Transport accessibility and land value: a case study of ... · She is a registered transport planner in P.R.China and with ten years of industrial experience in transport planning,

Transport accessibility and land value: a case study of Tyne and Wear

RICS Research paper seriesVolume 7 Number 3 June 2007

Research www.rics.org

Corporate Professional Local

Dr Hongbo Du

Dr Corinne MulleyNewcastle University, UK

Dr Hongbo Du received her PhD degree in transport engineering from theSchool of Civil Engineering and Geosciences at Newcastle University, UK byachieving unique expertise in relatively new transport geography field with theapplication of a cutting edge spatial analysis modelling, Geographically WeightedRegression in transport area. She is a registered transport planner in P.R.Chinaand with ten years of industrial experience in transport planning, her particularresearch interests lie in the policy issues arising from the interaction of land useand transport studies.

Dr Corinne Mulley is a Senior Lecturer in Transport Economics in the TransportOperations 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 includethe evaluation of transport projects, particularly where the impacts of urbantransport policies and investments have an economic perspective, the identificationof risk in transport evaluation; the quality of public transport provision bothinternally 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 qualitypublic transport in small and medium sized cities and Demand ResponsiveTransport Telematics Good Practice Guide, commissioned by the Department for Transport.

Transportaccessibility and land value: a case study ofTyne and Wear

Volume 7 Number 3

June 2007

© RICS

June 2007ISSN 1464-648XISBN 978-1-84219-328-0

Published by:

RICS12 Great George StreetLondon SW1P 3ADUnited Kingdom

The views expressed by the author(s) are notnecessarily those of the RICS nor any bodyconnected with it. Neither the author(s), northe RICS accept any liability arising from theuse of this publication.

This paper can be copied free of charge forteaching and research purposes, provided that:

• the permission of the RICS is sought inadvance

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

Aims and scope of the RICS ResearchPaper Series

The aim of the RICS Research Paper Series isto provide an outlet for the results of researchrelevant to the surveying profession. Papersrange from fundamental research workthrough to innovative practical applications ofnew and interesting ideas. Papers combineacademic rigour with an emphasis on theimplications in practice of the materialpresented. The Series is presented in areadable and lucid style which stimulates theinterest of all the members of the surveyingprofession.

4 l RICS Research

Editorial Team

Stephen BrownHead of Research RICS12 Great George StreetLondon SW1P 3ADUnited Kingdom

E: [email protected]: +44 (0)20 7334 3725

Amy RobertsRICS

E: [email protected]

Panel of referees

Kwame Addae-DapaahNational University of SingaporeSingapore

Akintola AkintoyaGlasgow Caledonian UniversityScotland

Saleh Al-HathloulAlhathloul lilumranKingdom of Saudi Arabia

Adarkwah AntwiUniversity of WolverhamptonEngland

Ghassan AouadUniversity of SalfordEngland

David BaldryUniversity of SalfordEngland

Malcolm BellLeeds Metropolitan UniversityEngland

Roderick BissetFaithful & GouldScotland

Alistair BlythUniversity of WestminsterEngland

Thomas BohnUniversity of LeipzigGermany

Paul BowenUniversity of Cape TownSouth Africa

Spike BoydellUniversity of Technology, SydneyAustralia

Jagjit ChadhaBNP Paribas UK Holdings LtdUnited Kingdom

S E CheungCity University of Hong KongHong Kong

Chris CloeteUniversity of PretoriaSouth Africa

Charles CowapHarper Adams University CollegeEngland

Hoffie CruywagenAssociation of Quantity Surveyorsin South Africa, South Africa

Julian Diaz IIIGeorgia State UniversityUnited States of America

Tim DixonOxford Brookes UniversityEngland

Keith DuffConsultantUnited Kingdom

Richard FellowsUniversity of Hong KongHong Kong

Christopher FortuneHeriot-Watt UniversityScotland

Karen GiblerGeorgia State UniversityUnited States of America

Andy HamiltonUniversity of SalfordEngland

Cliff HardcastleGlasgow Caledonian UniversityScotland

Barbara HawardUniversity of PortsmouthEngland

John HenneberryUniversity of SheffieldEngland

Robert HomeAnglia Ruskin UniversityEngland

Aart HordijkROZ Real Estate CouncilThe Netherlands

Michael HoxleyAnglia Ruskin UniversityEngland

Will HughesUniversity of ReadingEngland

Eddie HuiHong Kong Polytechnic UniversityHong Kong

Norman HutchisonUniversity of AberdeenScotland

L Jide IwarereHoward UniversityUnited States of America

Ramin KeivaniOxford Brookes UniversityEngland

J M Lusugga KirondeUCLASTanzania

Andrew KnightNottingham Trent UniversityEngland

Richard LaingRobert Gordon’s UniversityScotland

Chris LeishmanUniversity of GlasgowScotland

David LewisRoyal Agricultural CollegeEngland

Colin LizieriUniversity of ReadingEngland

SM LoCity University of Hong KongHong Kong

Sim Loo LeeNational University of SingaporeSingapore

Sui Pheng LowNational University of SingaporeSingapore

David LoweUniversity of ManchesterEngland

Ahmed M Saleh OufGovernment of SharjahUnited Arab Emirates

John MansfieldNottingham Trent UniversityEngland

Pat McAllisterUniversity of ReadingEngland

William McCluskeyUniversity of UlsterNorthern Ireland

James MeikleDavis LangdonUnited Kingdom

Stuart MorleyGVA GrimleyUnited Kingdom

Wordsworth Odame LarbiMinistry of Lands, Foresty & MinesGhana

Modupe OmirinUniversity of LagosNigeria

Jacob OpadeyiUniversity of the West IndiesTrinidad and Tobago

Allison OrrUniversity of GlasgowScotland

Ali ParsaUniversity of UlsterNorthern Ireland

Rob PickardUniversity of NorthumbriaEngland

Frances PlimmerKingston UniversityEngland

Gwilym PriceUniversity of GlasgowScotland

David ProverbsUniversity of WolverhamptonEngland

Les RuddockUniversity of SalfordEngland

Sarah SayceKingston UniversityEngland

Karl-Werner SchulteUniversity of RegensbergGermany

Rainer SchultzUniversity of AberdeenScotland

Martin SextonUniversity of SalfordEngland

Low Sui PhengNational University of SingaporeSingapore

Francois VirulyUniversity of WitwatersrandSouth Africa

Clive WarrrenUniversity of QueenslandAustralia

Christine WhiteheadLondon School of EconomicsEngland

Peter WyattUniversity of the West of EnglandEngland

Saad YahyaSaad Yahya AssociatesKenya

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 regressionare observed, where the relationship to be modelled holds everywhere in the study area. However, this is oftenviolated 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 ofspatial effects (Fotheringham et al. 2002), is employed to examine the relationship between transportaccessibility 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 linkedto the distance of the observation to the location of the regression point. This study identifies that therelationship between transport accessibility and land value varies over space. This suggests that a land valuecapture policy must take into account this lack of homogeneity and that the application of a uniform ‘tax’would be inappropriate.

Contact:

Grace Hongbo Du Email: [email protected]

Acknowledgments

The authors would like to acknowledge the support fromTrevor Arkless at Newcastle City Council, on behalf ofTyne and Wear Plan Partnership, for his assistance withthe acquisition of transport accessibility data. They alsoacknowledge the funding support from RICS EducationTrust, School of Civil Engineering and Geosciences ofNewcastle University, UK, Rees Jeffreys Road Fund, TheHenry Lester Trust Ltd and Lincoln Institute of LandPolicy, US. The work described in this paper is ashortened form of a thesis submitted for the degree ofPhD at the University. A copy of the thesis is availablefrom 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

lassical urban land economicsidentified the trade-off betweentransport cost and land rental thusexplaining why a parcel of land

located in city centre area with less transportcost has a higher value than a parcel of landlocated in the outskirts of the city. With thedevelopment of economies, the investment ontransport infrastructure to meet the increasingdemand for mobility requires substantial fundsfrom government for its implementation.However, there are many competing demandson public funds and lack of sufficient transportinfrastructure funding to meet all demands hasbeen a common issue worldwide. In thiscontext, the concept of a policy based on landvalue capture (LVC) has drawn much attentionwith various tax based or betterment basedapproaches having been explored as possiblemechanisms to finance transportinfrastructure. Indeed, behind these LVCmechanisms, there is the fundamentalquestion that needs to be answered. This is: towhat extent can transport infrastructure uplift

land value?

In an attempt to answer this question, theextension of the Tyne and Wear Metro toSunderland which opened in March 2002 wasexamined (Du and Mulley 2007) but this studyfailed to provide evidence for a significantpositive effect on land value resulting from theSunderland Metro Extension (SME). However,this study identified methodological concernswhich could only be resolved by taking a moredisaggregate approach to the issue. The moredisaggregated research is the focus of thisstudy.

This study also took the opportunity to expandthe case study from SME to the whole Tyneand Wear Region and specifically includes anexamination of transport accessibility and theimpact of Metro access.

Land value is the present value or capitalisedvalue of land (Mills 1972). Classical landtheories, as expounded by Mills (1972), areconcerned with two types of land –unimproved land which is without structuresand improved land where the value includesthe value of structures and other capital, suchas utilities pipes, invested in the land. In thisstudy, the focus is on residential land value.However, house prices capture the value ofthe land and the building that is located on it.As a result, the methodology standardises forthe aspects of the building so that therelationship between transport accessibilityand land value can be identified.

1 Introduction

RICS Research l 9www.rics.org/research

CC

...to what extentcan transportinfrastructureuplift land value?

“

“

here has been substantial researchinto this issue in the US, but relativelylimited research within the UK. This

review first considers the US experiencebefore turning to work focussed on UK cities.

In the US, a study on Philadelphia’s commuterrail service identified a 3.8% and a 10%premium for house values in suburbanPhiladelphia and Southern New Jerseyrespectively by comparing the median houseprices served and not served by a rail line(Voith 1991). In Atlanta, a study found positivepremiums in lower income households andnegative premiums in higher incomehouseholds associated with the MARTA EastLine (Nelson 1992), while another studyclaimed negative effects occured from railtransit, but positive effects for highways, interms of office rents (Bollinger et al. 1998).

Using a comparison method, Gatzlaff, Smith(1993) and Armstrong (1994) uncovered upto 5% and 6.7% premiums for houses in the

communities served by Miami Metrorail andcommuter rail service in Boston respectively.

By examining five transit systems with variousservice levels, as well as highways, inCalifornia, Landis et al. (1994) revealed thatthe systems with high ridership and frequentservice, such as BART and the San DiegoLight Rail, generate significant positivepremiums of up to $2.72 per metre to thenearest station, whilst the systems serving amore limited market, such as with the SanJose Light Rail and CalTrain, produceinsignificant or negative premiums in houseprices, with no effect found as a result ofproximity 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 priceincreases, up to 25%, resulting from DallasDART rail transit, whilst Weinstein and Clower(2002) showed positive results for bothresidential and commercial property in relationto the Dallas rail transit, with a relatively simplecomparison method.

Cervero has conducted a series of studies intothis topic since 1993 and obtainedinconsistent results. Positive but marginaleffects on commercial land was found inWashington D.C. and Atlanta (Cervero andLandis 1993; Cervero 1994; Cervero andDuncan 2001) while significant benefits tocommercial land from rail transit was found inSanta Clara County (Cervero and Duncan2001), which is similar to the result from thesame case study by Weinberger (2001). Morerecent studies have had mixed findings for theimpact of rail transit on both commercial andresidential land in San Diego County and LosAngeles County (Cervero and Duncan2002a;2002b).

2 Literature review

10 l RICS Research

TT

In the UK, Riley, a south London propertydeveloper, estimated that the increases in landvalue around the new London Jubilee LineExtension underground stations was nearlyfour 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 theadoption of an ‘agents survey’ in Phase 2(Chesterton 2002).

Although the latter demonstrates positiveresults, the methodology is not as robust asthe 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 timingissues (RICS 2004).

The study into the longer-term impacts of Tyne& Wear metro also failed to find statisticalevidence of land value uplift resulting from theintroduction of metro using a comparisonmethodology (TRRL 1993). With a hedonicprice approach, similar findings were identifiedin the Sheffield Supertram study wherenegative effects were found during theconstruction stage but these disappeared afteropening (Lawless and Gore 1999). In theManchester Metrolink study, a surprisinglysignificant negative result of up to -8.1% wasdiscovered (Forrest and Glen 1995).

Having been aware of the variation existing inhousing market, Adair et al (2000) appliedhedonic price models in a sub-market analysisand found significant positive premiums, of upto 14%, in only two out of fourteen areas, withinsignificant results in the other twelve areas.

No discernable effects have been found in anexamination of the changes in residential andcommercial property value following theoperation of Croydon Tramlink even when amore sophisticated approach is used (theKriging 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 theapplication of GWR in the examination on therelationship between transport accessibilityand land value although both identify that adetailed and relatively large data set isrequired.

Overall, the findings of the literature are mixedwith positive and negative results, and thevariations from case to case can be clearlyseen. As the study on this topic advances,these variations have become more apparentas attempts to model the sub-effect ofdifferent transport facilities or the sub-marketof housing progress.

However, there are limitations in many of theempirical studies since global statistics, ratherthan local statistics, were utilised in thesestudies using spatial data, which violate theassumption of the traditional multipleregression model which requires thatobservations are statistically independent fromone another. Some methods dealing withspatial data, such as Spatial Autoregressivemodels have been attempted to resolve theseproblems in this subject area but they havenot been successful. Hence, a local modelwhich is better at handling spatial data isexpected to offer insights in this area.


n the studies described above, a numberof different methodologies have beenutilised. In the context of changes in land

value following improvements in accessibilitycreated by a new transport infrastructure, themost basic methodology is the comparisonmethod. This method compares the change toan area benefiting from enhanced accessibilitywith a control area similar in nature but nothaving the advantage of improved accessibility.

Many studies used the more sophisticatedapproach of hedonic price modelling. If it isassumed that the ‘price’ of a house isdetermined by a bundle of characteristicsrelating to factors related to the property’sinternal features (such as number ofbedrooms) and external factors (such as thoserelating to the neighbourhood in which theproperty is located and transport accessibility)then a hedonic model allows the implicit pricesof these characteristics to be estimated.Consequently, property prices can be seen asa function of a group of variables contained bythree vectors:

Pi = f (H, N, T) (1)

Where H, N and T is a vector of propertyfeatures (H), the neighbourhood environment(N) and transport accessibility (T) respectively.

There are a number of problems withdetermining how a hedonic model is specified(So et al. 1997; Sirmans et al. 2005) but themost serious disadvantage in a studyinvestigating the effect of transportaccessibility on land value is that hedonicmodelling fails to take account of spatial

dependencies and only gives an averageimpression of the relationship between thevariables under study.

When looking at houses in differentneighbourhoods, hedonic modelling cannotfully explain the social processes giving rise to the spatial effects of spatial dependency(spatial autocorrelation) and spatial non-stationarity (spatial heterogeneity). This is thedistinction between a global model (asprovided by hedonic modelling) and a localmodel, where the spatial effects can beincorporated and which require local statisticsfor analysis.

This distinction between local and global canbe clearly understood by an example: whenlooking at the relationships between variables,such as the age of houses and house prices,would we expect any difference between theglobal statistics and local statistics? The globalstatistics for England may show that the ageof houses does not affect house pricessignificantly. In fact, in some parts of England,some old houses, such as houses built inVictorian times, might have character thusgenerating higher prices than newer houses.Whereas in other urban areas, older houseswhich might have had lower standards tohouse workers in rapidly expanding cities inthe middle of the nineteenth century, might bein poor condition, resulting in substantiallylower prices than newer houses.

The contrasting relationships in different areasmay cancel, to different degrees, thesignificant effects of one another and give aresult that, on average across England, theage of a house appears to have no impact on

3 Geographically WeightedRegression (GWR)

12 l RICS Research

II

the house price. This shows how misleadingthe results from global statistics could be inexamining the relationships in the aboveexample. Hence, a set of local statistics, inwhich data are analysed at local level, wouldprovide more accurate information on therelationship to be examined.

Another major difference between global andlocal statistics arises from the way in whichthese statistics can be presented. Globalstatistics are non-mappable single-valuedstatistics with averaged or summed data forthe area under consideration whilst localstatistics are mappable multi-valued statisticswith local disaggregations of global statistics.Hence a global model is calibrated by datawhich are not tied to any geographical ormappable reference in contrast to local modelyielding local parameter estimates is a modelcalibrated by data tied to specific geographicalloci.

Traditional multiple regression models areglobal models (as are hedonic price models inthis context) and Geographically WeightedRegression (GWR) models, developed toachieve local forms of spatial modelling, arelocal models. GWR models can improve upon

the explanation offered by the global modelsproduced by hedonic modelling since theyallow for spatial variation or non-stationarity inthe regression parameters. A traditionalregression model can be written as:

yi = β0 + Σk βk χik + εi (2)

By allowing local parameters to vary a GWRmodel is expanded to:

yi = β0 (ui, vi ) + Σk βk (ui, vi ) χik + εi (3)

where (ui, vi ) refers to the coordinates of thei th point in space. This can be fitted byweighted least squares to give an estimate ofthe parameters at the location (ui, vi ) and apredicted value. In this model, data nearer(ui, vi ) is weighted more heavily than datafurther away. It is notable that the location(ui, vi ) is regression point and so it is notnecessarily the location of a data point. By thisgeographically weighted calibration, estimatesof the parameters are made for each datapoint with coordinates, which is then mappedfor interpretation.

Before considering the weighting processused in GWR, it is worth digressing a little tounderstand how the presence of spatialeffects cause problems in traditional multipleregression. In multiple regression it is assumedthat the relationship to be modelled holdseverywhere in the study area. However, this isoften violated in the analysis of house pricesdue to the presence of spatial effects.

Spatial effects may occur in two differentforms: one is concerned with spatialdependency, or its weaker expression, spatial


...on average across

England, the age of a

house appears to

have no impact on

the house price.

“ “

autocorrelation (they are not identical thoughthey are often used interchangeably in theliterature) and the other form is spatialheterogeneity, namely spatial non-stationarity(Anselin 1999). Spatial autocorrelation can beseen as spatial interaction whilst spatialheterogeneity (spatial non-stationarity) refersto spatial structure (Anselin 1999). Spatialdependency and spatial non-stationarity havebeen the major challenges in spatial dataanalysis (Fotheringham et al. 2002).

In the GWR methodology, adaptive spatialkernels (see Figure 1 and 2) are used in theweighting process. These work by capturingthe data points to be regressed by moving theregression point across the region. In an areawhere data is dense, spatial kernels with anarrow bandwidth are in use, whilst in an areawhere data are sparse, spatial kernels with awide bandwidth are utilised. This approachuses a bi-square function where the weight ofthe jth data point at regression point i iscalculated by:

wij = exp[-(dij /b)2] ≤ when dij ≤ b

wij = 0 when dij > b

(4)

14 l RICS Research

Figure 1: A spatial kernel

Source: Fotheringham et al. (2002), Geographically Weighted Regression

The Akaike Information Criterion (AIC) (Akaike1973) is the method used in this report forevaluating the explanation given by GWRmodels. AIC takes account of ‘goodness of fit’(as does R

2in a regression or hedonic model)

as well as accounting for the complexity of themodel in question. In terms of interpretation,the better model has a smaller AIC value. As a rule of thumb, two models are consideredsignificantly different when their AIC differs by more than 3 (Fotheringham et al. 2002). The GWR software is programmed to choosethe optimal bandwidths by seeking to minimisethe AIC.

In summary, therefore, the methodologyemployed in this study is to calibrate ahedonic model or global regression and thento produce local estimations which takeaccount of spatial effects, elegantly treated in the GWR methodology by explicitlyincorporating the locations of data intocalibration. In terms of software, the resultshave been produced using the GWR 3.0software with the aid of GeographicalInformation System (GIS) and this producesfirst of all the global regression (GR) followedby local estimates which are plotted in mapform using GIS.


Figure 2: GWR with adaptive spatial kernels

Source: Fotheringham et al. (2002), Geographically Weighted Regression

yne and Wear is a MetropolitanRegion 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 ofNorth Tyneside and the Borough of SouthTyneside, plus hinterland stretching intoCounty Durham and Northumberland. The population is 1.08 million or 463,000households, accounting for approximately43% of the population in the North Eastregion.

The Tyne and Wear Metro, operated by Nexus,is the backbone of the area’s public transportsystem. Metro was the UK’s first light railtransit developed from the heavy track,

carrying 37 million passengers annually andreadily accessible to around 25% ofhouseholds in Tyne and Wear. The map of Tyneand 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 Metrodoes provide significant coverage.

The Metro system was planned over 30 yearsago and whilst much of the system was builtin the 1970s, many key structures date fromthe Victorian age. The first 55.5km of thenetwork was completed in 1984, followed bythe 1.5km Newcastle International Airportextension opening in 1991. An 18.5kmextension to the South Tyneside andSunderland areas was opened in March 2002,

4 The Case Study

16 l RICS Research

TT

Figure 3: Map of Tyne and Wear Metro

Source: based on Tyne and Wear

Accessibility Modelling and Edina Digimap

running over the existing heavy rail route andon the alignment of the former railway line.

The network is now 78 km long with 58stations as compared to the original systemwith 26 stations. Special features include themix of track conversion from heavy rail andnew build together with new major bridgesand 14 km of new tunnels.

The primary data for this study was collectedin 2004, twenty years after the opening of theTyne and Wear Metro. Data was collected forthe whole of the Tyne and Wear area andtherefore included areas which had benefitedfrom long exposure to Metro, areas, such asSunderland, where the Metro was relativelynew and areas with no access to Metro.


Property (House) Data

Transaction property data are normallyconsidered as ideal data for property relatedanalysis on the ground that these prices arethe agreed and accepted prices by the market.In contrast, asking prices can be seen asexpected prices which are valued by agencies.As a result, they are able to reflect marketvaluation although there may be someunrealistic extreme cases. At the beginning ofthis study, transaction property data, includingcommercial rents and residential sales, wereexplored but they are either unavailable inEngland due to confidentiality issues oravailable only with limited information aboutthe property characteristics. Therefore, askingprice data for residential properties, which areopen to the market and do not suffer fromconfidentiality, are considered as the data setfor this study available.

In many cases asking prices are reasonableprices since there is evidence that asking

house price and transaction house prices arehighly correlated with the actual sales priceachieved being above 93% on average ofasking price in the UK housing market since2002 (Hometrack 2005). In May 2004, whenthe data for this study were collected, thetransaction house price achieved was, onaverage, 98.6% of the asking price in theNorth Region (Hometrack 2005). It istherefore possible to examine thedeterminants of house prices by looking atasking prices without introducing too muchinaccuracy. In fact, asking price may be moreappropriate than transaction data for thepurpose of examining the effect of externalfactors, such as transport accessibility, sincetransaction data may vary more due to someinternal factors such as decoration style,garden or parking space whilst asking pricescan be seen more generalised in terms ofinternal factors.

At the time of data collection, a websitewww.icnewcastle.co.uk provided a servicecalled ‘icproperty’ which advertised propertiesfor sale in Tyne and Wear Region withsufficient information. This allowed thecollection of data on number of bedrooms andhouse types. The data for this study has beencollected at the postcode district (e.g. NE1)but with full postcode units (e.g. NE1 7RU) foreach property advertised in May 2004. Theinternet data source provided 2855 records ofasking house price with the information of thetype of house and the number of bedroomsfor 40 postcode districts covering the wholeTyne and Wear Region. Based on the fullpostcode, these records then were allocatedwith Output Areas (OAs) and the coordinatesof the centroids of OAs concerned.

5 Data Collection

18 l RICS Research

Neighbourhood EnvironmentData

Neighbourhood environment data includingsocial economic data, such as householdstatus in relation to household income andenvironment data, for example, the schoolingenvironment, is required to try and explain theexternal characteristics of house prices. Thepurpose here is to try and standardise for allcharacteristics so that the effect of transportaccessibility can be observed.

In this study, household status is captured by‘Higher managerial and professionaloccupations’ and ‘Long term unemployed’which are found in one of the widely-usedstandard socio-economic classifications in theUK, and available in census data format.Proximity to good schools has been identifiedas one of the key factors to determine thechoice in location of houses in Tyne and Wear(TRRL 1984).

This is confirmed by other empirical studies(Cheshire and Sheppard 2004; Gibbons andMachin 2005; Reback 2005) althoughwhether it is primary school quality orsecondary school quality which adds more tohouse price seems to vary from city to city(Cheshire and Sheppard 2004). To reflect thequality of school amenities this study uses theaverage point score of secondary schools, as aresult of comparison between primary andsecondary schools, in the school performanceleague table published annually by theDepartment for Education and Skills (DfES2004).

Transport Accessibility Data

The Tyne and Wear Plan Partners (TWPP)commissioned the development of the Tyneand Wear Transport Accessibility Model(TWTAM) as a result of the Local TransportPlan requirement (TWPP 2003). At the time ofthis study, this model produced travel time asan accessibility measurement. These werebased on either a ‘closest to facility’ method ora weighted Hansen method for publictransport (hourly between 0700 and 2300)and for car travel at different road states(capacity speed/half-capacity speed/fullspeed) to core services calibrated at oneminute intervals.

The closest method to education by publictransport, for example, is calculated as thetravel time to the nearest school; theassociated weighted Hansen accessibilitymeasure is more complicated and is calculatedby reference to a gravity-based formulation asfollows:

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 indexesavailable destination zones reachable fromzone i; Aj is the number of jobs at zone jaccessible to large employers; λ is adeterrence parameter consistent the tripdistribution and tij gives overall travel timefrom zone i to zone j (both journey time andwalking to/from bus stops).


It is worth noting that at the time of this study,public transport accessibility in the TWTAM isbased on Metro and bus combining to form asingle public transport network, making itimpossible to separate Metro from bus

accessibility. As Metro is a significant transportfacility in only part of the case-study Region,the effect of metro accessibility is separatelyinvestigated by the use of a group of dummyvariables for the thresholds of distance to thenearest Metro station.

20 l RICS Research

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) InternetSEMI Type of semi-detached (1=yes; 0=no) InternetDETA Type of detached (1=yes; 0=no) InternetFLATBED FLATBED = FLAT * BEDROOM Excel TERRBED TERRBED = TERR * BEDROOM ExcelSEMIBED SEMIBED = SEMI * BEDROOM ExcelDETABED DETABED = DETA * BEDROOM ExcelBEDROOM 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 TWPPCARCEPR Car transport accessibility to primary

schools at capacity speed TWPPCARCEMP Car transport accessibility to large employers

at capacity speed TWPPDISTMS1 the house within 200 metres of a Metro station

(1=yes; 0=no) GISDISTMS2 the house within 201-500 metres of a Metro station

(1=yes; 0=no) GISDISTMS3 the house within 501-1000 metres of a Metro station

(1=yes; 0=no) GISDISTMS4 the house beyond 1 kilometre of a Metro station

(1=yes; 0=no) GIS

Correlation Issues

Multicollinearity is a potential problem in allregression based analysis and has arisen inthis study. Whilst linear dependence betweenindependent variables frequently occurs, it isthe degree of linear dependence that cancause estimation problems since theconsequence of multicollinearity is thatparameter estimates become non-significantlydifferent from zero although R2 and adjustedR2, as measures of goodness of fit, remainunchanged.

A simple check on correlations betweenindependent variables is a common approachto examining the likelihood of multicollinearity.High correlations among independentvariables imply the likeliness ofmulticollinearity whilst the converse statementthat low correlations suggest the absence ofmulticollinearity may not be true. This isbecause, when there are a large number ofindependent variables, linear dependencies

may not show up through simple correlationbecause complex interactions may be involvedthat are not uncovered in a pairwisecomparison. Instability of parameter estimatesand enlarged standard errors can also beconsidered as a potential sign ofmulticollinearity (Thomas 1983).

Ad hoc tests such as these are often betterthan the more formal tests since many of theavailable tests make assumptions which thisstudy believes unrealistic (for example, the setof explanatory variables are jointly normallydistributed with zero population covariances(Thomas 1983)). Thus this study considersmonitoring the presence of multicollinearity bythe basic check of simple correlation betweenindependent variables and a carefulconsideration of the stability of parameterestimates and their standard errors.

6 Data Analysis


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

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 globalregression (GR) model, it is essential toestablish the best equation for the GR model.As it is not necessarily the case that the bestGR model leads to the best GWR model, thisstudy has sought to calibrate the best GRmodel consistent with sensible GWR results.Employing a stepwise approach to select theappropriate transport accessibility variables,the best equation in this study is identifiedbelow as Equation (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

Pi = α0 + α1FLATi + α2TERRi + α3DETAi +α4FLATBEDi + α5TERRBEDi + α6DETABEDi +α7BEDROOMi + α8EDU13PTi + α9%HPROFi +α10%UNEMi + α11PT08E13i + α12PT08E16i +α13CARCEPRi + α14CARCEMPi + α15DISTMS1i +α16 DISTMS2i + α17DISTMS3i (6)

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 parametersfor variables DETABED, BEDROOM,EDU13PT, %HPROF, DISTMS1, DISTMS2and 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 benegative (α4, α5, α10, α11, α12, α13, α14<0)as positive/negative relationships of the houseprices and these variables are expected.


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

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 **

The results of GR parameters are shown inTable 3 which contains the name of thevariable whose parameter is being estimated,the estimate of the parameter, and the tstatistic for the null hypothesis H0: α= 0against an alternative hypothesis of either H1:α>0 or α<0 as explained above and the levelof significance for rejecting H0 in a one tailedtest. The interpretation here, for the exampleof the average performance points of theclosest secondary school (EDU13PT), is thatan increase in one point will lead to £761(0.48%) increase in house price on average,holding everything else constant. The t-value is4.135 demonstrating that this GR parameter,EDU13PT, is greater than zero at a 1% levelof significance.

In this GR model, the internal factors areconsidered to combine the types of housewith number of bedrooms so the results arethe price per bedroom for each type of house.As the variable SEMIBED was dropped, theestimate for BEDROOM, α4, in factrepresents the price for one bedroom of semi-detached house. That is, one additionalbedroom of semi-detached house adds£35259 (22.05%) to the house price.

The estimate for FLATBED gives the value forone bedroom of flat as compared to a semi-detached house and similarly for a terraced ordetached house when considering TERRBEDor DETABED. The combined effect of oneadditional bedroom of flat/ terraced/ detachedis then given by adding α7 to α4 /α5 /α6

so that one additional bedroom in aflat/terraced/detached property is worth£17805/ £36496/ £48566 respectively,which takes up 11.13%/ 22.82%/ 30.37% ofthe mean price respectively. It can be seen, asexpected, that the internal features of houses,which are the main contributors to ‘improvedland’, give the principal explanation of houseprice.

However, the focus of this study is on theexternal factors which determine the landvalue rather than improved land value, and ontransport accessibility in particular eventhough these factors play a more minor role inthe determination of the house price ascompared to the internal features. Someresults, such as for EDU13PT, confirm theexpectation of the parameters mentionedabove. However, some results of the GRparameters (TERRBED, PT08EPR andCARCEMP) are the reverse to the expectationabove where they are significant. To see if theresults confirm the expectations mentionedabove or not, Table 4 displays a comparisonbetween outcomes and expectations. It showsthe alternative hypothesis, each parameter (asdiscussed above), the sign of the estimatedparameter and, where the parameter issignificantly different from zero, whether theoutcome matches a priori expectations. Thevariables where the outcome does not matchexpectations are in bold.

24 l RICS Research

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, thefactors of high professional andunemployment reflecting household statusdoes, as expected, significantly contributepositively and negatively to house valuerespectively. In addition, having a better schoolnearby, as identified by the point score, wasexpected to add value and this is true forEDU13PT variable.

In terms of car and public transportaccessibility, more travel time means worseaccessibility so, the alternative hypotheses,


# non-sig. at 5% level

Table 4: Outcomes and expectations

Variables H1 Estimated Outcome matchesparameters ( 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 + √



H1, for these parameters are expected to benegative so that one more minute of publictransport/car travel time (worse publictransport accessibility to schools and caraccessibility to larger employers) would lead tolower house prices as the underlying premiseis that, everything else being held constant,better accessibility would increase housevalue. Since car accessibility and publictransport accessibility are highly correlated,either car accessibility or public transportaccessibility to one service, such as secondaryschools, can be regarded as general transportaccessibility to that service as discussed in thesection above on correlation issues at thebeginning of this section.

Whilst the variables PT08E13 and PT08E16lift house value as expected and this issignificant at the 1% level, the variablesPT08EPR and CARCEMP reduce housevalue and are significantly negative in contrastto expectation. These results tell us that,globally, a house with better accessibility (oneminute less) to secondary schools andcolleges can add £1059 (0.66%)/ £610(0.38%) to house value whereas, one minutecloser to primary schools and larger employerswill reduce £2455 (0.76%) /£1217 (0.76%)house value.

Unlike some of the variables for car and publictransport accessibility, the results for variablesfor Metro access all agreed with expectationsalthough DISTMS3 (501m-1000m) gives aslightly higher value than DISTMS2 (201m-500m) which seems somewhat perverse.DISTMS1 is significant at the 1% level,suggesting that a house within 200m of aMetro station is worth £27396 (17.13%) more

than a house located more than 1km awayfrom a Metro station and this is very close tothe average estimation of a 20% premium forthe houses in proximity to Metro stations ascompared to houses further away made byestate agents in Newcastle upon Tyne (Hass-Klau et al. 2004). DISTMS2/3 gives£14193 (8.87%)/ £17598 (11.00%) positivepremiums 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 globalresults are ‘average’ results and that it ispossible that averaging effects over the Tyneand Wear study area could lead to unexpectedresults as identified by the example in thesection on GWR above. The focus of this workis to examine the contribution that a localmodel approach can add and the results fromGWR are considered next.

GWR results

The GWR software provides diagnosticinformation for the GWR model including anANOVA which tests the null hypothesis thatthe GWR model has no improvement over aglobal model. These are shown in Table 5where the F test suggests that this GWRmodel has a significant improvement over theglobal model in this study reported above. Inaddition, the adjusted R

2has increased from

0.615 in the global model to 0.728 in theGWR model implying that the GWR modelgives a better explanation, after taking intoaccount degrees of freedom.

26 l RICS Research

As identified above, GWR gives the ability toexamine spatial variability which can be hiddenin a GR model. All the local parameterestimates can be mapped but this studyconcentrates on transport accessibilityvariables – PT08E13, PT08E16, PT08EPR,CARCEMP and DISTMS1-3. The t- valuesand parameter estimates for these variablesare mapped in Figure 4 to Figure 7 by inversedistance weighted (IDW) interpolation withGIS. The best interpretation comes fromlooking at maps of local parameter estimatesalongside the maps of local t- value since thelocal t- values maps exhibit the significanceassociated with the local parameter estimates.In line with the discussion of the global resultsabove, the local estimates are presented interms of premiums (the percentage ofparameter estimate over observed houseprice) and these are presented on the samemap as the t-values to give values and theirsignificance together as a series of colouredsymbols.

The maps are presented in pairs. In the first ofeach set, the house price premiums are setalongside the significance of these values inthe local model. To interpret the maps, it isimportant to remember that a t-value havingan absolute value of approximately 2 or highersuggests, in a large data set, that an estimatoris significantly different from zero. In the localmodel, two tailed tests are carried out, incontrast to the global model where one tailedtests were used, since GWR is looking at eachobservation separately with no a prioriknowledge as to whether the spatial effectswould have a positive or negative effect. So, inthe discussion of PT08E13, PT08E16,CARCEPR and CARCEMP, the t- value isclassified by four bands: t>2.00, 2.00>t>0,0>t>-2.00 and t<-2.00. The percentage ofparameter estimates is classified in five bands,choosing 5% as the benchmark to showvisually the presence of positive and negativepremiums. In t- value maps, the lightest areasand darkest areas are significant with thelightest areas exhibiting positive housepremiums and the darkest negative house


Table 5: ANOVA results

Source Sum of Squares Degrees of freedom MS F value

OLS residuals 7534825890220.1 18.00

GWR improvement 2706671403008.0 265.16 10207867309.4160

GWR residuals 4828154408116.5 2553.84 1890543558.0949 5.399

sig. at 1% level

premiums. In each map, the value of the globalvalue, as percentage of parameter estimate, isset as single band to highlight the differencebetween global parameters and localparameters. The first Figure will be interpretedin detail to illustrate these points.

In the maps of parameter estimates, bandshave been set up to illustrate the parameterestimates in more detail. The yardstick used todetermine the bands is based on giving resultswhich provide clear and sufficient informationon parameter estimates, taking account of theglobal value for the particular parameterestimate. In the following subsections, theresults for each transport accessibility variableare demonstrated and interpreted in detail. As mentioned above, because of the highcorrelation between each pair of transportaccessibility by car and by public transport, nomatter the variables of public transportaccessibility or car accessibility are selectedfor the best equation, the results from thesevariables should represent the results fromtransport accessibility covering both publictransport and car. For example, the results forthe first pair of maps concerning the variablePT08E13 should be interpreted as the resultof transport accessibility to secondary schoolsrather than the single public transportaccessibility to secondary schools.

In the interpretation of the maps which follow,specific areas of Tyne and Wear are identifiedand these are identified on Figure 3 (above inSection 4).

Transport accessibility tosecondary schools (PT08E13)

Figure 4 shows the map of t- values andparameter estimates associated withPT08E13. As shown in Legend, the lightestareas are significant negative t- values whilethe darkest areas are significant positive t-values. The areas in light and dark grey arerespectively non-significant negative andpositive t-values. This therefore shows thegeographical areas where the localparameters are significantly different from zeroand need to be considered alongside thecoloured dots which represent the localparameter as a percentage of the globalestimate, which are classified by five bands of< -5% in red, -5% to -0.66% (which includesthe global value) in yellow, -0.66% to 0 in lightblue and > 5% in dark blue. As explainedabove, negative parameter estimates shouldbe interpreted as positive premiums for houseprices while positive parameter should beinterpreted as negative premiums.

Figure 5 shows the map of parameterestimates associated with PT08E13.Parameter estimates have been classified by 9bands of £2000 between the values of£10000 and - £10000 with a single band of£1000 to examine the global value (shown inTable 3). The greatest negative values ofparameter estimates (positive premiums) arein red and the smallest positive values(negative premiums) are in dark blue.

It is clear from Figure 4 and Figure 5 that boththe t-values and estimated parametersillustrate considerable spatial variation.

28 l RICS Research

Although the GR suggests that bettertransport accessibility to secondary schoolsleads to a globally statistically significanthouse price premiums (£1059 per minute at0.66% of the mean house price), Figure 4shows that, in most of the Region, the twovariables appear to be unrelated shown asmost of the map being light and dark greysuggesting t-values of less than an absolutevalue of 2 and a lack of significantly differentfrom zero local parameters for this variable.Only three major areas (the west end of theTyne and Wear Region, Newcastle central areaand West Monkseaton in North Tyneside tothe east of the study areas and two minorareas south of the River Tyne (one inGateshead and one in Sunderland)) emergewith this relationship greater than the GRvalue and up to 46.85% (red symbols). Thissuggests that areas in the vicinity of a Metrostation, transport accessibility seems to bepositively capitalised in relation to the Metroaccess to secondary schools for pupils whilst,in the other areas, bus or car access tosecondary schools must be associated withthe positive premiums.

Figure 5 demonstrates that transportaccessibility to secondary schools givespositive premiums up to £13377/minute formajority of the houses in Tyne and Wear, albeitnon significant for most areas (according to

Figure 4 above). As can be seen from Figure5, the value of global parameter (£1059) fallsinto the blue green band which only coverslimited and areas, which, according to Figure4, are non significant. This means that theglobal average value is not indicative of thelocal value for most houses.

As with PT08E13 above, apparent spatialvariation can be seen in Figure 6 and Figure 7.In the GR model, the relationship betweenhouse price and transport accessibility tofurther education, colleges, is positive and


“ ...transport accessibility tosecondary schools givespositive premiums...

“

30 l RICS Research

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

Figure 5: Map of parameter estimates associated with variablePT08E13 Transport accessibility to colleges (PT08E16)

statistically significant. However, Figure 6shows this is not true for most houses in Tyneand Wear with the exception of two areas tothe south east of the study area in Sunderlandwhich are showing up to 8.81% positivepremiums. In contrast, for houses in a majorarea of Gateshead and Newcastle, thereappears to be a negative relationship betweenhouse price and PT08E16 with negativepremiums of up to a surprising 50.8%. Thereason for such disparity might relate to theneighbourhood environments that thesecolleges serve.

In Figure 7, areas in orange tending towardsred exhibits positive premiums that transportaccessibility to colleges adds towards house

prices whilst areas in yellow tending throughto blue illustrate negative premiums. Takentogether with Figure 6, this means that somehouses in Sunderland, where the significantpositive relationship exists and shown inorange and red, receive £1000-£4089/minute positive house price premiumsfrom their accessibility to colleges.

Whereas, houses in the areas of Gatesheadand Newcastle, where the significant negativerelationship lies and shown in lime tending toblue, are devalued by their accessibility tocolleges from £2000-£12477/minute. It isnotable that, from Figure 7, the highest


Figure 6: Map of t-values and percentage of parameter estimatesassociated with variable PT08E16

parameter estimates (red shading) can befound in the north of Tyne and Wear where thet- value appears non-significant (pale or midgrey) in Figure 6.

Transport accessibility to primaryschools (CARCEPR)

Global results show transport accessibility toprimary schools reduces house prices by£2454/minute (1.53% of mean house price)and this might relate to some negative effectof primary schools which is unknown to thisstudy. However, GWR results shown in Figure8 demonstrate that this is not the case for themajority of houses in Tyne and Wear apartfrom five areas in black, suggesting that the

negative effect of primary schools varies orhouseholds value such negative effectdifferently. As the distance to primary schoolsis expected to be within walking distance,accessibility by car or public transport seemsunimportant for most households.

In the areas in black, being close to primaryschools decreases house price by up to46.42% suggesting the households in theseareas value the negative effect of being closeto primary schools highly. Significant positivepremiums for house prices that can be foundin one area in the south of the Region needsfurther investigation.

Again, Figure 9 illustrates the picture ofvarying parameter estimates associated with

32 l RICS Research

Figure 7: Map of parameter estimates associated with variablePT08E16


Figure 8: Map of t-values and percentage of parameter estimatesassociated with variable CARCEPR

Figure 9: Map of parameter estimates associated with variableCARCEPR

transport accessibility to primary schoolsalthough the t-values shown in Figure 8indicate they are non-significant in most areas.In the five areas covered by light blue and bluein Figure 9, where parameter estimates aregiving £3000-£6702/minute negativepremium. Nevertheless, there is one areawhere significant positive premiums in houseprices can be found which fall into the £4000-£5165 band.

Transport accessibility to largeemployers (CARCEMP)

The results from the GR are interpreted asbetter transport accessibility to largeemployers reduces house price but this is aglobal average figure. Figure 10 shows that inthe local GWR estimation, there are threeareas where a one minute quicker car orpublic transport access to large employersadds a premium of up to 29.81% of houseprice. In particular, a large area in the centre ofTyne and Wear where the positive premiums(shown by negative significance) confirms thelatest trend of gentrification in the UK (Hass-Klau et al. 2004). It also can be seen clearlythat the positive premiums appear to decaywith distance from large employers, particularlythe centre of Tyne and Wear, in line with theclassical Bid-rent land theory (Alonso 1964;Muth 1969).

The west area which shows negativepremiums on property prices (shown by thepositive relationship between transportaccessibility to large employers) ishypothesised to have other strongerenvironment feature, such as nice countrysidelandscape, which contributes more strongly to

house price than better access to employment.The northeast area with a high positiverelationship is thought partly to be the resultof proximity to a seaside amenity and partly tobe caused by the attractiveness of Metro as ameans of access to employment. However,this would not appear to be the reason behindnegative premiums in the southeast areawhich requires further study with perhapsmore detailed socio-economic information.

The non-stationarity of parameter estimatesassociated with CARCEMP is clearly revealedin Figure 11 with 13 bands that have beenset, spanning from £12941 (showing thehighest negative premiums) to -£16123(showing the lowest positive premiums).Amongst the three areas with significantpositive premiums shown in Figure 10, thehighest premiums of £16123 can be found inthe central area of Tyne and Wear (in red) withpositive premiums of £4000-£6000 in theother two areas (in golden). Whereas, amongstthose areas displaying the greatest negativepremiums (£12941) can be seen in thenortheast area (in blue) and less negativepremiums of £4000-£8000 are found in thesoutheast area (in light blue). The westernarea has been covered by bands with muchmore modest but statistically significantparameter estimates (£1000-£6000).

Houses within 200 metres of aMetro station (DISTMS1)

In the next pairs of maps, the interpretationturns to look at the effect of proximity of aMetro station on the house price. The t-valuesfor these maps have been differently classifiedwith the insertion of an additional band for a

34 l RICS Research


Figure 10: Map of t-values and percentage of parameter estimatesassociated with variable CARCEMP

Figure 11: Map of parameter estimates associated with variableCARCEMP

t-value between 0 and 1 giving five bandsoverall. For all the variables based on distancefrom a Metro station (DISTMS***), valuesbetween zero and one are problematic asthese variables are dummy variables takingthe value of 0 or 1 itself. The additional bandof 0 to1 band of t-values thus highlightswhere the value of the dummy variable itself iszero reflecting those properties which aresome distance from the Metro. These dummy variables have been enteredinto the global model to reflect differentthresholds of closeness to a Metro station.Again for interpretation, a pair of maps isprovided for each variable. In the first map for

each variable, t-values are shown as indicatedabove with the local parameter estimatesbeing shown as a percentage of observedhouse price as a coloured dot symbol. In thesecond map, the parameter estimates forhouses falling into the relevant threshold aremapped as coloured dot symbols along with t-values for all the observations in the data setso as to have better idea of parameterestimates. The dot symbols for parameterestimates in the second maps are slightlybigger than the first maps for percentage ofparameter estimates. As before, theinterpretation of the first of these variables isidentified in depth.

In contrast to the global value of £27396(17.13% of mean house price), the premium ofbeing within 200 metres of a Metro stationcatchment area from the GR, Figure 12 and

Figure 13 show that statistically significantpositive premiums do not occur in most areasof the Region (only small areas of white). This suggests the global model may haveoverestimated the value associated with Metroaccess for most houses. The mid grey colourcaptures where the t- values are 0 to thesouth and the southwest of Tyne and Wear,and this indicates where the local regressionis problematic because the value of thedummy variables are all zero (in these areas allhouses are some distance from Metro). Apartfrom this, two areas exhibit significant positivepremiums. One is to the west of Newcastle

city centre and the other is in the northeast ofthe Region: in both these cases, being within200 metres of a Metro station raises thehouse price by between 5% and 31.37% (seeFigure 12), adding £30000-£47267 of apositive premium shown by blue symbols onFigure 13. It should be noted that there are nosignificant negative effects found for anyhouse within 200 metres of a Metro station.

36 l RICS Research

...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 estimatesassociated with variable DISTMS1

Figure 13: Map of t-values and parameter estimates associated withvariable DISTMS1

Houses within 201-500 metres ofa Metro station (DISTMS2)

Globally, houses within 201-500 metres of aMetro station gain £14193 (8.87% of themean house price) in house price premiumsas compared to properties 1 kilometre awayfrom a Metro station. Once again, Figure 14shows that this is the case only in thenortheast of the Region, where being within200-500 metres proximity to a Metro stationis capitalised in a positive premium of between£20000-£52092 (see Figure 15), taking upbetween 5% to 50.09% of house price (seeFigure 14).

Despite being largely non significant, there isevidence of considerable positive premiums inthe Newcastle city centre area. However,significant negative premiums, shown by reddots in Figure 14 and Figure 15 are found inthree areas, where being 200-500 metresproximity to a Metro station gives the negativepremiums of £10000-£32056 (see Figure15), taking up from 5% to 30.38% of houseprice (see Figure 14).

It is notable that, unlike DISTMS1, a largenumber of houses within 200-500 metres of aMetro station have negative premiums,suggesting that being located between 200-

38 l RICS Research

Figure 14: Map of t-values and percentage of parameter estimatesassociated with variable DISTMS2

500 metres to a Metro station is not valuedpositively by the households in these areas.Positive premiums that are statisticallysignificant (shown by light blue or blue dots inFigure 14 and Figure15) can be seen to thesouth east of the study area, which coincidewith the South Hylton area and were newlyserved by the SME. Here, households seem tovalue Metro accessibility higher thanelsewhere in Sunderland, perhaps becauseSouth Hylton is more distant from theSunderland city centre as compared to otherareas newly served by the SME.


Figure 15: Houses within 501-1000 metres of a Metro station(DISTMS3)

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 16and Figure 17) shows that there are only twoareas, where this proximity has a significantpositive effect on house price with premiumsof between £20000-£43977 (see Figure 17),taking up between 5% and 50.46% of thehouse price (see Figure 16).

40 l RICS Research

Similarly to DISTMS1, one area is to the westof the Newcastle city centre and the other isthe northeast of the Region.

Also similarly to DISTMS2, two areas – one inSouth Tyneside and the other to the north ofRegent Centre (just north of Newcastle CityCentre) exhibit statistically significant negativepremiums of between £10000-£23630shown by red dots (see Figure 17), which isbetween 5% and 42.71% of the house price(see Figure 16).

Figure 16: Map of t- values and percentage of parameter estimatesassociated with variable DISTMS3


It is worth mentioning that, as can be seenfrom Figure 16 and Figure 17, there are morepositive and higher house price premiumsaround the stations of Tyne and Wear Metrowhich have been in service for decades ascompared to the new stations of SME whichhad been opened just for two years at thetime of data collection for this study.

This suggests that any effect of Metro on landvalue takes rather a long time to occur.

...any effect of Metro

on land value takes

rather a long time

to occur...

“

“Figure 17: Map of t- values and parameter estimates associated withvariable DISTMS3

42 l RICS Research

s raised at the beginning of thispaper, the fundamental question thisstudy set out to answer is “to what

extent can transport infrastructure uplift land

value?”

This empirical work suggests that the localmodel approach with GWR is more successfulthan the GR in examining the relationshipbetween transport accessibility and land value.From the results and analysis above, it can beseen that the relationship between transportaccessibility and land value is fairlycomplicated and varies greatly over physicalspace. The GR model offers the basis forexplaining average variation in house priceswith the additional results from GWR clearlyrevealing a spatially varying relationshipbetween house price and transportaccessibility variables.

The local model, GWR, gives a statisticallysignificant improvement over the global modelby increasing adjusted R

2even after

accounting for degrees of freedom. Theresults from GR model shows strong evidencethat proximity to Metro stations can uplift thehouse price significantly and better transportaccessibility to secondary schools also canadd significant premiums to a house whereasbeing close to large employers does notincrease the value of a house as the classicalurban land theory suggests.

Taking a closer look using the GWR model, itis clear that the relationships betweenneighbourhood features are not stationary.This finding that using a local model approach,GWR, gives greater insight to the relationshipsbetween transport accessibility and land value.

The primary motivation for this study was toprovide evidence for potential implementationof a Land Value Capture (LVC) policy toprovide a funding mechanism for the provisionof new transport infrastructure. The GR whichgives an average result for the examinedrelationship might be useful for settingspecific values in a LVC policy whereas theGWR outcomes which illustrate therelationship at the local detail would suggesthow that policy might impact in terms ofwinners and losers.

7 Discussion and conclusion

AA

“

“

...The resultsfrom GR modelshows strongevidence thatproximity toMetro stationscan uplift thehouse pricesignificantly...


The GR demonstrates that proximity to Metrostations can statistically significantly increasehouse prices and better transport accessibilityto secondary schools also can generatesignificant positive premiums to house prices.This implies that there are possibilities forcapturing land value uplift associated withMetro services and transport accessibility tosecondary schools.

However the local analysis, provided by GWRidentifies the presence of non-stationaritybetween transport accessibility and land valueleading to transport accessibility having apositive effect on land value in some areas but

in others a negative or zero effect. Althoughneighbourhood features may help to explainsuch variation, this means the implementationof a uniform LVC policy would necessarily giverise to winners and losers. However, a betterunderstanding of winners and losers mayallow such a LVC policy to be a moreprogressive if ‘winners’ are in the relativelypoor areas.

The picture for Tyne and Wear is complicatedby the way in which other external factors mayplay a more important role than transportaccessibility to employment in determiningland value. This suggests that a policy of LVCshould not only apply to transportinfrastructure, but also other importantamenities in order to justify the land value.This is in line with Land Value Taxation whichcharges the site value towards the land

created by the society, including transportimprovements as well as amenities, such asseaside amenity and green open spaces.

However, the difficulty for a Land ValueTaxation policy is how to evaluate the land(site) value accurately taking account of all thedifferent features so as to establish a taxationmechanism in a justified and transparent way.

Again, as demonstrated in this empirical work,using GWR is a good way to examine the sitevalue accurately once a good dataset withsufficient information can be achieved.

This empirical work clearly demonstratespositive house price premiums resulting fromgood Metro access, although this is notuniformly distributed. As such it is supportiveof the fact that a policy to recoup the effectsof land value betterment follows new andbetter transport infrastructure.

Whilst it is rational to try to capture some or allof the increased value associated with thebetter accessibility offered by Metro, the areawhich benefits in this way needs to becarefully identified.

In this case study, being within 200 metresproximity to a Metro station seems to be mostbeneficial,l but care ought to be taken withhouses that close to Metro stations exhibitingnon-significant negative premiums (see Figure12 and Figure 13).

44 l RICS Research

As with much research, this study answerssome questions but raises further questionswhich can only be resolved by future research.The major areas that have been seen forfurther work have been generated by theGWR results. GWR, indeed like a microscope,can bring out previously unimagined detailsuddenly into focus. As a result, the outputs ofthe GWR local estimations provided in themaps have uncovered a great deal of localinformation prompting potential work in thefuture.

In particular, GWR has been useful inidentifying spatially varying relationship. This could be developed to produce rationalpredictors associated with transportinvestment or other amenities on relevant land.

This study has revealed variability over spacein the relationship between transportaccessibility and land value, giving clues to theinteraction between transport infrastructureand land use, which have long been a keysubject in the transport planning field. Forinstance, according to this case study, beingwithin 200 metres proximity to a Metro stationgives positive premiums for house prices overthe majority of the case study area with thenoticeable exception of one Borough, whichhas the highest unemployment rate in Tyneand Wear Region (TWRI 2001). This suggeststhat there might be some interesting linkworthy of further investigation.


Adair, A. et al. (2000) House prices andaccessibility: The testing of relationships withinthe Belfast urban area, Housing Studies,15(5), 699-716.

Akaike, H. (1973) Information theory and anextension of the maximum likelihood principal,In Geographically Weighted Regression: theanalysis of spatially varying relationships, (Ed,Charlton, M.) John Wiley and sons, WestSussex.

Anselin, L. (1999) Spatial Econometrics,School of Social Sciences, University of Texasat Dallas.

Armstrong, R. (1994) Impacts of commuter railservice as reflected in single-family residentialproperty values, Transportation ResearchRecord, No. 1466, 88-98.

Atisreal (2005) Property Value Study -Assessing the Change in Values Attributableto the Jubilee Line Extension, Transport forLondon, London.

Bollinger, C. R. et al. (1998) Spatial variation inoffice rents within the Atlanta region, UrbanStudies, 35(7), 1097-1118.

Cervero, R. (1994) Rail transit and jointdevelopment: Land market impacts inWashington, DC and Atlanta, Journal of theAmerican Planning Association, 60(1), 83-94.

Cervero, R. and Duncan, M. (2001), Transit'svalue-added: Effects of light and commuter railservices on commercial land values, in 81stTRB Annual Meeting, Transportation ResearchBoard, National Research Council, WashingtonD.C.

Cervero, R. and Duncan, M. (2002a) Landvalue impacts of rail transit services in LosAngeles county, National Association ofRealtors; Urban Land Institute.

Cervero, R. and Duncan, M. (2002b) Landvalue impacts of rail transit services in sanDiego county, National Association of Realtors;Urban Land Institute.

Cervero, R. and Landis, J. (1993) Assessingthe impacts of urban rail transit on local realestate markets using quasi-experimentalcomparisons, Transportation Research Part A:Policy and Practice, 27(1), 13-22.

Chen, H. et al. (1997) Measuring the impact oflight rail systems on single family home values:A hedonic approach with GIS application,Centre for Urban Studies, College of Urbanand Public Affairs, Portland State University,http://www.upa.pdx.edu/CUS AccessedAugust 2005.

Cheshire, P. and Sheppard, S. (2004)Capitalising the value of free schools: Theimpact of supply characteristics anduncertainty, Economic Journal, 114(499),F397-F424.

Chesterton (2002) Second property marketactivity final report, Jubilee Line ExtensionImpact Study Unit, University of Westminster.

DfES (2004) Performance tables 2003: Keystage 2 test results, Department for Educationand Skills,http://www.dfes.gov.uk/performancetables/primary_03/glossary.shtml#q11 (AccessedMarch 2005.)

References

46 l RICS Research

Du, H. and Mulley, C. (2007) The short-termland value impacts of urban rail transit:Quantitative evidence from Sunderland, UK,Land Use Policy, 24(1), 223-233

Forrest, D. and Glen, J. (1995) The impact of alight rail system on the structure of houseprices, Journal of Transport Economics andPolicy, 30(1), 15-29.

Fotheringham, A. S. et al. (2002)Geographically Weighted Regression - theanalysis of spatially varying relationships, JohnWiley & Sons Ltd.

Gatzlaff, D. H. and Smith, M. T. (1993) Theimpact of the Miami Metrorail on the value ofresidences near station locations, LandEconomics, 69(1), 54-66.

Gibbons, S. and Machin, S. (2005) Valuing railaccess using transport innovations, Journal ofUrban Economics, 57(1), 148-169.

Hass-Klau, C., et al. (2004) Economic Impactof Light Rail: The Results of 15 Urban Areasin France, Germany, UK and North America,Environmental and Transport Planning.

Hack, J. (2002) The role of transit investmentin urban regeneration and spatialdevelopment: A review of research and currentpractice, In Land value and public transport,Stage one - Summary of findings, RISC andODPM.

Hometrack (2005) Key housing marketstatistics for England and Wales, UK HousePrice Statistics,http://www.themovechannel.com/sitefeatures/uk-house-prices (Accessed 2005.)

Landis, J. et al. (1994) Capitalization of transitinvestments into single-family home prices,The University of California TransportationCentre.

Lawless, P. and Gore, T. (1999) Urbanregeneration and transport investment: A casestudy of Sheffield 1992-96, Urban Studies,36(3), 527-545.

Nelson, A. C. (1992) Effects of elevatedheavy-rail transit stations on house prices withrespect to neighbourhood income, In ImpactsOf Rail Transit On Property Values, (Ed, Diaz,R. B.) APTA 1999 Rapid Transit Conference.

Reback, R. (2005) House prices and theprovision of local public services: Capitalizationunder school choice programs, Journal ofUrban Economics, 57(2), 275-301.

RICS (2004) Land value and public transport,Stage two - testing the methodology on theCorydon Tramlink, RICS, ODPM and DfT, 52.

Riley, D. (2001) Taken for a ride, the Centrefor Land Policy Studies, London.

Sirmans, G. S. et al. (2005) The composition ofhedonic pricing models, Journal of Real EstateLiterature, 13(1).


So, H. M. et al. (1997) Estimating theinfluence of transport on house prices:Evidence from Hong Kong, Journal ofProperty Valuation and Investment, 15(1), 40-47.

TRRL (1984) The impact of metro and publictransport integration in Tyne and Wear, SR478, Transport Road Research Laboratory.

TRRL (1993) The longer term effects of theTyne and Wear Metro, CR 357, Transport RoadResearch Laboratory.

TWPP (2003) Annual progress report 2002-3,Tyne and Wear Plan Partners.

Voith, R. (1991) Transportation, sorting andhouse value, In Impacts of Rail Transit onProperty Values, (Ed, Diaz, R. B.) APTA 1999Rapid Transit Conference, 1-8.

Weinberger, R. (2001), Light rail proximity:Benefit or detriment? - the case of SantaClara county, California, in TRB 80th AnnualMeeting, Washington D.C.

Weinstein, B. L. and Clower, T. L. (2002) Anassessment of the dart LRT on taxableproperty valuations and transit orienteddevelopment,http://www.apta.com/info/briefings/dart2002.(pdf Accessed June 2004.)

48 l RICS Research


Submission of papersThe RICS encourages academics and otherresearchers to submit work that is relevant tothe discipline of surveying in all its fields.Authors may submit either by post or by email.In the first instance, submissions should bemade to the RICS. Postal submissions shouldbe made in triplicate to:

Stephen BrownHead of Research Royal Institution of Chartered Surveyors12 Great George StreetLondon SW1P 3ADUNITED KINGDOM

Email submissions of Word or PDF filesshould be sent to [email protected]

Telephone number for enquiries:+44 (0)20 7334 3725

The refereeing process is double blind and allsubmitted papers will be referred to at leasttwo reviewers.

The manuscript There is no fixed limit on the length of papersand each paper is published as:

� A PDF on the RICS website, with nocharge made for download.

� A stand-alone published document

Papers are placed on the RICS websiteimmediately that they are completed andprinted copies available shortly thereafter.

While there is no word limit, it is anticipatedthat papers would be no less than 5,000words and generally no more than 30,000words.

The manuscript must be in English, typed indouble spacing on one side of A4 paper only,with a 4 cm margin on the left-hand side. Thepages should be numbered consecutively.There should be no loose agenda or notes orother explanatory material.

Title pageThe first page of the manuscript must containthe full title: the name(s), affiliation(s),address(es) of the author(s); a title of notmore than 75 characters and spaces; and fivekey words for the purpose of indexing. If thereis more than one author, the correspondingauthor should be indicated.

The abstractThe second page should contain the title andan abstract. It should not contain the name(s)of the author(s). The abstract should notexceed 200 words and must be a clearsummary of the contents of the manuscript,indicating the contribution that the papermakes to knowledge and providing a clearstatement of the key findings and outcomes ofthe paper.

Illustrations Any illustrations must accompany themanuscript but should not be included in thetext. Diagrams, charts, photographs and maps,should be referred to as 'Figure 1', 'Figure 2'and so on. They should be numbered in theorder in which they are referred to in the text.

Illustrations can be colour or monochrome.When providing electronic version ofillustrations, please ensure that these are of asufficient resolution for reproduction. Aresolution of 300 dots per inch is theminimum acceptable standard for images.They will normally be reduced in size on

RICS Research paper series

50 l RICS Research

reproduction and authors should bear this inmind, particularly when selecting font sizes.Appropriate electronic submission ispermissible.

Proofs Proofs, in the form of a PDF file, will be sentto the corresponding author for correction andapproval. The website is used to provideongoing comment and feedback on papersthat are posted on the website, and commentsthat are provided will be passed through to thecorresponding author. Authors are providedwith five free copies of their paper. Furthercopies can bought at a price of £2.00 each.

PublicityThe aim of the RICS is to ensure that papersare promoted and publicised to appropriateacademic, professional, policy and mediaaudiences. In pursuit of this, the RICS mayseek to develop and issue supporting materialfor papers published, such as press releasesand summary documents. The RICS will liaisewith the corresponding author on the draftingof this material and on the appropriate degreeof involvement of the author in this process.

ReferencesThe Harvard system should be used.References in the text should be quoted in thefollowing manner: Jones (1999) or (Edge andMoody, 2001) or, if there are more than twoauthors ... Thomas et al. (2002). If there is acitation of a page number or numbers, theformat should, as appropriate, be Smith (1999,20), Smith (1999, 20-5), (Smith, 1999,20) or(Smith, 1999, 20-5).

References should be collected at the end ofthe paper in alphabetical order by the firstauthor's surname. If references to the same

author have the same year, they should bedifferentiated by using 1998a, 1998b and soon. The style should follow the examplesbelow:

Chau, K.W., MacGregor, B.D. and Schwann, G.(2001) Price discovery in the Hong Kong realestate market, Journal of Property Research,18(3), 187-216.

Brown, G.R. and Matysiak, G.A. (2000) Realestate investment - a capital marketsapproach, Financial Times Prentice Hall,Harlow.

If no person is named as the author the bodyshould be used, for example:

Royal Institution of Chartered Surveyors(1994) Understanding the property cycle,London.

Copyright Submission of an article to the RICS ResearchPaper Series is taken to imply that itrepresents original, unpublished work, notunder consideration for publication elsewhere.When submitting a manuscript, authors will beasked to transfer the copyright for their articleto the Royal Institution of Chartered Surveyors,if and when the article is accepted forpublication. The Royal Institution of CharteredSurveyors will not refuse any reasonablerequest by the author for permission toreproduce any of his or her contributions tothe series in other forms.

Permission to publish illustrations must beobtained by the author before submission andany Acknowledgments should be included inthe figure captions.

www.rics.org

May

20

07/1

00

/RIC

S R

esea

rch/

38

83

7/S

terli

ng

The Royal Institution of Chartered Surveyors12 Great George StreetParliament SquareLondon SW1P 3ADUnited Kingdom

T +44 (0)20 7334 [email protected]

RICS (Royal Institution of Chartered Surveyors) is the largestorganisation for professionals in property, land, construction andrelated 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 propertyrelated knowledge, providing independent, impartial advice togovernments and global organisations.

Research

100%

Transport accessibility and land value: a case study of ... · She is a registered transport planner in P.R.China and with ten years of industrial experience in transport planning,

Documents