Corporate Professional Local

Transport accessibility and land value:

a case study of Tyne and Wear
RICS Research paper series
Volume 7 Number 3 June 2007

Dr Hongbo Du

Dr Corinne Mulley
Newcastle University, UK

Research www.rics.org
Dr Hongbo Du received her PhD degree in transport engineering from the
School of Civil Engineering and Geosciences at Newcastle University, UK by
achieving unique expertise in relatively new transport geography field with the
application of a cutting edge spatial analysis modelling, Geographically Weighted
Regression in transport area. She is a registered transport planner in P.R.China
and with ten years of industrial experience in transport planning, her particular
research interests lie in the policy issues arising from the interaction of land use
and transport studies.

Dr Corinne Mulley is a Senior Lecturer in Transport Economics in the Transport

Operations Research Group (TORG) at Newcastle University, UK.
Her research concentrates on the important interface between academia,
business, and government where economics and transport studies interact
(e.g. deregulation and privatisation) both in the UK and in Europe. Areas include
the evaluation of transport projects, particularly where the impacts of urban
transport policies and investments have an economic perspective, the identification
of risk in transport evaluation; the quality of public transport provision both
internally to the firm (internal benchmarking) and of passenger perceptions and
a long standing interest in road transport history. Recent publications include
co-authoring the Best Practice Guide entitled Network planning for high quality
public transport in small and medium sized cities and Demand Responsive
Transport Telematics Good Practice Guide, commissioned by the Department
for Transport.
Transport
accessibility
and land value:
a case study of
Tyne and Wear

Volume 7 Number 3

June 2007
© RICS Aims and scope of the RICS Research
Paper Series
June 2007
ISSN 1464-648X The aim of the RICS Research Paper Series is
ISBN 978-1-84219-328-0 to provide an outlet for the results of research
relevant to the surveying profession. Papers
Published by: range from fundamental research work
through to innovative practical applications of
RICS new and interesting ideas. Papers combine
12 Great George Street academic rigour with an emphasis on the
London SW1P 3AD implications in practice of the material
United Kingdom presented. The Series is presented in a
readable and lucid style which stimulates the
The views expressed by the author(s) are not interest of all the members of the surveying
necessarily those of the RICS nor any body profession.
connected with it. Neither the author(s), nor
the RICS accept any liability arising from the
use of this publication.

This paper can be copied free of charge for

teaching and research purposes, provided that:

• the permission of the RICS is sought in

advance
• the copies are not subsequently resold
• the RICS is acknowledged

4 l RICS Research
Editorial Team Charles Cowap J M Lusugga Kironde Ali Parsa
Harper Adams University College UCLAS University of Ulster
Stephen Brown England Tanzania Northern Ireland
Head of Research RICS
12 Great George Street Hoffie Cruywagen Andrew Knight Rob Pickard
London SW1P 3AD Association of Quantity Surveyors Nottingham Trent University University of Northumbria
United Kingdom in South Africa, South Africa England England
E: sbrown@rics.org Julian Diaz III Richard Laing Frances Plimmer
T: +44 (0)20 7334 3725 Georgia State University Robert Gordon’s University Kingston University
United States of America Scotland England
Amy Roberts
RICS Tim Dixon Chris Leishman Gwilym Price
Oxford Brookes University University of Glasgow University of Glasgow
E: aroberts@rics.org England Scotland Scotland
Panel of referees Keith Duff David Lewis David Proverbs
Consultant Royal Agricultural College University of Wolverhampton
Kwame Addae-Dapaah United Kingdom England England
National University of Singapore
Singapore Richard Fellows Colin Lizieri Les Ruddock
University of Hong Kong University of Reading University of Salford
Akintola Akintoya Hong Kong England England
Glasgow Caledonian University
Scotland Christopher Fortune SM Lo Sarah Sayce
Heriot-Watt University City University of Hong Kong Kingston University
Saleh Al-Hathloul Scotland Hong Kong England
Alhathloul lilumran
Kingdom of Saudi Arabia Karen Gibler Sim Loo Lee Karl-Werner Schulte
Georgia State University National University of Singapore University of Regensberg
Adarkwah Antwi United States of America Singapore Germany
University of Wolverhampton
England Andy Hamilton Sui Pheng Low Rainer Schultz
University of Salford National University of Singapore University of Aberdeen
Ghassan Aouad England Singapore Scotland
University of Salford
England Cliff Hardcastle David Lowe Martin Sexton
Glasgow Caledonian University University of Manchester University of Salford
David Baldry Scotland England England
University of Salford
England Barbara Haward Ahmed M Saleh Ouf Low Sui Pheng
University of Portsmouth Government of Sharjah National University of Singapore
Malcolm Bell England United Arab Emirates Singapore
Leeds Metropolitan University
England John Henneberry John Mansfield Francois Viruly
University of Sheffield Nottingham Trent University University of Witwatersrand
Roderick Bisset England England South Africa
Faithful & Gould
Scotland Robert Home Pat McAllister Clive Warrren
Anglia Ruskin University University of Reading University of Queensland
Alistair Blyth England England Australia
University of Westminster
England Aart Hordijk William McCluskey Christine Whitehead
ROZ Real Estate Council University of Ulster London School of Economics
Thomas Bohn The Netherlands Northern Ireland England
University of Leipzig
Germany Michael Hoxley James Meikle Peter Wyatt
Anglia Ruskin University Davis Langdon University of the West of England
Paul Bowen England United Kingdom England
University of Cape Town
South Africa Will Hughes Stuart Morley Saad Yahya
University of Reading GVA Grimley Saad Yahya Associates
Spike Boydell England United Kingdom Kenya
University of Technology, Sydney
Australia Eddie Hui Wordsworth Odame Larbi
Hong Kong Polytechnic University Ministry of Lands, Foresty & Mines
Jagjit Chadha Hong Kong Ghana
BNP Paribas UK Holdings Ltd
United Kingdom Norman Hutchison Modupe Omirin
University of Aberdeen University of Lagos
S E Cheung Scotland Nigeria
City University of Hong Kong
Hong Kong L Jide Iwarere Jacob Opadeyi
Howard University University of the West Indies
Chris Cloete United States of America Trinidad and Tobago
University of Pretoria
South Africa Ramin Keivani Allison Orr
Oxford Brookes University University of Glasgow
England Scotland
Transport accessibility and land value:
a case study of Tyne and Wear
Hongbo Du and Corinne Mulley, Newcastle University, UK

Abstract
This study is concerned with the relationship between transport accessibility and increases in land value,
motivated by a desire to inform the land value capture debate. Focusing on residential land, here land value
is examined through the form of house prices, using the Tyne and Wear Region of the UK as a case study.
In a typical hedonic price model, adopted widely in property studies, the assumptions of multiple regression
are observed, where the relationship to be modelled holds everywhere in the study area. However, this is often
violated in the analysis of house prices due to the presence of spatial effects.

Thus, a relatively new technique, Geographically Weighted Regression (GWR), which addresses the issue of
spatial effects (Fotheringham et al. 2002), is employed to examine the relationship between transport
accessibility and land value. By embodying spatial coordinates into the traditional global regression model,
GWR provides a set of local estimates using a weighted least squares process, where the weights are linked
to the distance of the observation to the location of the regression point. This study identifies that the
relationship between transport accessibility and land value varies over space. This suggests that a land value
capture policy must take into account this lack of homogeneity and that the application of a uniform ‘tax’
would be inappropriate.

Contact: Acknowledgments

Grace Hongbo Du The authors would like to acknowledge the support from
Email: duhongbo@gmail.com Trevor Arkless at Newcastle City Council, on behalf of
Tyne and Wear Plan Partnership, for his assistance with
the acquisition of transport accessibility data. They also
acknowledge the funding support from RICS Education
Trust, School of Civil Engineering and Geosciences of
Newcastle University, UK, Rees Jeffreys Road Fund, The
Henry Lester Trust Ltd and Lincoln Institute of Land
Policy, US. The work described in this paper is a
shortened form of a thesis submitted for the degree of
PhD at the University. A copy of the thesis is available
from the University Library.
Contents
1 Introduction 5

2 Literature Review 6

3 Geographically Weighted Regression (GWR) 8

4 The Case Study 12

5 Data Collection 14

6 Data Analysis 17

7 Discussion and Conclusion 38

References 41

RICS research paper series 44

1 Introduction

lassical urban land economics This study also took the opportunity to expand

C identified the trade-off between

transport cost and land rental thus
explaining why a parcel of land
located in city centre area with less transport
cost has a higher value than a parcel of land
located in the outskirts of the city. With the
development of economies, the investment on
transport infrastructure to meet the increasing
demand for mobility requires substantial funds
from government for its implementation.
However, there are many competing demands
on public funds and lack of sufficient transport
infrastructure funding to meet all demands has
been a common issue worldwide. In this
context, the concept of a policy based on land
value capture (LVC) has drawn much attention
with various tax based or betterment based
approaches having been explored as possible
mechanisms to finance transport
infrastructure. Indeed, behind these LVC
mechanisms, there is the fundamental
question that needs to be answered. This is: to
what extent can transport infrastructure uplift
land value?
the case study from SME to the whole Tyne
and Wear Region and specifically includes an
examination of transport accessibility and the
impact of Metro access.
Land value is the present value or capitalised
value of land (Mills 1972). Classical land
theories, as expounded by Mills (1972), are
concerned with two types of land –
unimproved land which is without structures
and improved land where the value includes
the value of structures and other capital, such
as utilities pipes, invested in the land. In this
study, the focus is on residential land value.
However, house prices capture the value of
the land and the building that is located on it.
As a result, the methodology standardises for
the aspects of the building so that the
relationship between transport accessibility
and land value can be identified.

In an attempt to answer this question, the

“ ...to what extent
extension of the Tyne and Wear Metro to
Sunderland which opened in March 2002 was
can transport
examined (Du and Mulley 2007) but this study
failed to provide evidence for a significant
infrastructure
positive effect on land value resulting from the
Sunderland Metro Extension (SME). However,
this study identified methodological concerns
uplift land value? “
which could only be resolved by taking a more
disaggregate approach to the issue. The more
disaggregated research is the focus of this
study.

www.rics.org/research RICS Research l 9

2 Literature review
T
here has been substantial research
into this issue in the US, but relatively
limited research within the UK. This
review first considers the US experience
before turning to work focussed on UK cities.
In the US, a study on Philadelphia’s commuter
rail service identified a 3.8% and a 10%
premium for house values in suburban
Philadelphia and Southern New Jersey
respectively by comparing the median house
prices served and not served by a rail line
(Voith 1991). In Atlanta, a study found positive
premiums in lower income households and
negative premiums in higher income
households associated with the MARTA East
Line (Nelson 1992), while another study
claimed negative effects occured from rail
transit, but positive effects for highways, in
terms of office rents (Bollinger et al. 1998).
Using a comparison method, Gatzlaff, Smith
(1993) and Armstrong (1994) uncovered up
to 5% and 6.7% premiums for houses in the
10 l RICS Research
communities served by Miami Metrorail and
commuter rail service in Boston respectively.
By examining five transit systems with various
service levels, as well as highways, in
California, Landis et al. (1994) revealed that
the systems with high ridership and frequent
service, such as BART and the San Diego
Light Rail, generate significant positive
premiums of up to $2.72 per metre to the
nearest station, whilst the systems serving a
more limited market, such as with the San
Jose Light Rail and CalTrain, produce
insignificant or negative premiums in house
prices, with no effect found as a result of
proximity to highways. Chen et al (1997)
claimed a 10.5% uplift in the value of single-
family properties from the Portland light rail.
Hack (2002) gave evidence of house price
increases, up to 25%, resulting from Dallas
DART rail transit, whilst Weinstein and Clower
(2002) showed positive results for both
residential and commercial property in relation
to the Dallas rail transit, with a relatively simple
comparison method.
Cervero has conducted a series of studies into
this topic since 1993 and obtained
inconsistent results. Positive but marginal
effects on commercial land was found in
Washington D.C. and Atlanta (Cervero and
Landis 1993; Cervero 1994; Cervero and
Duncan 2001) while significant benefits to
commercial land from rail transit was found in
Santa Clara County (Cervero and Duncan
2001), which is similar to the result from the
same case study by Weinberger (2001). More
recent studies have had mixed findings for the
impact of rail transit on both commercial and
residential land in San Diego County and Los
Angeles County (Cervero and Duncan
2002a;2002b).
In the UK, Riley, a south London property
developer, estimated that the increases in land
value around the new London Jubilee Line
Extension underground stations was nearly
four times the building cost of the line
(Riley 2001). However, a recent study failed
to identify any statistically significant effect
for Phase 1 of the study using a hedonic
price model and this was substituted by the
adoption of an ‘agents survey’ in Phase 2
(Chesterton 2002).
Although the latter demonstrates positive
results, the methodology is not as robust as
the hedonic model used to examine in
Phase 1. No significant effect was found
in the Croydon Tramlink stage two study,
although this was suspected to do with timing
issues (RICS 2004).
The study into the longer-term impacts of Tyne
& Wear metro also failed to find statistical
evidence of land value uplift resulting from the
introduction of metro using a comparison
methodology (TRRL 1993). With a hedonic
price approach, similar findings were identified
in the Sheffield Supertram study where
negative effects were found during the
construction stage but these disappeared after
opening (Lawless and Gore 1999). In the
Manchester Metrolink study, a surprisingly
significant negative result of up to -8.1% was
discovered (Forrest and Glen 1995).
Having been aware of the variation existing in
housing market, Adair et al (2000) applied
hedonic price models in a sub-market analysis
and found significant positive premiums, of up
to 14%, in only two out of fourteen areas, with
insignificant results in the other twelve areas.
www.rics.org/research
No discernable effects have been found in an
examination of the changes in residential and
commercial property value following the
operation of Croydon Tramlink even when a
more sophisticated approach is used (the
Kriging and Inverse Diverse Weighted (IDW)
method) (RICS 2004).
The experiment with GWR in RICS (2004)
study and the empirical study of Atisreal
(2005) have given a positive sign for the
application of GWR in the examination on the
relationship between transport accessibility
and land value although both identify that a
detailed and relatively large data set is
required.
Overall, the findings of the literature are mixed
with positive and negative results, and the
variations from case to case can be clearly
seen. As the study on this topic advances,
these variations have become more apparent
as attempts to model the sub-effect of
different transport facilities or the sub-market
of housing progress.
However, there are limitations in many of the
empirical studies since global statistics, rather
than local statistics, were utilised in these
studies using spatial data, which violate the
assumption of the traditional multiple
regression model which requires that
observations are statistically independent from
one another. Some methods dealing with
spatial data, such as Spatial Autoregressive
models have been attempted to resolve these
problems in this subject area but they have
not been successful. Hence, a local model
which is better at handling spatial data is
expected to offer insights in this area.
RICS Research l 11
3 Geographically Weighted
Regression (GWR)
I
n the studies described above, a number
of different methodologies have been
utilised. In the context of changes in land
value following improvements in accessibility
created by a new transport infrastructure, the
most basic methodology is the comparison
method. This method compares the change to
an area benefiting from enhanced accessibility
with a control area similar in nature but not
having the advantage of improved accessibility.
Many studies used the more sophisticated
approach of hedonic price modelling. If it is
assumed that the ‘price’ of a house is
determined by a bundle of characteristics
relating to factors related to the property’s
internal features (such as number of
bedrooms) and external factors (such as those
relating to the neighbourhood in which the
property is located and transport accessibility)
then a hedonic model allows the implicit prices
of these characteristics to be estimated.
Consequently, property prices can be seen as
a function of a group of variables contained by
three vectors:
Pi = f (H, N, T) (1)
Where H, N and T is a vector of property
features (H), the neighbourhood environment
(N) and transport accessibility (T) respectively.
There are a number of problems with
determining how a hedonic model is specified
(So et al. 1997; Sirmans et al. 2005) but the
most serious disadvantage in a study
investigating the effect of transport
accessibility on land value is that hedonic
modelling fails to take account of spatial
12 l RICS Research
dependencies and only gives an average
impression of the relationship between the
variables under study.
When looking at houses in different
neighbourhoods, hedonic modelling cannot
fully explain the social processes giving rise
to the spatial effects of spatial dependency
(spatial autocorrelation) and spatial non-
stationarity (spatial heterogeneity). This is the
distinction between a global model (as
provided by hedonic modelling) and a local
model, where the spatial effects can be
incorporated and which require local statistics
for analysis.
This distinction between local and global can
be clearly understood by an example: when
looking at the relationships between variables,
such as the age of houses and house prices,
would we expect any difference between the
global statistics and local statistics? The global
statistics for England may show that the age
of houses does not affect house prices
significantly. In fact, in some parts of England,
some old houses, such as houses built in
Victorian times, might have character thus
generating higher prices than newer houses.
Whereas in other urban areas, older houses
which might have had lower standards to
house workers in rapidly expanding cities in
the middle of the nineteenth century, might be
in poor condition, resulting in substantially
lower prices than newer houses.
The contrasting relationships in different areas
may cancel, to different degrees, the
significant effects of one another and give a
result that, on average across England, the
age of a house appears to have no impact on
 the house price. This shows how misleading
the results from global statistics could be in
examining the relationships in the above
example. Hence, a set of local statistics, in
which data are analysed at local level, would
provide more accurate information on the
relationship to be examined.
“
...on average across
England, the age of a
house appears to “ where (ui, vi ) refers to the coordinates of the
have no impact on
the house price.
Another major difference between global and
local statistics arises from the way in which
these statistics can be presented. Global
statistics are non-mappable single-valued
statistics with averaged or summed data for
the area under consideration whilst local
statistics are mappable multi-valued statistics
with local disaggregations of global statistics.
Hence a global model is calibrated by data
which are not tied to any geographical or
mappable reference in contrast to local model
yielding local parameter estimates is a model
calibrated by data tied to specific geographical
loci.
Traditional multiple regression models are
global models (as are hedonic price models in
this context) and Geographically Weighted
Regression (GWR) models, developed to
achieve local forms of spatial modelling, are
local models. GWR models can improve upon
www.rics.org/research
the explanation offered by the global models
produced by hedonic modelling since they
allow for spatial variation or non-stationarity in
the regression parameters. A traditional
regression model can be written as:
yi = β0 + Σk βk χik + εi (2)
By allowing local parameters to vary a GWR
model is expanded to:
yi = β0 (ui, vi ) + Σk βk (ui, vi ) χik + εi (3)
i th point in space. This can be fitted by
weighted least squares to give an estimate of
the parameters at the location (ui, vi ) and a
predicted value. In this model, data nearer
(ui, vi ) is weighted more heavily than data
further away. It is notable that the location
(ui, vi ) is regression point and so it is not
necessarily the location of a data point. By this
geographically weighted calibration, estimates
of the parameters are made for each data
point with coordinates, which is then mapped
for interpretation.
Before considering the weighting process
used in GWR, it is worth digressing a little to
understand how the presence of spatial
effects cause problems in traditional multiple
regression. In multiple regression it is assumed
that the relationship to be modelled holds
everywhere in the study area. However, this is
often violated in the analysis of house prices
due to the presence of spatial effects.
Spatial effects may occur in two different
forms: one is concerned with spatial
dependency, or its weaker expression, spatial
RICS Research l 13
autocorrelation (they are not identical though
they are often used interchangeably in the
literature) and the other form is spatial
heterogeneity, namely spatial non-stationarity
(Anselin 1999). Spatial autocorrelation can be
seen as spatial interaction whilst spatial
heterogeneity (spatial non-stationarity) refers
to spatial structure (Anselin 1999). Spatial
dependency and spatial non-stationarity have
been the major challenges in spatial data
analysis (Fotheringham et al. 2002).
Figure 1: A spatial kernel
Source: Fotheringham et al. (2002), Geographically Weighted Regression
14 l RICS Research
In the GWR methodology, adaptive spatial
kernels (see Figure 1 and 2) are used in the
weighting process. These work by capturing
the data points to be regressed by moving the
regression point across the region. In an area
where data is dense, spatial kernels with a
narrow bandwidth are in use, whilst in an area
where data are sparse, spatial kernels with a
wide bandwidth are utilised. This approach
uses a bi-square function where the weight of
the jth data point at regression point i is
calculated by:
wij = exp[-(dij /b)2] ≤ when dij ≤ b
wij = 0 when dij > b
(4)
Figure 2: GWR with adaptive spatial kernels
Source: Fotheringham et al. (2002), Geographically Weighted Regression
The Akaike Information Criterion (AIC) (Akaike
1973) is the method used in this report for
evaluating the explanation given by GWR
models. AIC takes account of ‘goodness of fit’
2
(as does R in a regression or hedonic model)
as well as accounting for the complexity of the
model in question. In terms of interpretation,
the better model has a smaller AIC value.
As a rule of thumb, two models are considered
significantly different when their AIC differs
by more than 3 (Fotheringham et al. 2002).
The GWR software is programmed to choose
the optimal bandwidths by seeking to minimise
the AIC.
www.rics.org/research
In summary, therefore, the methodology
employed in this study is to calibrate a
hedonic model or global regression and then
to produce local estimations which take
account of spatial effects, elegantly treated
in the GWR methodology by explicitly
incorporating the locations of data into
calibration. In terms of software, the results
have been produced using the GWR 3.0
software with the aid of Geographical
Information System (GIS) and this produces
first of all the global regression (GR) followed
by local estimates which are plotted in map
form using GIS.
RICS Research l 15
4 The Case Study
T
yne and Wear is a Metropolitan
Region located in the North East
of England. It comprises of the City of
Newcastle upon Tyne, the City of Sunderland,
the Borough of Gateshead, the Borough of
North Tyneside and the Borough of South
Tyneside, plus hinterland stretching into
County Durham and Northumberland.
The population is 1.08 million or 463,000
households, accounting for approximately
43% of the population in the North East
region.
The Tyne and Wear Metro, operated by Nexus,
is the backbone of the area’s public transport
system. Metro was the UK’s first light rail
transit developed from the heavy track,
Figure 3: Map of Tyne and Wear Metro
16 l RICS Research
carrying 37 million passengers annually and
readily accessible to around 25% of
households in Tyne and Wear. The map of Tyne
and Wear Metro is exhibited as Figure 3.
It can be seen that for a large part of the area,
with the exception of Gateshead, the Metro
does provide significant coverage.
The Metro system was planned over 30 years
ago and whilst much of the system was built
in the 1970s, many key structures date from
the Victorian age. The first 55.5km of the
network was completed in 1984, followed by
the 1.5km Newcastle International Airport
extension opening in 1991. An 18.5km
extension to the South Tyneside and
Sunderland areas was opened in March 2002,
Source: based on Tyne and Wear
Accessibility Modelling and Edina Digimap
running over the existing heavy rail route and
on the alignment of the former railway line.

The network is now 78 km long with 58

stations as compared to the original system
with 26 stations. Special features include the
mix of track conversion from heavy rail and
new build together with new major bridges
and 14 km of new tunnels.

The primary data for this study was collected

in 2004, twenty years after the opening of the
Tyne and Wear Metro. Data was collected for
the whole of the Tyne and Wear area and
therefore included areas which had benefited
from long exposure to Metro, areas, such as
Sunderland, where the Metro was relatively
new and areas with no access to Metro.

www.rics.org/research RICS Research l 17

5 Data Collection
Property (House) Data
Transaction property data are normally
considered as ideal data for property related
analysis on the ground that these prices are
the agreed and accepted prices by the market.
In contrast, asking prices can be seen as
expected prices which are valued by agencies.
As a result, they are able to reflect market
valuation although there may be some
unrealistic extreme cases. At the beginning of
this study, transaction property data, including
commercial rents and residential sales, were
explored but they are either unavailable in
England due to confidentiality issues or
available only with limited information about
the property characteristics. Therefore, asking
price data for residential properties, which are
open to the market and do not suffer from
confidentiality, are considered as the data set
for this study available.
In many cases asking prices are reasonable
prices since there is evidence that asking
18 l RICS Research
house price and transaction house prices are
highly correlated with the actual sales price
achieved being above 93% on average of
asking price in the UK housing market since
2002 (Hometrack 2005). In May 2004, when
the data for this study were collected, the
transaction house price achieved was, on
average, 98.6% of the asking price in the
North Region (Hometrack 2005). It is
therefore possible to examine the
determinants of house prices by looking at
asking prices without introducing too much
inaccuracy. In fact, asking price may be more
appropriate than transaction data for the
purpose of examining the effect of external
factors, such as transport accessibility, since
transaction data may vary more due to some
internal factors such as decoration style,
garden or parking space whilst asking prices
can be seen more generalised in terms of
internal factors.
At the time of data collection, a website
www.icnewcastle.co.uk provided a service
called ‘icproperty’ which advertised properties
for sale in Tyne and Wear Region with
sufficient information. This allowed the
collection of data on number of bedrooms and
house types. The data for this study has been
collected at the postcode district (e.g. NE1)
but with full postcode units (e.g. NE1 7RU) for
each property advertised in May 2004. The
internet data source provided 2855 records of
asking house price with the information of the
type of house and the number of bedrooms
for 40 postcode districts covering the whole
Tyne and Wear Region. Based on the full
postcode, these records then were allocated
with Output Areas (OAs) and the coordinates
of the centroids of OAs concerned.
Neighbourhood Environment
Data
Neighbourhood environment data including
social economic data, such as household
status in relation to household income and
environment data, for example, the schooling
environment, is required to try and explain the
external characteristics of house prices. The
purpose here is to try and standardise for all
characteristics so that the effect of transport
accessibility can be observed.
In this study, household status is captured by
‘Higher managerial and professional
occupations’ and ‘Long term unemployed’
which are found in one of the widely-used
standard socio-economic classifications in the
UK, and available in census data format.
Proximity to good schools has been identified
as one of the key factors to determine the
choice in location of houses in Tyne and Wear
(TRRL 1984).
This is confirmed by other empirical studies
(Cheshire and Sheppard 2004; Gibbons and
Machin 2005; Reback 2005) although
whether it is primary school quality or
secondary school quality which adds more to
house price seems to vary from city to city
(Cheshire and Sheppard 2004). To reflect the
quality of school amenities this study uses the
average point score of secondary schools, as a
result of comparison between primary and
secondary schools, in the school performance
league table published annually by the
Department for Education and Skills (DfES
2004).
www.rics.org/research
Transport Accessibility Data
The Tyne and Wear Plan Partners (TWPP)
commissioned the development of the Tyne
and Wear Transport Accessibility Model
(TWTAM) as a result of the Local Transport
Plan requirement (TWPP 2003). At the time of
this study, this model produced travel time as
an accessibility measurement. These were
based on either a ‘closest to facility’ method or
a weighted Hansen method for public
transport (hourly between 0700 and 2300)
and for car travel at different road states
(capacity speed/half-capacity speed/full
speed) to core services calibrated at one
minute intervals.
The closest method to education by public
transport, for example, is calculated as the
travel time to the nearest school; the
associated weighted Hansen accessibility
measure is more complicated and is calculated
by reference to a gravity-based formulation as
follows:
j j
ti = ln{Σ Aj exp(-λ tij)/Σ Aj }/(-λ) (5)
j=1 j=1
where ti is travel time in zone i (OAi); j indexes
available destination zones reachable from
zone i; Aj is the number of jobs at zone j
accessible to large employers; λ is a
deterrence parameter consistent the trip
distribution and tij gives overall travel time
from zone i to zone j (both journey time and
walking to/from bus stops).
RICS Research l 19
It is worth noting that at the time of this study,
public transport accessibility in the TWTAM is
based on Metro and bus combining to form a
single public transport network, making it
impossible to separate Metro from bus
accessibility. As Metro is a significant transport
facility in only part of the case-study Region,
the effect of metro accessibility is separately
investigated by the use of a group of dummy
variables for the thresholds of distance to the
nearest Metro station.

Table 1: Summary of data collection

Vector Variable Description Source

House data P House price (dependent variable) Internet
(H) FLAT Type of flat (1=yes; 0=no) Internet
TERR Type of terraced (1=yes; 0=no) Internet
SEMI Type of semi-detached (1=yes; 0=no) Internet
DETA Type of detached (1=yes; 0=no) Internet
FLATBED FLATBED = FLAT * BEDROOM Excel
TERRBED TERRBED = TERR * BEDROOM Excel
SEMIBED SEMIBED = SEMI * BEDROOM Excel
DETABED DETABED = DETA * BEDROOM Excel
BEDROOM Number of bedroom Internet

Neighbourhood data
(N) EDU13PT The average point score of the secondary
school in 2003 nearest to the house DfES
%UNEMP % of long term unemployment in OA where
the house is located 2001 census
%HPROF % of higher professional occupations in OA
where the house is located 2001 census

Transport PT08E13 Public transport accessibility to secondary

accessibility data schools at 0800 am TWPP
(T) PT08E16 Public transport accessibility to colleges at 0800 am TWPP
PT08EPR Public transport accessibility to primary
schools at 0800 am TWPP
PT09EMP Public transport accessibility to large
employers at capacity speed TWPP
CARCE13 Car transport accessibility to secondary
schools at capacity speed TWPP
CARCE16 Car transport accessibility to colleges at capacity speed TWPP
CARCEPR Car transport accessibility to primary
schools at capacity speed TWPP
CARCEMP Car transport accessibility to large employers
at capacity speed TWPP
DISTMS1 the house within 200 metres of a Metro station
(1=yes; 0=no) GIS
DISTMS2 the house within 201-500 metres of a Metro station
(1=yes; 0=no) GIS
DISTMS3 the house within 501-1000 metres of a Metro station
(1=yes; 0=no) GIS
DISTMS4 the house beyond 1 kilometre of a Metro station
(1=yes; 0=no) GIS

20 l RICS Research
6 Data Analysis

Correlation Issues may not show up through simple correlation

Multicollinearity is a potential problem in all
regression based analysis and has arisen in
this study. Whilst linear dependence between
independent variables frequently occurs, it is
the degree of linear dependence that can
cause estimation problems since the
consequence of multicollinearity is that
parameter estimates become non-significantly
different from zero although R2 and adjusted
R2, as measures of goodness of fit, remain
unchanged.
A simple check on correlations between
independent variables is a common approach
to examining the likelihood of multicollinearity.
High correlations among independent
variables imply the likeliness of
multicollinearity whilst the converse statement
that low correlations suggest the absence of
multicollinearity may not be true. This is
because, when there are a large number of
independent variables, linear dependencies
because complex interactions may be involved
that are not uncovered in a pairwise
comparison. Instability of parameter estimates
and enlarged standard errors can also be
considered as a potential sign of
multicollinearity (Thomas 1983).
Ad hoc tests such as these are often better
than the more formal tests since many of the
available tests make assumptions which this
study believes unrealistic (for example, the set
of explanatory variables are jointly normally
distributed with zero population covariances
(Thomas 1983)). Thus this study considers
monitoring the presence of multicollinearity by
the basic check of simple correlation between
independent variables and a careful
consideration of the stability of parameter
estimates and their standard errors.

Table 2: Results of highly correlated variables

Variable 1 Variable 2 Sample correlation coefficient

FLAT FLATBED 0.93

TERR TERRBED 0.93
SEMI SEMIBED 0.96
DETA DETABED 0.96
CARCEMP PT09EMP 0.91
CARCE13 PT08E13 0.86
CARCE16 PT08E16 0.96
CARCEPR PT08EPR 0.95

www.rics.org/research RICS Research l 21

The correlation analysis has been undertaken
using Microsoft Excel software for all the
independent variables considered in the
process of analysis. Correlation matrix is large
but those pairs of variables having a
coefficient greater than 0.75 are shown in
Table 2 above. The nature of the interaction
terms ***BED (for example FLATBED and
TERRBED) yield high correlations with the
corresponding type variables
(FLAT/TERR/SEMI/DETA).
The pairs of car and public transport
accessibility to the same service are not
surprisingly highly correlated since the
accessibility model used to provide data in this
study calculates car accessibility to a particular
service using the network of public transport
accessibility to this particular service. Hence,
extra care needs to be taken with potential
multicollinearity between the variables shown
in this Table.
One option with highly collinear variables is to
exclude one of the variables from the
calibration of the model. When this is done,
the interpretation requires care for example, if
CARCEMP is selected as the appropriate
variable in the final equation and it is highly
correlated with PT09EMP which is excluded
from the calibration process, then the
interpretation of the results for CARCEMP is
the effect of both CARCEMP and PT09EMP.
i.e. either CARCEMP or PT09EMP could
represent the transport accessibility to large
employment regardless by car and by public
transport.
The Best Equation
As GWR is an extension of the global
regression (GR) model, it is essential to
establish the best equation for the GR model.
As it is not necessarily the case that the best
GR model leads to the best GWR model, this
study has sought to calibrate the best GR
model consistent with sensible GWR results.
Employing a stepwise approach to select the
appropriate transport accessibility variables,
the best equation in this study is identified
below as Equation (6):
Pi = α0 + α1FLATi + α2TERRi + α3DETAi +
α4FLATBEDi + α5TERRBEDi + α6DETABEDi +
α7BEDROOMi + α8EDU13PTi + α9%HPROFi +
α10%UNEMi + α11PT08E13i + α12PT08E16i +
α13CARCEPRi + α14CARCEMPi + α15DISTMS1i +
α16 DISTMS2i + α17DISTMS3i
(6)
The description of the variables can be found
in Table 1 above. FLATBED, TERRBED, and
DETABED are a group of interaction terms
obtained by multiplying the dummy variables,
the type of houses, and the number of
bedrooms. The reason for introducing this
group of interaction terms is because there is
joint effect of the type of houses and the
number of bedrooms on the house price over
and above their separate effects. i.e. the price
of one bedroom relates to what type of house.
e.g. on average, one additional bedroom of a
detached house would worth more than one
additional bedroom of a flat. The introduction
of the interaction terms for these variables can

22 l RICS Research
provide an accurate estimation of the true
relationship between house price and types of
house and numbers of bedrooms and can also
explain more of the variation in the dependent
variable. As these are a set of dummy
variables, SEMIBED is omitted in the
calibration and so the interpretation of the
results for FLATBED, TERRBED and
DETABED are relative to this omitted variable.
GR Results
For the GR model, the estimated parameters
for variables DETABED, BEDROOM,
EDU13PT, %HPROF, DISTMS1, DISTMS2
and DISTMS3 are expected to be positive (α6,
α7, α8, α9, α15, α16, α17>0) and FLATBED,
TERRBED, %UNEM, PT08E13, PT08E16,
PT08EPR and CARCEMP are expected to be
negative (α4, α5, α10, α11, α12, α13, α14<0)
as positive/negative relationships of the house
prices and these variables are expected.

Table 3: Results of GR parameters

Parameter Estimate Percentage of estimate/mean price t- value

INTERCEPT -28625.173 N/A -2.823 **

FLAT 16788.900 N/A 1.618 non-sig.
TERR -22586.883 N/A -2.376 **
DETA 3394.188 N/A 0.289 non-sig.
FLATBED -17454.151 N/A -4.407 **
TERRBED 1236.163 N/A 0.396 non-sig.
DETABED 13306.508 N/A 3.714 **
BEDROOM 35259.560 22.05% 13.859 **
EDU13PT 761.203 0.48% 4.135 **
%HPROF 4905.114 3.07% 26.251 **
%UNEM -4339.039 -2.71% -5.446 **
PT08ED13 -1059.150 -0.66% -3.154 **
PT08ED16 -609.663 -0.38% -3.011 **
CARCEDPR 2454.737 1.53% 4.987 non-sig.
CARCEMP 1217.490 0.76% 6.351 non-sig.
DISTMS1 27396.385 17.13% 4.681 **
DISTMS2 14193.270 8.87% 4.577 **
DISTMS3 17598.439 11.00% 6.777 **

** = significant at 1% level for one tailed tests

www.rics.org/research RICS Research l 23

The results of GR parameters are shown in
Table 3 which contains the name of the
variable whose parameter is being estimated,
the estimate of the parameter, and the t
statistic for the null hypothesis H0: α= 0
against an alternative hypothesis of either H1:
α>0 or α<0 as explained above and the level
of significance for rejecting H0 in a one tailed
test. The interpretation here, for the example
of the average performance points of the
closest secondary school (EDU13PT), is that
an increase in one point will lead to £761
(0.48%) increase in house price on average,
holding everything else constant. The t-value is
4.135 demonstrating that this GR parameter,
EDU13PT, is greater than zero at a 1% level
of significance.
In this GR model, the internal factors are
considered to combine the types of house
with number of bedrooms so the results are
the price per bedroom for each type of house.
As the variable SEMIBED was dropped, the
estimate for BEDROOM, α4, in fact
represents the price for one bedroom of semi-
detached house. That is, one additional
bedroom of semi-detached house adds
£35259 (22.05%) to the house price.
24 l RICS Research
The estimate for FLATBED gives the value for
one bedroom of flat as compared to a semi-
detached house and similarly for a terraced or
detached house when considering TERRBED
or DETABED. The combined effect of one
additional bedroom of flat/ terraced/ detached
is then given by adding α7 to α4 /α5 /α6
so that one additional bedroom in a
flat/terraced/detached property is worth
£17805/ £36496/ £48566 respectively,
which takes up 11.13%/ 22.82%/ 30.37% of
the mean price respectively. It can be seen, as
expected, that the internal features of houses,
which are the main contributors to ‘improved
land’, give the principal explanation of house
price.
However, the focus of this study is on the
external factors which determine the land
value rather than improved land value, and on
transport accessibility in particular even
though these factors play a more minor role in
the determination of the house price as
compared to the internal features. Some
results, such as for EDU13PT, confirm the
expectation of the parameters mentioned
above. However, some results of the GR
parameters (TERRBED, PT08EPR and
CARCEMP) are the reverse to the expectation
above where they are significant. To see if the
results confirm the expectations mentioned
above or not, Table 4 displays a comparison
between outcomes and expectations. It shows
the alternative hypothesis, each parameter (as
discussed above), the sign of the estimated
parameter and, where the parameter is
significantly different from zero, whether the
outcome matches a priori expectations. The
variables where the outcome does not match
expectations are in bold.
Table 4: Outcomes and expectations

Variables H1 Estimated Outcome matches

parameters ( i) expectations
(agree √, disagree x )

FLATBED α4<0 - √
TERRBED# α5<0 + x
DETABED α6>0 + √
BEDROOM α7>0 + √
EDU13PT α8>0 + √
%HPROF α9>0 + √
%UNEM α10<0 - √
PT08E13 α11<0 - √
PT08E16 α12<0 - √
CARCEPR α13<0 + x
CARCEMP α14<0 + x
DISTMS1 α15>0 + √
DISTMS2 α16>0 + √
DISTMS3 α17>0 + √

# non-sig. at 5% level

For the internal factors, FLATBED and
TERRBED were expected to have less value
than SEMIBED while DETABED was thought to
be more expensive than SEMIBED. From Table
4, it can be seen that the result for TERRBED is
not as expected but, in addition, is non-significant
at 5% level. With respect to the socio-economic
factors, %UNEM is significant in decreasing the
property value in the GR as expected whilst
%HPROF, as another proxy for income, would be
expected to be positively associated with
property value and this is the case.
So we can see that, in the GR model, the
factors of high professional and
unemployment reflecting household status
does, as expected, significantly contribute
positively and negatively to house value
respectively. In addition, having a better school
nearby, as identified by the point score, was
expected to add value and this is true for
EDU13PT variable.
In terms of car and public transport
accessibility, more travel time means worse
accessibility so, the alternative hypotheses,

www.rics.org/research RICS Research l 25

H1, for these parameters are expected to be
negative so that one more minute of public
transport/car travel time (worse public
transport accessibility to schools and car
accessibility to larger employers) would lead to
lower house prices as the underlying premise
is that, everything else being held constant,
better accessibility would increase house
value. Since car accessibility and public
transport accessibility are highly correlated,
either car accessibility or public transport
accessibility to one service, such as secondary
schools, can be regarded as general transport
accessibility to that service as discussed in the
section above on correlation issues at the
beginning of this section.
Whilst the variables PT08E13 and PT08E16
lift house value as expected and this is
significant at the 1% level, the variables
PT08EPR and CARCEMP reduce house
value and are significantly negative in contrast
to expectation. These results tell us that,
globally, a house with better accessibility (one
minute less) to secondary schools and
colleges can add £1059 (0.66%)/ £610
(0.38%) to house value whereas, one minute
closer to primary schools and larger employers
will reduce £2455 (0.76%) /£1217 (0.76%)
house value.
Unlike some of the variables for car and public
transport accessibility, the results for variables
for Metro access all agreed with expectations
although DISTMS3 (501m-1000m) gives a
slightly higher value than DISTMS2 (201m-
500m) which seems somewhat perverse.
DISTMS1 is significant at the 1% level,
suggesting that a house within 200m of a
Metro station is worth £27396 (17.13%) more
26 l RICS Research
than a house located more than 1km away
from a Metro station and this is very close to
the average estimation of a 20% premium for
the houses in proximity to Metro stations as
compared to houses further away made by
estate agents in Newcastle upon Tyne
(Hass-Klau et al. 2004). DISTMS2/3 gives
£14193 (8.87%)/ £17598 (11.00%) positive
premiums to a house within 201m-
500m/501m-1000m compared to a
house located beyond 1km away from a
Metro station.
It should be remembered that the global
results are ‘average’ results and that it is
possible that averaging effects over the Tyne
and Wear study area could lead to unexpected
results as identified by the example in the
section on GWR above. The focus of this work
is to examine the contribution that a local
model approach can add and the results from
GWR are considered next.
GWR results
The GWR software provides diagnostic
information for the GWR model including an
ANOVA which tests the null hypothesis that
the GWR model has no improvement over a
global model. These are shown in Table 5
where the F test suggests that this GWR
model has a significant improvement over the
global model in this study reported above. In
2
addition, the adjusted R has increased from
0.615 in the global model to 0.728 in the
GWR model implying that the GWR model
gives a better explanation, after taking into
account degrees of freedom.
Table 5: ANOVA results
Source Sum of Squares Degrees of freedom
OLS residuals 7534825890220.1 18.00
GWR improvement 2706671403008.0 265.16
GWR residuals 4828154408116.5 2553.84
As identified above, GWR gives the ability to
examine spatial variability which can be hidden
in a GR model. All the local parameter
estimates can be mapped but this study
concentrates on transport accessibility
variables – PT08E13, PT08E16, PT08EPR,
CARCEMP and DISTMS1-3. The t- values
and parameter estimates for these variables
are mapped in Figure 4 to Figure 7 by inverse
distance weighted (IDW) interpolation with
GIS. The best interpretation comes from
looking at maps of local parameter estimates
alongside the maps of local t- value since the
local t- values maps exhibit the significance
associated with the local parameter estimates.
In line with the discussion of the global results
above, the local estimates are presented in
terms of premiums (the percentage of
parameter estimate over observed house
price) and these are presented on the same
map as the t-values to give values and their
significance together as a series of coloured
symbols.
www.rics.org/research
MS F value
10207867309.4160
1890543558.0949 5.399
sig. at 1% level
The maps are presented in pairs. In the first of
each set, the house price premiums are set
alongside the significance of these values in
the local model. To interpret the maps, it is
important to remember that a t-value having
an absolute value of approximately 2 or higher
suggests, in a large data set, that an estimator
is significantly different from zero. In the local
model, two tailed tests are carried out, in
contrast to the global model where one tailed
tests were used, since GWR is looking at each
observation separately with no a priori
knowledge as to whether the spatial effects
would have a positive or negative effect. So, in
the discussion of PT08E13, PT08E16,
CARCEPR and CARCEMP, the t- value is
classified by four bands: t>2.00, 2.00>t>0,
0>t>-2.00 and t<-2.00. The percentage of
parameter estimates is classified in five bands,
choosing 5% as the benchmark to show
visually the presence of positive and negative
premiums. In t- value maps, the lightest areas
and darkest areas are significant with the
lightest areas exhibiting positive house
premiums and the darkest negative house
RICS Research l 27
premiums. In each map, the value of the global
value, as percentage of parameter estimate, is
set as single band to highlight the difference
between global parameters and local
parameters. The first Figure will be interpreted
in detail to illustrate these points.
In the maps of parameter estimates, bands
have been set up to illustrate the parameter
estimates in more detail. The yardstick used to
determine the bands is based on giving results
which provide clear and sufficient information
on parameter estimates, taking account of the
global value for the particular parameter
estimate. In the following subsections, the
results for each transport accessibility variable
are demonstrated and interpreted in detail.
As mentioned above, because of the high
correlation between each pair of transport
accessibility by car and by public transport, no
matter the variables of public transport
accessibility or car accessibility are selected
for the best equation, the results from these
variables should represent the results from
transport accessibility covering both public
transport and car. For example, the results for
the first pair of maps concerning the variable
PT08E13 should be interpreted as the result
of transport accessibility to secondary schools
rather than the single public transport
accessibility to secondary schools.
In the interpretation of the maps which follow,
specific areas of Tyne and Wear are identified
and these are identified on Figure 3 (above in
Section 4).
28 l RICS Research
Transport accessibility to
secondary schools (PT08E13)
Figure 4 shows the map of t- values and
parameter estimates associated with
PT08E13. As shown in Legend, the lightest
areas are significant negative t- values while
the darkest areas are significant positive t-
values. The areas in light and dark grey are
respectively non-significant negative and
positive t-values. This therefore shows the
geographical areas where the local
parameters are significantly different from zero
and need to be considered alongside the
coloured dots which represent the local
parameter as a percentage of the global
estimate, which are classified by five bands of
< -5% in red, -5% to -0.66% (which includes
the global value) in yellow, -0.66% to 0 in light
blue and > 5% in dark blue. As explained
above, negative parameter estimates should
be interpreted as positive premiums for house
prices while positive parameter should be
interpreted as negative premiums.
Figure 5 shows the map of parameter
estimates associated with PT08E13.
Parameter estimates have been classified by 9
bands of £2000 between the values of
£10000 and - £10000 with a single band of
£1000 to examine the global value (shown in
Table 3). The greatest negative values of
parameter estimates (positive premiums) are
in red and the smallest positive values
(negative premiums) are in dark blue.
It is clear from Figure 4 and Figure 5 that both
the t-values and estimated parameters
illustrate considerable spatial variation.
 Although the GR suggests that better Figure 4 above). As can be seen from Figure
transport accessibility to secondary schools 5, the value of global parameter (£1059) falls
leads to a globally statistically significant into the blue green band which only covers
house price premiums (£1059 per minute at limited and areas, which, according to Figure
0.66% of the mean house price), Figure 4 4, are non significant. This means that the
shows that, in most of the Region, the two global average value is not indicative of the
variables appear to be unrelated shown as local value for most houses.
most of the map being light and dark grey
suggesting t-values of less than an absolute As with PT08E13 above, apparent spatial
value of 2 and a lack of significantly different variation can be seen in Figure 6 and Figure 7.
from zero local parameters for this variable. In the GR model, the relationship between
Only three major areas (the west end of the house price and transport accessibility to
Tyne and Wear Region, Newcastle central area further education, colleges, is positive and
and West Monkseaton in North Tyneside to
the east of the study areas and two minor
areas south of the River Tyne (one in
Gateshead and one in Sunderland)) emerge
with this relationship greater than the GR
value and up to 46.85% (red symbols). This
suggests that areas in the vicinity of a Metro
station, transport accessibility seems to be
positively capitalised in relation to the Metro
access to secondary schools for pupils whilst,
in the other areas, bus or car access to
secondary schools must be associated with
the positive premiums.

“ ...transport accessibility to
secondary schools gives “
positive premiums...
Figure 5 demonstrates that transport
accessibility to secondary schools gives
positive premiums up to £13377/minute for
majority of the houses in Tyne and Wear, albeit
non significant for most areas (according to

www.rics.org/research RICS Research l 29

Figure 4: Map of t-values and percentage of parameter estimates
associated with variable PT08E13

Figure 5: Map of parameter estimates associated with variable

PT08E13 Transport accessibility to colleges (PT08E16)

30 l RICS Research
statistically significant. However, Figure 6
shows this is not true for most houses in Tyne
and Wear with the exception of two areas to
the south east of the study area in Sunderland
which are showing up to 8.81% positive
premiums. In contrast, for houses in a major
area of Gateshead and Newcastle, there
appears to be a negative relationship between
house price and PT08E16 with negative
premiums of up to a surprising 50.8%. The
reason for such disparity might relate to the
neighbourhood environments that these
colleges serve.
In Figure 7, areas in orange tending towards
red exhibits positive premiums that transport
accessibility to colleges adds towards house
Figure 6: Map of t-values and percentage of parameter estimates
associated with variable PT08E16
www.rics.org/research
prices whilst areas in yellow tending through
to blue illustrate negative premiums. Taken
together with Figure 6, this means that some
houses in Sunderland, where the significant
positive relationship exists and shown in
orange and red, receive £1000-
£4089/minute positive house price premiums
from their accessibility to colleges.
Whereas, houses in the areas of Gateshead
and Newcastle, where the significant negative
relationship lies and shown in lime tending to
blue, are devalued by their accessibility to
colleges from £2000-£12477/minute. It is
notable that, from Figure 7, the highest
RICS Research l 31
Figure 7: Map of parameter estimates associated with variable
PT08E16

parameter estimates (red shading) can be
found in the north of Tyne and Wear where the
t- value appears non-significant (pale or mid
grey) in Figure 6.
Transport accessibility to primary
negative effect of primary schools varies or
households value such negative effect
differently. As the distance to primary schools
is expected to be within walking distance,
accessibility by car or public transport seems
unimportant for most households.

schools (CARCEPR)
In the areas in black, being close to primary
schools decreases house price by up to
Global results show transport accessibility to
46.42% suggesting the households in these
primary schools reduces house prices by
areas value the negative effect of being close
£2454/minute (1.53% of mean house price)
to primary schools highly. Significant positive
and this might relate to some negative effect
premiums for house prices that can be found
of primary schools which is unknown to this
in one area in the south of the Region needs
study. However, GWR results shown in Figure
further investigation.
8 demonstrate that this is not the case for the
majority of houses in Tyne and Wear apart
Again, Figure 9 illustrates the picture of
from five areas in black, suggesting that the
varying parameter estimates associated with

32 l RICS Research
Figure 8: Map of t-values and percentage of parameter estimates
associated with variable CARCEPR

Figure 9: Map of parameter estimates associated with variable

CARCEPR

www.rics.org/research RICS Research l 33

transport accessibility to primary schools
although the t-values shown in Figure 8
indicate they are non-significant in most areas.
In the five areas covered by light blue and blue
in Figure 9, where parameter estimates are
giving £3000-£6702/minute negative
premium. Nevertheless, there is one area
where significant positive premiums in house
prices can be found which fall into the £4000-
£5165 band.
house price than better access to employment.
The northeast area with a high positive
relationship is thought partly to be the result
of proximity to a seaside amenity and partly to
be caused by the attractiveness of Metro as a
means of access to employment. However,
this would not appear to be the reason behind
negative premiums in the southeast area
which requires further study with perhaps
more detailed socio-economic information.

Transport accessibility to large The non-stationarity of parameter estimates

employers (CARCEMP) associated with CARCEMP is clearly revealed
in Figure 11 with 13 bands that have been
The results from the GR are interpreted as
set, spanning from £12941 (showing the
better transport accessibility to large
highest negative premiums) to -£16123
employers reduces house price but this is a
(showing the lowest positive premiums).
global average figure. Figure 10 shows that in
Amongst the three areas with significant
the local GWR estimation, there are three
positive premiums shown in Figure 10, the
areas where a one minute quicker car or
highest premiums of £16123 can be found in
public transport access to large employers
the central area of Tyne and Wear (in red) with
adds a premium of up to 29.81% of house
positive premiums of £4000-£6000 in the
price. In particular, a large area in the centre of
other two areas (in golden). Whereas, amongst
Tyne and Wear where the positive premiums
those areas displaying the greatest negative
(shown by negative significance) confirms the
premiums (£12941) can be seen in the
latest trend of gentrification in the UK (Hass-
northeast area (in blue) and less negative
Klau et al. 2004). It also can be seen clearly
premiums of £4000-£8000 are found in the
that the positive premiums appear to decay
southeast area (in light blue). The western
with distance from large employers, particularly
area has been covered by bands with much
the centre of Tyne and Wear, in line with the
more modest but statistically significant
classical Bid-rent land theory (Alonso 1964;
parameter estimates (£1000-£6000).
Muth 1969).

The west area which shows negative
premiums on property prices (shown by the
positive relationship between transport
accessibility to large employers) is
hypothesised to have other stronger
environment feature, such as nice countryside
landscape, which contributes more strongly to
Houses within 200 metres of a
Metro station (DISTMS1)
In the next pairs of maps, the interpretation
turns to look at the effect of proximity of a
Metro station on the house price. The t-values
for these maps have been differently classified
with the insertion of an additional band for a

34 l RICS Research
Figure 10: Map of t-values and percentage of parameter estimates
associated with variable CARCEMP

Figure 11: Map of parameter estimates associated with variable

CARCEMP

www.rics.org/research RICS Research l 35

t-value between 0 and 1 giving five bands
overall. For all the variables based on distance
from a Metro station (DISTMS***), values
between zero and one are problematic as
these variables are dummy variables taking
the value of 0 or 1 itself. The additional band
of 0 to1 band of t-values thus highlights
where the value of the dummy variable itself is
zero reflecting those properties which are
some distance from the Metro.
These dummy variables have been entered
into the global model to reflect different
thresholds of closeness to a Metro station.
Again for interpretation, a pair of maps is
provided for each variable. In the first map for
each variable, t-values are shown as indicated
above with the local parameter estimates
being shown as a percentage of observed
house price as a coloured dot symbol. In the
second map, the parameter estimates for
houses falling into the relevant threshold are
mapped as coloured dot symbols along with t-
values for all the observations in the data set
so as to have better idea of parameter
estimates. The dot symbols for parameter
estimates in the second maps are slightly
bigger than the first maps for percentage of
parameter estimates. As before, the
interpretation of the first of these variables is
identified in depth.
In contrast to the global value of £27396
(17.13% of mean house price), the premium of
being within 200 metres of a Metro station
catchment area from the GR, Figure 12 and
36 l RICS Research
Figure 13 show that statistically significant
positive premiums do not occur in most areas
of the Region (only small areas of white).
This suggests the global model may have
overestimated the value associated with Metro
access for most houses. The mid grey colour
captures where the t- values are 0 to the
south and the southwest of Tyne and Wear,
and this indicates where the local regression
is problematic because the value of the
dummy variables are all zero (in these areas all
houses are some distance from Metro). Apart
from this, two areas exhibit significant positive
premiums. One is to the west of Newcastle
city centre and the other is in the northeast of
the Region: in both these cases, being within
200 metres of a Metro station raises the
house price by between 5% and 31.37% (see
Figure 12), adding £30000-£47267 of a
positive premium shown by blue symbols on
Figure 13. It should be noted that there are no
significant negative effects found for any
house within 200 metres of a Metro station.
“...there are no significant
negative effects found for
any house within 200
metres of a Metro station...
“
Figure 12: Map of t-values and percentage of parameter estimates
associated with variable DISTMS1

Figure 13: Map of t-values and parameter estimates associated with

variable DISTMS1

www.rics.org/research RICS Research l 37

Houses within 201-500 metres of
a Metro station (DISTMS2)
Globally, houses within 201-500 metres of a
Metro station gain £14193 (8.87% of the
mean house price) in house price premiums
as compared to properties 1 kilometre away
from a Metro station. Once again, Figure 14
shows that this is the case only in the
northeast of the Region, where being within
200-500 metres proximity to a Metro station
is capitalised in a positive premium of between
£20000-£52092 (see Figure 15), taking up
between 5% to 50.09% of house price (see
Figure 14).
Figure 14: Map of t-values and percentage of parameter estimates
associated with variable DISTMS2
38 l RICS Research
Despite being largely non significant, there is
evidence of considerable positive premiums in
the Newcastle city centre area. However,
significant negative premiums, shown by red
dots in Figure 14 and Figure 15 are found in
three areas, where being 200-500 metres
proximity to a Metro station gives the negative
premiums of £10000-£32056 (see Figure
15), taking up from 5% to 30.38% of house
price (see Figure 14).
It is notable that, unlike DISTMS1, a large
number of houses within 200-500 metres of a
Metro station have negative premiums,
suggesting that being located between 200-
500 metres to a Metro station is not valued
positively by the households in these areas.
Positive premiums that are statistically
significant (shown by light blue or blue dots in
Figure 14 and Figure15) can be seen to the
south east of the study area, which coincide
with the South Hylton area and were newly
served by the SME. Here, households seem to
value Metro accessibility higher than
elsewhere in Sunderland, perhaps because
South Hylton is more distant from the
Sunderland city centre as compared to other
areas newly served by the SME.
Figure 15: Houses within 501-1000 metres of a Metro station
(DISTMS3)
www.rics.org/research
In the GR model, the parameter estimate for
DISTMS3 (being between 501-1000 metres
of a Metro station) is £17598 (11.00% of
mean house price), which is surprisingly higher
than DISTMS2 (representing being between
200-500 metres of a Metro station).
Using the GWR results and mapping the t-
values associated with DISTMS3 (Figure 16
and Figure 17) shows that there are only two
areas, where this proximity has a significant
positive effect on house price with premiums
of between £20000-£43977 (see Figure 17),
taking up between 5% and 50.46% of the
house price (see Figure 16).
RICS Research l 39
Similarly to DISTMS1, one area is to the west
of the Newcastle city centre and the other is
the northeast of the Region.

Also similarly to DISTMS2, two areas – one in

South Tyneside and the other to the north of
Regent Centre (just north of Newcastle City
Centre) exhibit statistically significant negative
premiums of between £10000-£23630
shown by red dots (see Figure 17), which is
between 5% and 42.71% of the house price
(see Figure 16).

Figure 16: Map of t- values and percentage of parameter estimates

associated with variable DISTMS3

40 l RICS Research
It is worth mentioning that, as can be seen

“
from Figure 16 and Figure 17, there are more
positive and higher house price premiums
around the stations of Tyne and Wear Metro
which have been in service for decades as ...any effect of Metro
compared to the new stations of SME which
had been opened just for two years at the
on land value takes
time of data collection for this study.

This suggests that any effect of Metro on land

value takes rather a long time to occur.
rather a long time
to occur...
“
Figure 17: Map of t- values and parameter estimates associated with
variable DISTMS3

www.rics.org/research RICS Research l 41

7 Discussion and conclusion

s raised at the beginning of this The primary motivation for this study was to

A paper, the fundamental question this

study set out to answer is “to what
extent can transport infrastructure uplift land
value?”
This empirical work suggests that the local
model approach with GWR is more successful
than the GR in examining the relationship
between transport accessibility and land value.
From the results and analysis above, it can be
seen that the relationship between transport
accessibility and land value is fairly
provide evidence for potential implementation
of a Land Value Capture (LVC) policy to
provide a funding mechanism for the provision
of new transport infrastructure. The GR which
gives an average result for the examined
relationship might be useful for setting
specific values in a LVC policy whereas the
GWR outcomes which illustrate the
relationship at the local detail would suggest
how that policy might impact in terms of
winners and losers.

“
complicated and varies greatly over physical
space. The GR model offers the basis for
explaining average variation in house prices
with the additional results from GWR clearly
revealing a spatially varying relationship
...The results
between house price and transport from GR model
accessibility variables.
shows strong
The local model, GWR, gives a statistically
significant improvement over the global model
2
evidence that
by increasing adjusted R even after
accounting for degrees of freedom. The proximity to
results from GR model shows strong evidence
that proximity to Metro stations can uplift the Metro stations
house price significantly and better transport
accessibility to secondary schools also can can uplift the “
add significant premiums to a house whereas
being close to large employers does not house price
increase the value of a house as the classical
urban land theory suggests.
significantly...
Taking a closer look using the GWR model, it
is clear that the relationships between
neighbourhood features are not stationary.
This finding that using a local model approach,
GWR, gives greater insight to the relationships
between transport accessibility and land value.

42 l RICS Research
The GR demonstrates that proximity to Metro
stations can statistically significantly increase
house prices and better transport accessibility
to secondary schools also can generate
significant positive premiums to house prices.
This implies that there are possibilities for
capturing land value uplift associated with
Metro services and transport accessibility to
secondary schools.
However the local analysis, provided by GWR
identifies the presence of non-stationarity
between transport accessibility and land value
leading to transport accessibility having a
positive effect on land value in some areas but
in others a negative or zero effect. Although
neighbourhood features may help to explain
such variation, this means the implementation
of a uniform LVC policy would necessarily give
rise to winners and losers. However, a better
understanding of winners and losers may
allow such a LVC policy to be a more
progressive if ‘winners’ are in the relatively
poor areas.
The picture for Tyne and Wear is complicated
by the way in which other external factors may
play a more important role than transport
accessibility to employment in determining
land value. This suggests that a policy of LVC
should not only apply to transport
infrastructure, but also other important
amenities in order to justify the land value.
This is in line with Land Value Taxation which
charges the site value towards the land
www.rics.org/research
created by the society, including transport
improvements as well as amenities, such as
seaside amenity and green open spaces.
However, the difficulty for a Land Value
Taxation policy is how to evaluate the land
(site) value accurately taking account of all the
different features so as to establish a taxation
mechanism in a justified and transparent way.
Again, as demonstrated in this empirical work,
using GWR is a good way to examine the site
value accurately once a good dataset with
sufficient information can be achieved.
This empirical work clearly demonstrates
positive house price premiums resulting from
good Metro access, although this is not
uniformly distributed. As such it is supportive
of the fact that a policy to recoup the effects
of land value betterment follows new and
better transport infrastructure.
Whilst it is rational to try to capture some or all
of the increased value associated with the
better accessibility offered by Metro, the area
which benefits in this way needs to be
carefully identified.
In this case study, being within 200 metres
proximity to a Metro station seems to be most
beneficial,l but care ought to be taken with
houses that close to Metro stations exhibiting
non-significant negative premiums (see Figure
12 and Figure 13).
RICS Research l 43
As with much research, this study answers
some questions but raises further questions
which can only be resolved by future research.
The major areas that have been seen for
further work have been generated by the
GWR results. GWR, indeed like a microscope,
can bring out previously unimagined detail
suddenly into focus. As a result, the outputs of
the GWR local estimations provided in the
maps have uncovered a great deal of local
information prompting potential work in the
future.

In particular, GWR has been useful in

identifying spatially varying relationship.
This could be developed to produce rational
predictors associated with transport
investment or other amenities on relevant land.

This study has revealed variability over space

in the relationship between transport
accessibility and land value, giving clues to the
interaction between transport infrastructure
and land use, which have long been a key
subject in the transport planning field. For
instance, according to this case study, being
within 200 metres proximity to a Metro station
gives positive premiums for house prices over
the majority of the case study area with the
noticeable exception of one Borough, which
has the highest unemployment rate in Tyne
and Wear Region (TWRI 2001). This suggests
that there might be some interesting link
worthy of further investigation.

44 l RICS Research
References
Adair, A. et al. (2000) House prices and
accessibility: The testing of relationships within
the Belfast urban area, Housing Studies,
15(5), 699-716.
Akaike, H. (1973) Information theory and an
extension of the maximum likelihood principal,
In Geographically Weighted Regression: the
analysis of spatially varying relationships, (Ed,
Charlton, M.) John Wiley and sons, West
Sussex.
Anselin, L. (1999) Spatial Econometrics,
School of Social Sciences, University of Texas
at Dallas.
Armstrong, R. (1994) Impacts of commuter rail
service as reflected in single-family residential
property values, Transportation Research
Record, No. 1466, 88-98.
Atisreal (2005) Property Value Study -
Assessing the Change in Values Attributable
to the Jubilee Line Extension, Transport for
London, London.
Bollinger, C. R. et al. (1998) Spatial variation in
office rents within the Atlanta region, Urban
Studies, 35(7), 1097-1118.
Cervero, R. (1994) Rail transit and joint
development: Land market impacts in
Washington, DC and Atlanta, Journal of the
American Planning Association, 60(1), 83-94.
Cervero, R. and Duncan, M. (2001), Transit's
value-added: Effects of light and commuter rail
services on commercial land values, in 81st
TRB Annual Meeting, Transportation Research
Board, National Research Council, Washington
D.C.
www.rics.org/research
Cervero, R. and Duncan, M. (2002a) Land
value impacts of rail transit services in Los
Angeles county, National Association of
Realtors; Urban Land Institute.
Cervero, R. and Duncan, M. (2002b) Land
value impacts of rail transit services in san
Diego county, National Association of Realtors;
Urban Land Institute.
Cervero, R. and Landis, J. (1993) Assessing
the impacts of urban rail transit on local real
estate markets using quasi-experimental
comparisons, Transportation Research Part A:
Policy and Practice, 27(1), 13-22.
Chen, H. et al. (1997) Measuring the impact of
light rail systems on single family home values:
A hedonic approach with GIS application,
Centre for Urban Studies, College of Urban
and Public Affairs, Portland State University,
http://www.upa.pdx.edu/CUS Accessed
August 2005.
Cheshire, P. and Sheppard, S. (2004)
Capitalising the value of free schools: The
impact of supply characteristics and
uncertainty, Economic Journal, 114(499),
F397-F424.
Chesterton (2002) Second property market
activity final report, Jubilee Line Extension
Impact Study Unit, University of Westminster.
DfES (2004) Performance tables 2003: Key
stage 2 test results, Department for Education
and Skills,
http://www.dfes.gov.uk/performancetables/pri
mary_03/glossary.shtml#q11 (Accessed
March 2005.)
RICS Research l 45
Du, H. and Mulley, C. (2007) The short-term
land value impacts of urban rail transit:
Quantitative evidence from Sunderland, UK,
Land Use Policy, 24(1), 223-233
Hometrack (2005) Key housing market
statistics for England and Wales, UK House
Price Statistics,
http://www.themovechannel.com/sitefeatures/
uk-house-prices (Accessed 2005.)

Forrest, D. and Glen, J. (1995) The impact of a

Landis, J. et al. (1994) Capitalization of transit
light rail system on the structure of house
investments into single-family home prices,
prices, Journal of Transport Economics and
The University of California Transportation
Policy, 30(1), 15-29.
Centre.

Fotheringham, A. S. et al. (2002)

Lawless, P. and Gore, T. (1999) Urban
Geographically Weighted Regression - the
regeneration and transport investment: A case
analysis of spatially varying relationships, John
study of Sheffield 1992-96, Urban Studies,
Wiley & Sons Ltd.
36(3), 527-545.

Gatzlaff, D. H. and Smith, M. T. (1993) The

Nelson, A. C. (1992) Effects of elevated
impact of the Miami Metrorail on the value of
heavy-rail transit stations on house prices with
residences near station locations, Land
respect to neighbourhood income, In Impacts
Economics, 69(1), 54-66.
Of Rail Transit On Property Values, (Ed, Diaz,
R. B.) APTA 1999 Rapid Transit Conference.
Gibbons, S. and Machin, S. (2005) Valuing rail
access using transport innovations, Journal of
Reback, R. (2005) House prices and the
Urban Economics, 57(1), 148-169.
provision of local public services: Capitalization
under school choice programs, Journal of
Hass-Klau, C., et al. (2004) Economic Impact
Urban Economics, 57(2), 275-301.
of Light Rail: The Results of 15 Urban Areas
in France, Germany, UK and North America,
RICS (2004) Land value and public transport,
Environmental and Transport Planning.
Stage two - testing the methodology on the
Corydon Tramlink, RICS, ODPM and DfT, 52.
Hack, J. (2002) The role of transit investment
in urban regeneration and spatial
Riley, D. (2001) Taken for a ride, the Centre
development: A review of research and current
for Land Policy Studies, London.
practice, In Land value and public transport,
Stage one - Summary of findings, RISC and
Sirmans, G. S. et al. (2005) The composition of
ODPM.
hedonic pricing models, Journal of Real Estate
Literature, 13(1).

46 l RICS Research
So, H. M. et al. (1997) Estimating the
influence of transport on house prices:
Evidence from Hong Kong, Journal of
Property Valuation and Investment, 15(1),
40-47.

TRRL (1984) The impact of metro and public

transport integration in Tyne and Wear, SR
478, Transport Road Research Laboratory.

TRRL (1993) The longer term effects of the

Tyne and Wear Metro, CR 357, Transport Road
Research Laboratory.

TWPP (2003) Annual progress report 2002-3,

Tyne and Wear Plan Partners.

Voith, R. (1991) Transportation, sorting and

house value, In Impacts of Rail Transit on
Property Values, (Ed, Diaz, R. B.) APTA 1999
Rapid Transit Conference, 1-8.

Weinberger, R. (2001), Light rail proximity:

Benefit or detriment? - the case of Santa
Clara county, California, in TRB 80th Annual
Meeting, Washington D.C.

Weinstein, B. L. and Clower, T. L. (2002) An

assessment of the dart LRT on taxable
property valuations and transit oriented
development,
http://www.apta.com/info/briefings/dart2002.
(pdf Accessed June 2004.)

www.rics.org/research RICS Research l 47

48 l RICS Research
RICS Research paper series
Submission of papers
The RICS encourages academics and other
researchers to submit work that is relevant to
the discipline of surveying in all its fields.
Authors may submit either by post or by email.
In the first instance, submissions should be
made to the RICS. Postal submissions should
be made in triplicate to:
Stephen Brown
Head of Research
Royal Institution of Chartered Surveyors
12 Great George Street
London SW1P 3AD
UNITED KINGDOM
Email submissions of Word or PDF files
should be sent to sbrown@rics.org
Telephone number for enquiries:
+44 (0)20 7334 3725
The refereeing process is double blind and all
submitted papers will be referred to at least
two reviewers.
The manuscript
There is no fixed limit on the length of papers
and each paper is published as:
 A PDF on the RICS website, with no
charge made for download.
 A stand-alone published document
Papers are placed on the RICS website
immediately that they are completed and
printed copies available shortly thereafter.
While there is no word limit, it is anticipated
that papers would be no less than 5,000
words and generally no more than 30,000
words.
www.rics.org/research
The manuscript must be in English, typed in
double spacing on one side of A4 paper only,
with a 4 cm margin on the left-hand side. The
pages should be numbered consecutively.
There should be no loose agenda or notes or
other explanatory material.
Title page
The first page of the manuscript must contain
the full title: the name(s), affiliation(s),
address(es) of the author(s); a title of not
more than 75 characters and spaces; and five
key words for the purpose of indexing. If there
is more than one author, the corresponding
author should be indicated.
The abstract
The second page should contain the title and
an abstract. It should not contain the name(s)
of the author(s). The abstract should not
exceed 200 words and must be a clear
summary of the contents of the manuscript,
indicating the contribution that the paper
makes to knowledge and providing a clear
statement of the key findings and outcomes of
the paper.
Illustrations
Any illustrations must accompany the
manuscript but should not be included in the
text. Diagrams, charts, photographs and maps,
should be referred to as 'Figure 1', 'Figure 2'
and so on. They should be numbered in the
order in which they are referred to in the text.
Illustrations can be colour or monochrome.
When providing electronic version of
illustrations, please ensure that these are of a
sufficient resolution for reproduction. A
resolution of 300 dots per inch is the
minimum acceptable standard for images.
They will normally be reduced in size on
RICS Research l 49
reproduction and authors should bear this in author have the same year, they should be
mind, particularly when selecting font sizes. differentiated by using 1998a, 1998b and so
Appropriate electronic submission is on. The style should follow the examples
permissible. below:

Proofs Chau, K.W., MacGregor, B.D. and Schwann, G.

Proofs, in the form of a PDF file, will be sent (2001) Price discovery in the Hong Kong real
to the corresponding author for correction and estate market, Journal of Property Research,
approval. The website is used to provide 18(3), 187-216.
ongoing comment and feedback on papers
that are posted on the website, and comments Brown, G.R. and Matysiak, G.A. (2000) Real
that are provided will be passed through to the estate investment - a capital markets
corresponding author. Authors are provided approach, Financial Times Prentice Hall,
with five free copies of their paper. Further Harlow.
copies can bought at a price of £2.00 each.
If no person is named as the author the body
Publicity
should be used, for example:
The aim of the RICS is to ensure that papers
are promoted and publicised to appropriate
Royal Institution of Chartered Surveyors
academic, professional, policy and media
(1994) Understanding the property cycle,
audiences. In pursuit of this, the RICS may
London.
seek to develop and issue supporting material
for papers published, such as press releases Copyright
and summary documents. The RICS will liaise Submission of an article to the RICS Research
with the corresponding author on the drafting Paper Series is taken to imply that it
of this material and on the appropriate degree represents original, unpublished work, not
of involvement of the author in this process. under consideration for publication elsewhere.
When submitting a manuscript, authors will be
References
asked to transfer the copyright for their article
The Harvard system should be used.
to the Royal Institution of Chartered Surveyors,
References in the text should be quoted in the
if and when the article is accepted for
following manner: Jones (1999) or (Edge and
publication. The Royal Institution of Chartered
Moody, 2001) or, if there are more than two
Surveyors will not refuse any reasonable
authors ... Thomas et al. (2002). If there is a
request by the author for permission to
citation of a page number or numbers, the
reproduce any of his or her contributions to
format should, as appropriate, be Smith (1999,
the series in other forms.
20), Smith (1999, 20-5), (Smith, 1999,20) or
(Smith, 1999, 20-5).
Permission to publish illustrations must be
obtained by the author before submission and
References should be collected at the end of
any Acknowledgments should be included in
the paper in alphabetical order by the first
the figure captions.
author's surname. If references to the same

50 l RICS Research
www.rics.org

RICS (Royal Institution of Chartered Surveyors) is the largest

organisation for professionals in property, land, construction and
related environmental issues worldwide. We promote best practice,
regulation and consumer protection to business and the public.
With 130 000 members, RICS is the leading source of property
related knowledge, providing independent, impartial advice to
governments and global organisations.

May 2007/100/RICS Research/38837/Sterling

The Royal Institution

of Chartered Surveyors
12 Great George Street
Parliament Square
London SW1P 3AD
United Kingdom

T +44 (0)20 7334 3725

contactrics@rics.org 100%

www.rics.org

Research