to assess both the absolute
and relative performance of individual schools.
The focus of this paper is on the capitalisation of state funded secondary school performance
in England into residential property prices. The single measure of school performance
employed in the analysis is the percentage of students, in any one academic year, gaining five
or more GCSE passes at grades A*-C. In doing so we are guided by Lambert and Lines
(2000), who state that:
6
‘ …..GCSEs are now seen as a way, by aggregation, of certifying the whole school.’5
Downes and Zabel (2002) in their study of the impact of school characteristics on house
prices in Chicago found that homeowners paid attention to school outputs, i.e. test scores, and
not inputs in the form of per pupil expenditures by schooling authorities. Brasington and
Haurin (2006) examined the extent to which homeowners valued traditional measures of
school performance (absolute test scores), of which GCSE examination pass rates are an
example. These traditional measures contrast with the relatively new value-added measures
of school quality, which attempt to measure the capability of students as they enter the
secondary schooling process and compare this with the test scores on exit, the difference
between the two being the value added. With this approach it is the measure of value added
which acts as a measure of school quality rather than the absolute performance at the end of
the process.
Whilst educationalists may argue that the value-added approach represents a much more
effective evaluation of school performance, Brasington and Haurin (2006) report that
homeowners value average test scores and levels of expenditure on education above any
measure of student value-added when assessing local school performance.
They argue
therefore that it is average test scores which are capitalised into local property prices rather
than value-added measures of school performance.
Perhaps the most telling manifestation of the impact of school performance on property
prices is to be found on the UK internet property search site “Rightmove”. One of the
services provided by Rightmove to prospective purchasers of property is the capability to
click through from any property to information on the nearest state funded secondary schools.
The information includes straight-line distance to those schools, examination performance of
those schools, as measured by GCSE pass rates and the most recent OFSTED report.
5
The primacy of GCSE results as an indicator of state funded, secondary school performance is most probably
due to the disproportionate amount of media coverage which GCSE results are given, at both the local and
national level, compared to Key Stages, 1, 2 and 3 and A-level results.
7
3. Description of the Dataset
This section describes the data employed in this study, including the data cleansing
procedures adopted. The results reported in this paper are based on an exceptionally large
dataset developed by HBoS (the largest residential mortgage lender in the UK). The dataset
covers all the house purchase transactions on which the HBoS plc group has provided
mortgage loans between 1983 and 2007 (totalling more than 5 million transactions) although,
for the purpose of this paper, we have only selected data for the period from 2001 to 2007.
This period is chosen as a result of:
•
a desire to present estimations of capitalisation which are contemporary rather than
historical.
•
following Bartik (1988), a desire to allow for a period of time to elapse whereby
house purchasers were aware of differing school performances and could reflect this
in their willingness to pay for new properties.
In relation to the second point, we feel that the availability of data electronically on the
Department for Education and Science website has been an important conduit via which
‘reliable’ information has been made available to parents and house purchasers.
Our analysis examines the impact of school performance on residential property values in
seven selected major cities in England, namely Greater London, Birmingham, Bristol, Leeds,
Greater Manchester, Liverpool and Newcastle. The data on school performance measure the
performance of individual schools in terms of the percentage of students attaining five passes
at GCSE grades A* to C. This percentage score was entered as a continuous variable in our
preferred hedonic house price regression model. The data on school performance were
obtained from The Department for Education and Science website. The website also provided
us with the required locational data in terms of the postcode of each school.
In order to allow for the capitalisation of school performance into property prices we have
lagged the school performance data one period. Thus, we have regressed 2001 school
performance results on 2002 property price values. The rationale for this approach is that it is
historical school performance data that will impact current offer prices for residential
8
properties. In addition, we are suggesting that it is the most recent historical school
performance data that will have the most immediate impact on the price offered by house
purchasers.
The data on house price characteristics is provided from the HBoS mortgage data set. The
richness of this dataset allows us to analyse the impact of school performance on residential
property values to an extent which we feel has not been possible for researchers previously in
the UK. The database includes purchase price, recorded at the mortgage approval stage, as
well as a detailed breakdown of the information on the following house characteristics:
•
property type, i.e. whether the house is detached, semi-detached, terraced, bungalow,
flat or maisonette;
•
number of habitable rooms;
•
floor space area;
•
number of bathrooms;
•
number of toilets;
•
availability of central heating;
•
number of garages and garage spaces;
•
garden;
•
age of property;
Over and above these physical house characteristics, the database also includes information
on the location of each property in terms of postcodes. In order to locate each property and
each school in our dataset, we employed Royal Mail Post Office postcodes. We then utilised
the Royal Mail's POSTZON software to provide us with an Ordinance Survey (OS) grid map
reference. This grid reference effectively places each property on the bottom left-hand corner
of a 100 metre by 100 metre grid. To each of the Eastings and Northings of the OS grid
reference a five was added, which effectively located each of our properties at the centre of
the 100 metre by 100 metre grid; i.e., at the expected location of any property within a 100
metre grid.
We then developed a simple computer macro which applied Pythagoras's
theorem and calculated the straight line distance between each property in the dataset and the
nearest school.
9
Rosenthal (2003) reports that POSTZON locates each school and house within a maximum of
70 metres of its true location and concludes that for the purpose of locating the nearest school
to a given residential property, the use of the POSTZON database is unlikely to result in a
serious misallocation of houses to the nearest school. Gibbons and Machin (2008) suggest
that a lack of clearly defined catchment areas for state schools introduces an amount of
ambiguity in the link between residential property locations and accessibility to school.
Rosenthal (2003) also comments that a lack of national information on school catchment
areas makes it difficult to test the assumption that the geographically closest state secondary
school is the relevant school, in terms of catchment area, for any given property. Using data
from a particular English city (Stoke-on-Trent) Rosenthal tested the extent to which the
closest school to a given property, as calculated using the POSTZON dataset, coincided with
the catchment area within which a given property was placed. Reassuringly, in 83% of all
observations the nearest school coincided with the catchment area zoning of a particular
property.
The HBoS dataset includes a special locational classification regime used in the UK, known
as ACORN, A Classification of Residential Neighbourhoods 6.
This regime allocates ratings according to geo-demographic information, allowing for the
categorisation of the location of each property at two levels: at the micro level (immediate
residential neighbourhood) and at the macro level (wider surrounding area).
In our study the inclusion of ACORN codes, in so far as they allow for the influence of the
socioeconomic characteristics of a neighbourhood population, should pick up some of the socalled peer group affects; i.e., our study allows for the socio-economic impacts on school
performance noted by Crone (1998).
Based on postcode information, the location of each property is classified into an ACORN
group. The HBoS dataset employs eight main ACORN groups, as shown in Table 2 below.
6
For the full details of the ACORN classification system, based on postcodes, see CACI (2006).
10
Table 2: Description of Main ACORN Groups
ACORN Group
Main Characteristics of Group
A
Areas where residents are wealthy investors
B
Prospering families
C
Areas of traditional money
D
Young urbanites
EFG
Areas of middle aged families (comfortable), contented
pensioners and families and individuals looking to settle down
(Middle Aged Comfort (E), Contented Pensioners (F) and Settling
down (G))
H
Moderate living
IK
Meagre means and impoverished pensioners
J
Inner city existence (low income singles and couples, with little
financial activity; multi-ethnic young singles renting flats; high
rise poverty, relying on welfare; poor young, multi-ethnic who are
financially inactive)
Data Cleansing
Prior to modelling the relationship between school performance and local property prices, the
dataset was ‘cleansed’. The purpose of this cleansing is to take account of coding errors in
the information and to eliminate properties which are atypical or can be described as outliers.
Therefore, the cleansed dataset only includes those properties which meet the following
selection criteria:
•
purchase price between £ 30,000 and £6 million;
•
number of bedrooms, bathrooms, toilets, living rooms, garages or garage spaces must
not exceed nine in each case;
•
number of habitable rooms must be less than 20;
•
size property must be less than 500 square metres;
•
mortgage advance must be less than 10 times either the main or the additional
applicant’s income.
11
Certain transactions are regarded as atypical and have therefore been excluded from the
analysis. These include:
•
houses sold at ‘ nonmarket’ prices, such as council houses sold at a discounted prices
to their tenants and other sales to sitting tenants in the private sector;
•
houses sold without vacant possession or which are unusual in being mixed freehold
and leasehold;
•
transactions for which there was an error in recording or coding of the data at source;
In addition to these standard cleansing procedures we added a further condition:
•
a house is excluded if the nearest school is further than 10 km away.
The purpose of this last condition is to provide an outer boundary to the urban conurbations
which we have studied.
4. Modelling Framework
In this study we employ the hedonic approach to price measurement in order to determine a
monetary valuation which parents place on the performance of schools. The hedonic pricing
method makes use of price differentials in properties with different attributes to determine the
value of those attributes. These attributes include both physical characteristics of a particular
property such as number of bedrooms, floor area (square metres) and number of bathrooms,
as well as locational variables which in this study include the ‘quality’ of the most proximate
state school. For the purpose of this analysis we assume that the housing stock in the major
urban conurbations that we are studying is not sufficiently elastic in its supply so as to negate
any capitalisation of difference in housing amenity resulting from school performance.
The hedonic approach assumes that when consumers purchase houses they are buying
physical house characteristics, such as floor area, the existence or otherwise of central heating
and a number of bedrooms, as well as the neighbourhood and environmental characteristics
which the houses location offers to a potential purchaser.
By regressing the physical,
neighbourhood and environmental characteristics of a set of properties against the purchase
price as the dependent variable, it is possible to calculate implicit prices for each of the
characteristics.
12
Hedonic pricing methods have been employed to calculate the value of a wide range of both
“welcome” and “unwelcome” local amenities. These include aircraft noise (Levesque, 1994),
air quality (Chay and Greenstone, 1998), hazardous waste sites (McCluskey and Rausser,
2000), woodland (Garrod and Willis, 1992) and proximity to churches (Do et al., 1994;
Carroll et al., 1996). A number of studies, including those by Blomquist et al., (1988) and
Thayer et al., (1992), have attempted to measure more than one environmental “good” using
the same hedonic model.
A key criticism of the hedonic pricing model approach concerns the treatment of omitted
variables, i.e., when a specific local amenity is identified and included in the hedonic
estimation it may be the case that the included variable is picking up the influence of another
excluded local variable, where it may not have been possible to include a reliable indicator of
the impact of the excluded variable. If this is the case, OLS estimation techniques will
provide biased parameter estimates. It may be the case, as Gibbons and Machin (2008) argue,
that it is difficult to measure the supply of an amenity that is accessible from a given location.
In this study, we would argue that locational data allows us to identify the closest school and
therefore, in so far as it is practicably possible, we feel we have identified access to the
‘focus’ amenity of school quality.
As with many previous studies we employ examination performance as the measure of school
quality. However, it is the case that examination performance may reflect the socioeconomic
background of the students who are in the school catchment area. Clapp et al., (2007) found
evidence in Connecticut that students’ test scores have a positive impact on property values.
Furthermore, they found that the ethnic composition of a neighbourhood also impacted
property values. For example, the proportion of Hispanics in a specific neighbourhood
appeared to have a negative impact on property values, although there was evidence that the
impact declines over time.
Hilber and Mayer (2004) argue that because of the capitalisation of school performance into
local property prices, those residents who do not have children have an incentive to see an
improvement in school performance. Furthermore, they argue that such residents have an
incentive to see increased expenditure on public schooling. Hilber and Mayer (2004) also
13
found that the capitalisation of school performance was strongest in those areas where the
supply of housing was inelastic.
Choice of functional form
Box and Cox (1964) have developed a statistical procedure based on likelihood ratio tests to
identify the functional form providing the best fit. This procedure is used to identify the
adequate transformation of the dependent variable, as well as the appropriate transformations
of the explanatory variables considered. However, one limitation of this test is that it cannot
assess any of the qualitative characteristics, due to the fact that they are represented by
dummy variables. As a result, the incorporation of qualitative variables to which price may be
related non-linearly cannot be tested.
The Box-Cox procedure is described in Spitzer (1982) and results reported in Fleming and
Nellis (1985) have indicated that the regression results are substantially more sensitive to
transformations of the dependent variable than of the explanatory variables. According to the
Box-Cox test procedure, a logarithmic transformation of the dependent variable (in this case
house price) is the most appropriate functional form for the hedonic regression model. In
many instances this preference is related to the ease with which parameter estimates can be
interpreted. Furthermore, as Cropper et al., (1988) suggest, the lack of theoretical guidance
on functional form for hedonic price functions has led most researchers to adopt goodness-offit criteria. Indeed, Garrod and Willis (1992) favour the semi-log specification on this basis
while Laasko (1992) points to the ease of interpretation. Consequentially, a semi log model
is used in the results reported below.
Optimal combination of explanatory variables
Analyses were also carried out to determine the optimal combinations and appropriate
transformations of the explanatory variables in the regression equations, based on the
locational and physical house characteristics. In particular, the results showed that floor area
consistently outperforms the number of habitable rooms in explaining house prices and thus
the former variable is incorporated in the final regression model.
Multicollinearity
With such a rich data set it is desirable to use as many variables as possible to define the
house price equation. In practice, some explanatory variables may be correlated with each
14
other – in other words, the problem of multicollinearity may exist. Garrod and Willis (1992)
indicate that multicollinearity is a common problem in hedonic price functions and one which
is often conveniently ignored. However, as multicollinearity is such a significant issue in
hedonic price equations we also undertook a number of other checks to test for it. As outlined
in Belsey et al. (1980), eigenvalues (λ) can be employed to generate conditional indices and
the conditional number (K). Where there is no multicollinearity present the eigenvalues for
each parameter estimate would be equal to 1. As eigenvalues move away from 1 this
indicates the existence of multicollinearity. The conditional indices matrix can be calculated
by dividing eigenvalues λi/λj for all parameter estimates in the preferred equation. As a rule of
thumb, values between 100 and 1000 in the conditional indices matrix would suggest
moderate multicollinearity while values over 1000 would suggest that the problem of
multicollinearity is a serious issue. Furthermore, it is possible to calculate a conditional
number (K) by dividing the maximum eigenvalues attached to a parameter in our preferred
equation by the minimum eigenvalues attached to a parameter in the equation, and taking the
square root of the resulting value; i.e.:
K = √(λmax/λmin)
Again, a general rule of thumb is that if K lies between 10 and 30 there is moderate to strong
multicollinearity and if it exceeds 30 there is severe multicollinearity.
Finally, as the preferred regression equation covered more than one time period we have
indexed the equation using the ‘time dummy’ method (Fleming and Nellis, 1985). Therefore,
with time incorporated as a dummy variable, percentage changes in price can be observed
directly as the coefficient on the time variable, our base year in all cases being 2001.
5. Empirical Results
Our preferred equation emerged from a process of experimentation with functional form and
combinations of explanatory variables. Ultimately our choice of preferred equation focused
upon a combination of ‘goodness of fit’, statistical significance of explanatory variables and
minimisation of multicollinearity.
In order to facilitate comparison across different
geographical spaces we have applied the same functional form (i.e. semi-log) to all of the
equations in our analysis. Furthermore, our preferred equation includes the same set of
15
explanatory variables in all instances. The robustness and consistency of the model in part
reflects the enormous amount of work which has been done by HBoS, as part of their house
price index analysis, over a large number of years7.
The performance of what Graves et al. (1988) term ‘focus’ variables – in this paper, the
school performance variables - is of primary importance. We experimented with dummy
variables including those which separated the dataset into discrete deciles and excluded the
deciles which included the national average and local average school performance; we also
experimented with excluding lower deciles and higher deciles. Finally, we included dummies
to pick up the effect of better than and worse than performances with respect to national and
local averages. In all of these cases the dummy variables failed to work as effectively as the
continuous specification of the school performance variable that we ultimately included in
our preferred equation.
The final regression model uses the natural log of house price as the dependent variable (ln P)
and the following independent variables: dummy variables to capture the impact of house
type (detached, semi-detached, terraced, flat and bungalow); the existence of a garden; none
or partial central heating and the age of the property (split into four distinct classifications properties built between 1919-45, 1945-60, post-1960 and new build). A set of ACORN
ratings dummy variables were included to allow for variability in neighbourhood
characteristics while annual time dummy variables were included to allow for the influence
of inflationary effects over the period 2001-2007 not associated with the physical,
neighbourhood or environmental characteristics of the properties in the dataset. A dummy
variable was incorporated to link each property with the nearest secondary school.
Continuous measures of numbers of bathrooms, toilets, garages and garage spaces and
floorspace (measured in square metres) were also included in the final equation. As noted
earlier, the data set allowed us to experiment with specifications which included the number
of habitable rooms but this specification was rejected in favour of a continuous floorspace
variable on the basis of goodness-of-fit and the overall performance of other variables in our
preferred equation.
7
We are grateful for the guidance and advice which were received from the HBoS economics team in this
respect.
16
Therefore, the model provides an estimate of the influence of property characteristics with
reference to a ‘standard’ house (Forrest et al. 1996). In this study our standard property in
each city was determined on the basis of identifying the modal characteristics in each of the
seven cities. Thus in all cities the standard house had full central heating and no garden while
the excluded house types, house ages and ACORN types are reported in Table 3 below. The
modal characteristics are excluded from our equation and the coefficients on the included
dummy variables can be interpreted as a percentage premium or discount relative to the
excluded modal characteristic.
Table 3: Excluded house type, house age and ACORN code by city
City
House type
House Age8
ACORN
Birmingham
Bristol
Leeds
Liverpool
Greater London
Greater Manchester
Newcastle
Semi-detached
Terraced
Terraced
Semi-detached
Flat
Terraced
Semi-detached
1960+
Pre 1919
1960
1960+
Pre 1919
1960+
1960+
IK
EFG
IK
IK
D
IK
IK
The regression analysis are presented in Tables 4a-g below for each of the cities in turn
8
1960+ denotes properties built after 1960 but not new build.
17
Table 4a: Birmingham
Variable
Coefficient
Standard
Error
T-statistic
Constant
10.482
0.007
1398.444**
House Type
Detached
Terraced
Bungalow
Flat
0.157
-0.114
0.133
-0.200
0.005
0.004
0.012
0.006
30.731**
-30.089**
11.459**
-31.713**
Number of Bathrooms
Number of Toilets
Number of Garages
Number of Garage Spaces
Garden
Size (in square metres)
0.116
0.051
0.083
0.035
0.106
0.003
0.004
0.003
0.003
0.002
0.008
0.000
31.813**
18.532**
29.270**
22.201**
14.004**
72.675**
Availability of Central Heating
None
Partial
-0.066
-0.055
0.004
0.005
-17.616**
-11.718**
Age of Property
Pre-1919
1919-1945
1945-1960
New
0.070
0.071
0.029
0.097
0.005
0.004
0.005
0.006
15.173**
18.879**
6.248**
16.021**
Acorn Classification
A
B
C
D
EFG
H
J
0.617
0.433
0.435
0.444
0.284
0.173
-0.035
0.007
0.006
0.008
0.007
0.004
0.004
0.005
83.179**
66.959**
57.806**
60.832**
71.109**
42.342**
-6.402**
Time
Yr 2002
Yr 2003
Yr 2004
Yr 2005
Yr 2006
Yr 2007
0.234
0.417
0.515
0.580
0.624
0.669
0.006
0.005
0.006
0.005
0.005
0.005
41.514**
76.379**
93.308**
110.609**
119.508**
129.969**
School Quality
0.001
0.000
18.133**
Sample size
Adjusted R²
F-Statistic
**Significant at the 5% or higher level of significance.
38,456
0.772
4348.164
18
Table 4b: Bristol
Variable
Coefficient
Standard
Error
T-statistic
Constant
10.938
0.011
1031.013**
House Type
Detached
Semi
Bungalow
Flat
0.190
0.084
0.125
-0.073
0.008
0.006
0.013
0.008
24.540**
15.107**
9.278**
-8.813**
Number of Bathrooms
Number of Toilets
Number of Garages
Number of Garage Spaces
Garden
Size (in square metres)
0.134
0.061
0.089
0.034
-0.038
0.003
0.005
0.004
0.004
0.003
0.009
0.000
27.555**
14.417**
22.016**
13.584**
-4.268**
56.701**
Availability of Central Heating
None
Partial
-0.014
-0.065
0.007
0.009
-2.098**
-7.418**
Age of Property
1919-1945
1945-1960
1960+
New
-0.048
-0.117
-0.123
-0.103
0.007
0.008
0.006
0.009
-6.862**
-14.913**
-20.193**
-11.172**
Acorn Classification
A
B
C
D
H
IK
J
0.252
0.116
0.103
0.188
-0.068
-0.224
-0.176
0.010
0.008
0.011
0.007
0.006
0.007
0.014
25.580**
15.137**
9.386**
25.879**
-11.032**
-34.156**
-12.246**
Time
Yr 2002
Yr 2003
Yr 2004
Yr 2005
Yr 2006
Yr 2007
0.256
0.346
0.417
0.439
0.534
0.609
0.009
0.008
0.008
0.008
0.007
0.007
29.399**
41.816**
51.046**
56.215**
73.467**
83.455**
School Quality
0.003
0.000
24.614**
Sample size
Adjusted R²
F-Statistic
**Significant at the 5% or higher level of significance.
13,698
0.774
1561.433
19
Table 4c: Leeds
Variable
Coefficient
Standard
Error
T-statistic
Constant
10.133
0.007
1466.668**
House Type
Detached
Semi
Bungalow
Flat
0.230
0.102
0.217
0.042
0.005
0.004
0.007
0.006
43.328**
27.111**
30.926**
6.517**
Number of Bathrooms
Number of Toilets
Number of Garages
Number of Garage Spaces
Garden
Size (in square metres)
0.139
0.071
0.104
0.037
-0.025
0.003
0.004
0.003
0.003
0.001
0.006
0.000
38.986**
23.791**
38.990**
25.507**
-4.177**
83.712**
Availability of Central Heating
None
Partial
-0.103
-0.074
0.003
0.006
-31.379**
-12.351**
Age of Property
Pre-1919
1919-1945
1945-1960
News
-0.005
0.014
-0.025
0.052
0.004
0.004
0.005
0.006
-1.300
3.660**
-5.289**
8.580**
Acorn Classification
A
B
C
D
EFG
H
J
0.612
0.391
0.391
0.505
0.288
0.173
-0.082
0.007
0.006
0.007
0.007
0.004
0.004
0.007
84.666**
68.846**
52.240**
73.340**
77.070**
47.179**
-11.656**
Time
Yr 2002
Yr 2003
Yr 2004
Yr 2005
Yr 2006
Yr 2007
0.211
0.402
0.571
0.686
0.756
0.824
0.005
0.005
0.005
0.005
0.005
0.005
40.816**
80.616**
112.998**
140.757**
160.369**
174.231**
School Quality
0.002
0.000
27.721**
Sample size
Adjusted R²
F-Statistic
**Significant at the 5% or higher level of significance.
44,278
0.794
5681.224
20
Table 4d: Liverpool
Variable
Coefficient
Standard
Error
T-statistic
Constant
10.267
0.010
977.921**
House Type
Detached
Terraced
Bungalow
Flat
0.125
-0.152
0.128
-0.034
0.008
0.006
0.013
0.008
15.994**
-26.672**
9.581**
-4.066**
Number of Bathrooms
Number of Toilets
Number of Garages
Number of Garage Spaces
Garden
Size (in square metres)
0.117
0.057
0.102
0.043
-0.056
0.003
0.005
0.004
0.005
0.003
0.007
0.000
21.849**
13.697**
22.469**
16.398**
-8.352**
52.314**
Availability of Central Heating
None
Partial
-0.065
-0.012
0.006
0.005
-11.774**
-2.249**
Age of Property
Pre-1919
1919-1945
1945-1960
New
0.070
0.130
0.039
0.110
0.007
0.006
0.007
0.008
10.153**
22.866**
5.570**
13.522**
Acorn Classification
A
B
C
D
EFG
H
J
0.598
0.416
0.470
0.452
0.330
0.224
0.030
0.013
0.010
0.011
0.011
0.006
0.006
0.013
44.689**
41.192**
42.314**
39.993**
56.030**
38.875**
2.257**
Time
Yr 2002
Yr 2003
Yr 2004
Yr 2005
Yr 2006
Yr 2007
0.158
0.386
0.625
0.739
0.783
0.814
0.008
0.008
0.008
0.007
0.007
0.007
20.013**
51.218**
80.198**
100.985**
109.475**
112.168**
School Quality
0.001
0.000
12.589**
Sample size
Adjusted R²
F-Statistic
19,506
0.772
2200.021
**Significant at the 5% or higher level of significance.
21
Table 4e: Greater London
Variable
Coefficient
Standard
Error
T-statistic
Constant
11.499
0.004
2940.455**
House Type
Detached
Semi
Terraced
Bungalow
0.223
0.208
0.165
0.267
0.005
0.003
0.002
0.008
45.849**
67.335**
70.117**
34.091**
Number of Bathrooms
Number of Toilets
Number of Garages
Number of Garage Spaces
Garden
Size (in square metres)
0.254
0.064
0.035
0.012
0.048
0.004
0.002
0.002
0.002
0.001
0.002
0.000
137.899**
40.188**
17.343**
9.728**
22.272**
167.389**
Availability of Central Heating
None
Partial
-0.023
-0.029
0.002
0.004
-9.640**
-7.933**
Age of Property
1919-1945
1945-1960
1960+
New
-0.100
-0.183
-0.211
-0.178
0.002
0.003
0.002
0.004
-44.586**
-59.782**
-96.259**
-42.480**
Acorn Classification
A
B
C
EFG
H
IK
J
0.020
-0.126
-0.227
-0.294
-0.373
-0.483
-0.291
0.005
0.005
0.008
0.002
0.003
0.003
0.002
4.085**
-27.624**
-29.390**
-124.190**
-120.579**
-138.289**
-124.842**
Time
Yr 2002
Yr 2003
Yr 2004
Yr 2005
Yr 2006
Yr 2007
0.163
0.245
0.279
0.330
0.420
0.523
0.003
0.004
0.003
0.003
0.003
0.003
47.055**
69.459**
82.525**
107.994**
140.998**
178.558**
School Quality
0.001
0.000
24.127**
Sample size
Adjusted R²
F-Statistic
**Significant at the 5% or higher level of significance.
167,807
0.699
13003.140
22
Table 4f: Greater Manchester
Variable
Coefficient
Standard
Error
T-statistic
Constant
10.082
0.006
1580.512**
House Type
Detached
Semi
Bungalow
Flat
0.233
0.133
0.218
0.111
0.005
0.003
0.007
0.006
46.175**
38.038**
30.836**
20.015**
Number of Bathrooms
Number of Toilets
Number of Garages
Number of Garage Spaces
Garden
Size (in square metres)
0.153
0.054
0.087
0.061
-0.083
0.003
0.003
0.003
0.003
0.001
0.004
0.000
47.627**
20.702**
33.226**
41.786**
-20.070**
94.968**
Availability of Central Heating
None
Partial
-0.036
-0.062
0.004
0.007
-10.072**
-9.517**
Age of Property
Pre-1919
1919-1945
1945-1960
New
0.102
0.082
0.006
0.110
0.004
0.004
0.005
0.005
26.737**
22.602**
1.408
22.106**
Acorn Classification
A
B
C
D
EFG
H
J
0.712
0.449
0.437
0.508
0.337
0.191
-0.023
0.007
0.005
0.006
0.006
0.004
0.004
0.007
106.397**
93.784**
67.623**
81.769**
95.624**
52.921**
-3.168**
Time
Yr 2002
Yr 2003
Yr 2004
Yr 2005
Yr 2006
Yr 2007
0.168
0.332
0.545
0.651
0.708
0.759
0.005
0.005
0.005
0.004
0.004
0.004
34.905**
70.920**
117.642**
147.466**
166.047**
180.704**
School Quality
0.002
0.000
21.378**
60,502
Sample size
Adjusted R² 0.776
7006.727
F-Statistic
**Significant at the 5% or higher level of significance.
23
Table 4g: Newcastle
Variable
Coefficient
Standard
Error
T-statistic
Constant
10.509
0.012
856.103**
House Type
Detached
Terraced
Bungalow
Flat
0.084
-0.089
0.102
-0.173
0.007
0.005
0.009
0.007
11.668**
-16.156**
10.794**
-23.150**
Number of Bathrooms
Number of Toilets
Number of Garages
Number of Garage Spaces
Garden
Size (in square metres)
0.144
0.052
0.121
0.035
-0.014
0.003
0.005
0.004
0.004
0.003
0.007
0.000
28.842**
13.255**
30.089**
11.565**
-2.056**
55.719**
Availability of Central Heating
None
Partial
0.017
-0.064
0.006
0.010
2.962**
-6.671**
Age of Property
Pre-1919
1919-1945
1945-1960
New
0.114
0.062
0.016
0.093
0.006
0.006
0.006
0.008
17.712**
11.064**
2.555**
11.883**
Acorn Classification
A
B
C
D
EFG
H
J
0.569
0.348
0.350
0.465
0.270
0.151
0.071
0.012
0.009
0.010
0.009
0.005
0.006
0.011
47.837**
40.009**
33.478**
54.551**
49.393**
25.714**
6.552**
Time
Yr 2002
Yr 2003
Yr 2004
Yr 2005
Yr 2006
Yr 2007
0.237
0.382
0.527
0.575
0.611
0.663
0.010
0.009
0.009
0.009
0.009
0.009
23.525**
41.649**
58.073**
62.962**
68.039**
73.474**
School Quality
0.001
0.000
3.566**
15,206
Sample size
0.729
Adjusted R²
1361.268
F-Statistic
**Significant at the 5% or higher level of significance
24
Table 5 below reports the regression coefficient on the school performance variable in each
of the seven major English cities that we have studied. The table also includes the adjusted R2
values and number of observations for each estimated equation.
Table 5: Coefficient on school performance variable
City
Coefficient
Birmingham
Bristol
Leeds
Liverpool
London
Greater Manchester
Newcastle
0.001
0.003
0.002
0.001
0.001
0.002
0.001
Adjusted R2
0.772
0.774
0.794
0.772
0.699
0.776
0.729
Observations
38,456
13,698
44,278
19,506
167,807
60,502
15,206
The above results suggest that in Birmingham, Liverpool, London and Newcastle, for every
10% point improvement in the average GCSE pass rate (% or more GCSEs at grades A*-C),
property prices increase by 1%. This would imply that, all other things being equal, a school
which had a pass rate of 80% would give rise to a local property price premium of 4%
compared to properties where the local state-funded secondary school had a pass rate of 40%.
In order to facilitate comparison across a number of studies, Gibbons and Machin (2008)
suggest examination of the premium in local property prices created by a performance which
is one standard deviation away from the mean performance in the local schooling area. Table
6 below reports these results for the seven cities/urban conurbations we have considered in
this study. The table shows the average standard deviation of results over the period of this
study (2001-2007) and the resulting premia based on the school performance coefficients we
have calculated as part of this study.
The results presented in Table 6 are in line with a range of both UK and international studies
surveyed by Gibbons and Machin (2008). The depth and range of our study provides
additional weight to the hypothesis that good school performance is capitalised into property
prices and the extent of that capitalisation. At the same time, the results reported here suggest
that school performance one standard deviation above the local average produces different
results in different English cities.
25
For example, in Birmingham an additional one standard deviation of school performance
increases residential property prices by 1.7% whereas in Bristol an additional standard
deviation of school performance increases property prices by 5.4%. The table presents data
on average school performance in all seven cities and the range of performance, indicating
that in cities where the average school performance is higher and the range of average school
performance scores narrower we might expect the residential price premium to be lower. In
other words if there are few high performing schools in a particular city we would expect
parents to rush to buy houses in areas which guaranteed access to that high performing
school.
Table 6: The residential house price premium for a local school with GCSE
performance one standard deviation above the local average
City
Birmingham
Bristol
Leeds
Liverpool
London
Greater Manchester
Newcastle
Coefficient
0.001
0.003
0.002
0.001
0.001
0.002
0.001
Average
standard
deviation of
GCSE results
House price
premium for
an additional
standard
deviation
17.31
18.07
18.33
21.08
18.64
15.96
16.06
+1.7%
+5.4%
+3.6%
+2.1%
+1.9%
+3.2%
+1.6%
Table 7 below reports the percentage of students obtaining 5 or more GCSE’s at grades A* to
C, in the seven cities studied, over the period 2001 to 2006. Students’ performance in 2007 is
not reported in this table as our model suggests that house prices in any one year will be
contingent upon the previous years GCSE performance of the nearest state school.
A comparison of Table 7 and Table 6 shows that the top four performing cities in terms of
secondary school performance, namely Greater London, Liverpool, Birmingham and
Newcastle, all have low school performance coefficients of 0.001. The next two cities in
terms of school performance, as measured by GCSE pass rates at grades A* to C - Greater
Manchester and Leeds - have school performance coefficients of 0.002 and the worst
26
performing city in terms of school performance, Bristol, has the highest school performance
coefficient of 0.003.
Table 7: Percentage of students gaining five or more GCSE passes at grades A* to C
(2001-2006)
2001
2002
2003
2004
2005
2006
Birmingham
Mean
36.9
41.3
46.1
48.1
53.9
56.7
Bristol
Mean
32.9
32.2
38
36.9
37.5
46.4
Leeds
Mean
38
39.6
43.2
45.2
50.3
52.3
Liverpool
Mean
36.7
39.5
41.7
45.3
51.6
57.2
G.London
Mean
44.6
47.3
50.2
53
54.9
58.1
G.Manchester
Mean
41.9
43.7
46
46.4
50.4
53.7
Newcastle
Mean
39.8
41.9
41
45.2
59.6
56.1
Our results are essentially stating that in cities where state school performance is poor, the
rate at which ‘good’ school performance is capitalised into local residential property values is
the greatest. This is entirely in line with our a priori expectations. Therefore, Bristol which
has the poorest state school performance of the seven cities in terms of percentage of
students passing five or more GCSE’s at grades A* to C is the city in our sample where
parents are willing to pay most to purchase a property which is close to a high performing
school.
7. Conclusions
The following conclusions follow from the results reported here:
•
The difference in state funded secondary school performance in GCSE examinations
is consistently capitalised into residential property prices in all of the seven major
27
English cities/conurbations that we have studied, ie., residential properties which are
located proximate to high performing state-funded secondary schools attract a price
premium.
•
The extent of the price premium varies between different cities. In Greater London,
Liverpool, Birmingham and Newcastle a 10% point differential in GCSE performance
produces a 1% point differential in residential properties located close to the high
performing school. In Greater Manchester and Leeds the same differential in school
performance produces a 2% point increase in residential properties located close to
the high performing school. Finally, in Bristol a 10% point differential in GCSE
performance by a specific school produces a 3% point increase in residential
properties located close to the high performing school. Note that this implies that in
Bristol, a property where the closest school had a pass rate of 80% compared to a
property where the closest school had a pass rate of 40% would attract a price
premium of 12% points, all other things being equal.
•
Those properties which attracted the highest price premia for ‘good’ school
performance were in cities where the average pass rate in GCSE examinations was the
lowest. In other words, in those cities where the supply of good schools was lower,
the price premium for a ‘good’ school, in terms of the rate at which school
performance was capitalised into local property prices, was higher.
•
This last point has implications for students’ ability to access high performing
schools, in so far as their parents have an inability to pay the price premium required
to purchase a house which would guarantee access. The response of the current UK
government to the paradox of pupils being denied access to state funded education
due to their parents inability to pay a ‘shadow’ price that exists in the local residential
property market is to introduce a lottery into the allocation of places at
‘oversubscribed’ schools. We would suggest that this is inappropriate in that it deals
with the symptoms of the problem rather than the causes. We would argue that the
government should focus on improving school quality as doing so would remove the
extent of the price premium that ‘high performing’ schools attract, thereby reducing
the extent to which students would be excluded from state funded secondary
education because of their parents inability to ‘pay the price’ in local residential
property markets.
28
•
Finally, it may be the case that the private sector may be willing to fund
improvements in state funded secondary schooling, as low performing schools will
mean that they may find it difficult to attract workers with siblings of secondary
school age to areas where secondary school performance is poor. In such
circumstances workers may require wages which allow them to educate their children
at private schools or, in extremis they may not move to a particular city. This point is
worthy of further investigation.
References
Bartik, T.J. (1988) Measuring the benefits of amenity improvements in hedonic price models.
Land Economics, Vol. 64, pp 172-183.
Bogart, W.T. and Cromwell, B.A. (1997) How much is a good school district worth?
National Tax Journal, Vol. 50, pp 215-232.
Box, G.E.P. and Cox, D.R. (1964) An analysis of transformations. Journal of the Royal
Statistical Society, Series B, Vol. 21, pp 1-43.
Brasington, D. (2002) Edge versus center: Finding common ground in the capitalisation
debate. Journal of Urban Economics, Vol. 52, pp 245-268.
Brasington, D. and Haurin, D.R. (2006) Educational outcomes and house values: A test of the
value-added approach. Journal of Regional Science, Vol. 46, No. 2, pp 245-268.
CACI, (2006) The Acorn Users Guide, London
Carroll, T.M., Clauretie, T.M. and Jensen, J. (1996) Living next to godliness: Residential
property values and churches. Journal of Real Estate Finance and Economics, Vol.12, pp
319-330.
Chay, K.Y. and Greenstone, M. (1998) Does air quality matter? Evidence from the housing
market. NBER Working Paper 6826, December.
29
Cheshire, P. and Shepherd, S. (2004) Capitalising the value of free schools: the impact of
supply characteristics and uncertainty. Economic Journal, Vol.114, pp 397-424.
Clapp, J., Nanda, A. and Ross, S.L. (2007) Which school attributes matter? The influence of
school district performance and demographic composition on property values. Journal of
Urban Economics, Vol. 63, pp 451-466.
Crone, T.M. (1998) House prices in the quality of public schools: what are we buying?
Business Review Federal Reserve Bank of Philadelphia, (September/October), pp 3-14.
Cropper, M. L., Deck, Leland B. and McConnell, Kenneth, E. (1988) On the Choice of
Functional Form for Hedonic Price Functions. Review of Economics and Statistics, Vol. 70,
No. 4, pp 668-675.
Do, A.Q., Wilber, R.W. and Short, J.L. (1994) An empirical examination of the externalities
of neighbourhood churches on housing values. Journal of Real Estate Finance and
Economics, Vol. 9, pp 127-136.
Downes, T.A. and Zabel, J.E. (2002) The impact of school characteristics on house prices:
Chicago 1987-1991. Journal of Urban Economics, Vol. 52, pp 1-25.
Elkins, T. and Elliott, J. (2004) Competition and control: the impact of government regulation
on teaching and learning in English schools. Research Papers in Education, Vol. 19, No. 1,
pp 15-30.
Fleming, M.C. and Nellis, J.G. (1985) The Application of Hedonic Indexing Methods: A
Study of House Prices in the United Kingdom. Statistical Journal of the United Nations, ECE
3, pp. 249-270.
Forrest, D., Glen, J. and Ward, R. (1996) The impact of a light rail system on the structure of
house prices: a hedonic longitudinal study. Journal of Transport Economics and Policy, Vol.
30, No. 1, pp 15-29.
30
Garrod, G.D. and Willis, K.G. (1992) Valuing goods’ characteristics: an application of the
hedonic price method to environmental attributes. Journal of Environmental Management,
Vol. 34, pp 56-76.
Gibbons, S. and Machin, S. (2008) Valuing school quality, better transport, and lower crime:
evidence from house prices. Oxford Review of Economic Policy, Vol. 24, No. 1, pp 99-119.
Graves, P., Murdoch, D.C., Thayer M.A. and Waldman D. (1998) The Robustness of
Hedonic Price Estimation: Urban Air Quality. Land Economics, Vol. 64, No. 3, pp 220-233.
Haurin, D. and Brasington, D. (1996) School quality and real house prices: inter and intrametropolitan effects. Journal of Housing Economics, Vol. 5, pp 351-368.
Hayes, K.J. and Taylor, L.L. (1997) Neighbourhood school characteristics: What signals
quality to homebuyers? Federal Reserve Bank of Dallas Economic Review, Vol. 4, pp 2-9.
Hilber, C.A.L. and Mayer, C. (2004) Why do households without children support local
public schools? Linking house price capitalisation to school spending. Working Paper, No.
10804, NBER, October.
Hoyt, W.H. (1999) Leviathan, local government expenditures, and capitalization. Regional
Science and Urban Economics, Vol. 29, pp 155-171.
Lambert, D. and Lines, D. (2000) Understanding assessment: Purposes, perceptions, practice
London: Routledge Falmer.
Leech, D. and Campos, E. (2003) Is comprehensive education really free? A case study of the
effects of secondary school admissions policies on house prices in one local area. Journal of
the Royal Statistical Society, Vol. 166, Part 1, pp 135-154.
Levensque, T.J. (1994) Modelling the effects of airport noise on residential housing markets:
A case study of Winnipeg International Airport. Journal of Transport Economics and Policy.
Vol. 28, pp 199-210.
31
McCluskey, J.J and Rausser, G.C. (2000) Hazardous waste sites and housing appreciation
rates, mimeo. University of California: Berkeley.
Rosenthal, L. (2003) The value of secondary school quality. Oxford Bulletin of Economics
and Statistics, Vol. 65, No. 3, pp 329-355.
Starrett, D. (1981) Land value capitalization in local public finance. Journal of Political
Economy, Vol. 89, pp 306-327.
Taylor, L.L. (2005) Revealed-preference measures of school quality. In L. Stiefel, A.E.
Schwartz, R. Rubinstein, J. Zabel (Eds), Measuring school performance and efficiency:
Implications for practice and research. Eye on Education: Larchmont, NV, pp 163-183.
Weimer, D.L. and Wolkoff, M.J. (2001) School performance and housing values using non
contiguous district and incorporation boundaries to identify school effects. National Tax
Journal, Vol. 45, pp 231-254.
Wildasin, D.E. (1987) Theoretical Analysis of Local Public Economics. In E.S. Mills (Ed),
Handbook of Regional and Urban Economics, Vol. II, North Holland, Amsterdam, pp 11321178.
32