Valuing Location in an Urban Housing Market

Scott Orford
School of Geographical Sciences, University of Bristol, Bristol BS8 1SS. UK.


It is often said that the most important factor in determining house prices is location, location, location. So it is surprising what little research has been done in a UK context on the importance of locational externalities, particularly within urban housing markets. A reason for this lies in a lack of understanding to how socio-economic processes operate at various spatial scales, and how this relates to the fabric of the built environment. For instance, it is typical to treat urban housing markets as spatially uniform, with little or no consideration to the buildings, streets and neighbourhoods within which it operates. This has inevitably led to modelling difficulties and inconclusive results. However, by modelling house prices and locational externalities within a multi-levelled framework, many of the previous conceptual and technical problems may be resolved. This paper will demonstrate how an urban GIS has been developed that links data across several spatial scales, and how the Ordnance Survey ADDRESS-POINT product has allowed locational data to be georeferenced at the level of the individual property. By utilising ARC / INFO tools such as GRID and NETWORK, the paper shows how the complexities of locational externalities can be untangled. Finally, the paper will discuss some of the results with respect to the valuation of specific externalities, such as access to parks, schools and the influence of non-residential land-use. The paper concludes that an appreciation of the spatial scales of the built environment is crucial if the effects of location are to be valued correctly.

1. Introduction

This paper is divided into four main parts. The first two parts evaluates location in previous work, and proposes a new concept of how location affects property values. The third part discusses the nature of the study area, and the construction of a GIS to simulate this new concept of location, and generate locational data. The fourth part describes how the results of the hedonic models can be visualised within the GIS, allowing an interpretation of locational effects to be made.

2. Background

Hedonic house price models regard housing as a composite commodity, made up of distinct bundles of housing attributes that vary between properties (Rosen, 1974). The price of a property is a realisation of the price of these attributes, although they are never directly observed in property transactions. Housing attributes have traditionally been divided into structural attributes and locational attributes. The former describe the physical structure of the property and the land parcel within which it is located, whilst the latter are concerned with the property's location.

Although this division between structural and locational attributes is fundamental, theory offers little guidance in determining exactly what attributes to include in the hedonic model (Ohsfeldt, 1988). This is demonstrated in the literature, where the lack of widespread agreement has resulted in a diverse range of variables entering the hedonic specification (Graves et al., 1988). In addition, far more attention has been paid to date to structural attributes than to locational ones (Cheshire & Sheppard, 1995). This is not surprising. Data relating to the location of a property are arguably much harder to quantify and conceptually less tangible than data relating to its structure. Their derivation also requires more time and effort. As a result, model specifications have tended to contain a similar range of structural attributes, with the range of locational attributes being diverse and marginal in comparison.

2.1 The traditional view of location in hedonic models

In much of the previous work, the location of a property has been conceived in terms of relative and fixed locational attributes. Relative locational attributes are measures of the local neighbourhood, whilst fixed locational attributes quantify the location of a property with respect to the whole urban area, and pertain to some form of accessibility measure. Together, relative and fixed location attributes are said to equal the total utility of location (Krumm, 1980). Thus, Follain & Jemenez (1985) have classified locational attributes from previous studies into categories of (fixed) accessibility and (relative) neighbourhood quality measures.

Accessibility has been the most fiercely contested and the single most important measure of location in hedonic house price studies. Usually, accessibility has been gauged by various measures of access to the CBD, representing the influence of the bid-rent curve on property price as proposed in the urban economics literature. More recently, access to non-CBD centres have also been considered (e.g. Wadell et al, 1993). Neighbourhood quality has typically been quantified by surrogate measures, usually generated from census data aggregated to Enumeration District or Ward level. Dubin & Sung (1990) have classified neighbourhood quality measures in previous work using three broad categories relating to social economic class, local municipal services, and racial composition. These categories illustrate the surrogate nature of much of the work on locational attributes, shown by the lack of any explicit measures of environmental quality. Interestingly, Minchge & Brown (1980) argued ten years previously for such explicit measures of the local neighbourhood, which they defined in terms of aesthetic attributes, pollution levels and proximity to local amenities. Although these attributes can be regarded as direct measures of neighbourhood quality, it would appear that few researchers have adopted them.

A review of previous work indicates that similar types of research is often repeated. One of the probable reason for this lies in the conflicting and often spurious results notable of much house price research (Waddel et al (1993), Follain & Jimenez (1985), Ball, (1973)). This is especially true of work on accessibility, which has been plagued by insignificant, positive findings that contradict urban economic theory and has often resulted in cries of 'what happened to the CBD-distance gradient?' (Heikkila et al, 1989). More often then not this is a result of using poor and incomplete locational data. Ozanne & Malpezzi (1985) have warned that in many studies, the lack of more specific locational data may create significant errors in the hedonic house price model. Since the estimation of the hedonic model is usually by ordinary least squares regression, many of the contradictory and spurious findings could be attributable to violations of the assumptions of the ordinary least squares specification.

3. Towards a new conceptualisation of location

The differences in the treatment of locational and structural attributes in hedonic house price research illustrates the continuing uncertainty to the relevance of locational factors in determining property prices. This is remarkable since it has been over thirty years since the first urban economic models were developed. There has been very little consensus to the types of locational attributes that influence house prices, and empirical evidence is contradictory. In particular, evidence supporting the existence of a negative rent gradient from the CBD outwards has been conflicting, and has cast doubts upon the validity of the assumptions under-pinning the urban economic models. It would appear that locational attributes suffer from both conceptual and measurement problems. This section will re-evaluate the influence of location in hedonic house price research in terms of the quantification of locational attributes and the spatial structure of the housing market.

3.1 Locational externalities

The above has illustrated that the effects of location in house price studies is far from unproblematic. Lack of adequately measured data is only part of the problem. The main stumbling block is a conceptual one. Locational attributes are in essence measures of locational externalities. Pinch (1985) distinguishes between locational externalities depending upon whether or not they impose a cost or benefit upon the householder. Those associated with costs are known as negative externalities and have a detrimental effect upon house prices. Positive externalities are benefits received as a by-product of an amenity and these tend to have a positive influence upon house prices. Frequently a single amenity may emit both positive and negative externalities, such as a local shopping centre. This imposes negative externalities in the form of traffic congestion and noise pollution and positive externalities through convenience to the local residents.

Most externalities are local in their impact, with a distance decay effect in their extent and intensity. Generally, households closest to the source of the externality will be the most affected, with the intensity of this effect diminishing with distance. In addition, the larger the facility, the greater the intensity and range of these effects. In many cases the decay effect may not be a monotonic function of distance, since many externalities have both a positive and negative impact. Therefore the distance decay function will be non-monotonic, with the optimal location viewed as a trade off between the benefits of increased accessibility and the costs of proximity.

This conceptualisation of locational attributes as measures of locational externalities blurs the traditional distinction between relative and fixed location. This was Minchge & Brown's (1980) argument when they advocated specific measures of the local neighbourhood. By doing this they supported a more precise specification of locational externalities; that is, one which captured the magnitude and distance decay nature of the externalities. This cannot be achieved with the use of poorly specified 'blanket measures', which are typical of the types of proxy data that have been used.

3.2 The spatial dynamics of urban housing markets

The spatial dynamics of housing markets in hedonic research have tended to be conceived as operating in one spatial dimension, with house prices varying continuously across urban space. Warnings against such a view, such as by Schnare & Struyk (1976), have generally been ignored. However, urban space is divided up by transportation routes, housing stock and landuse into discrete units. Spatial spill over effects from these units implies that property prices are better conceived as contiguous rather than continuous. Moreover, these units can be nested into a hierarchy operating at several spatial scales. The most basic scale can be regarded as the individual property. Properties are nested into streets, and streets into neighbourhoods. The spatial structure of property values should reflect this structure of urban space. Hence, the spatial dynamics of the housing markets should be conceptualised as being multi-scaled, or multi-levelled.

This concept of the housing market as multi-levelled has very important implications for hedonic house price modelling. The main implication is that a differentiation should be made between compositional and contextual effects of location on house price. Contextual effects are the differences a place makes, such as neighbourhood quality, whilst compositional effects are the differences caused by the variations in the housing stock within each place. The need for such a differentiation was acknowledged in very early hedonic house price work. For instance, Kain & Quigley (1970), appreciated that the quality of the neighbourhood is to some extent influenced by the dwelling stock, and commented that the difficulty in separating the two was 'perhaps the most vexing problem encountered in evaluating the several attribute bundles of residential services' (pp. 533). However, differentiating between compositional and contextual effects is difficult in the traditional specification of the hedonic model, since ordinary least squares regression presumes only one single level of variation. Jones & Bullen (1994) have argued that single level hedonic models are problematic since locational (contextual) attributes are modelled at the same level as structural (compositional) attributes:

'[Neighbourhood] quality has been made an attribute of each housing unit, with no distinction between houses and the [neighbourhoods] in which they are located. [Neighbourhoods] and houses are treated as equivalent observations, although houses are likely to be more numerous than [neighbourhoods], and houses within a [neighbourhood] are likely to be more similar than houses in a different [neighbourhood]. When there is only one observation per [neighbourhood], the within-place variation is totally confounded with the between-place variation and no separate estimates of these distinct components is possible.' (Jones & Bullen, 1994. pp. 255).

Inferential errors are also likely to occur when in-appropriate single-level models are used to model multi-level data. These problems can be overcome by specifying the hedonic model, not as varying at a single level, but as varying simultaneously over a number of levels. This is the basis of multi-level modelling, a recent statistical technique that allows multi-hierarchical space to be explicitly incorporated into standard econometric models (Goldstein, 1995; Jones, 1991). Multi-level modelling of house prices have been investigated in the work by Jones & Bullen (1993, 1994), and these possibly form the basis of the next generation of hedonic house price models. However, in such models urban space is initially built in context free. Locational externalities are still needed to 'colour in' this space to reflect the unique context of the property's location. Therefore, the concept of multi-levelled location infers that locational externalities will operate at different spatial scales.

3.3 Multi-level locational externalities

Table 1 summarises how externalities may operate over different spatial scales. The externalities in each level can be differentiated by how they are affected by the activities of households and the attributes of property, and the range and extent of their influence. Hence, since fixed locational attributes, such as accessibility to the CBD, and relative locational attributes, such as proximity to non-residential landuse, are unique for each property, they can be conceived as operating at the property level. However, locational attributes, such as street quality are influenced by the activities of residents in the street and hence can be regarded as a street level externality. Relative locational attributes at the neighbourhood level are those externalities that affect prices across wider areas. For instance, the effects of racial and social composition of a neighbourhood may be seen as operating at this level, as well as concepts such as neighbourhood 'desirability'. An externality may also operate at more than one spatial scale. For instance, the externality effects of local amenities such as a school may operate at the property level with respect to issues of proximity, but also at the street level with respect to the catchment area. Both these two effects will influence property prices differently.

Table 1
A Multi-Level Conceptualisation of Locational Externalities

Property Level Externalities
Accessibility to CBD
Accessibility to Major Non-CBD Centres
Motorway Exits
Railway Stations
Shopping Centres
Suburban Employment Centres
Proximity Measures to Non-residential Landuses
Local Shops
Recreational Centres
Cultural / Educational Centres
Street Level Externalities
Street Environment
Class of Street
Street Quality
Non-residential Activity
School Catchment Areas
Neighbourhood Level Externalities
Housing Density
Proportion of Non-residential Landuse
Proportion of Open Space
Quality of Local Amenities
Social Composition
Racial Composition
Prestige / Desirability

3.4 GIS and locational externalities

It should be clear from the discussion that one of the main problems encountered with hedonic house price models is the treatment of locational data, whether in terms of modelling of geographic space or in the measurement of locational attributes. These issues are representative of spatial data analysis in general (Anselin and Griffith, 1988) and can be linked to both the disregard of the potential problems posed by spatial data, and of the availability of well defined spatial data which has historically been poor. However, conceptualising locational attributes in terms of locational externalities operating across several spatial levels removes the artificial dichotomy of fixed and relative location, and may also resolve some of the technical problems associated with hedonic models such as multicollinearity (Powe et al, 1995; Garrod & Willis, 1992) and spatial autocorrelation (Can, 1990; 1992). Moreover, it also implies that detailed disaggregated data is need to measure these externalities. Simple proxy measures of neighbourhood quality, representative of much previous research, are inadequate and unsatisfactory. Instead, the remainder of the paper will describe how a GIS can be utilised very effectively in hedonic house price research as a means of generating, manipulating and visualising large and complex location-specific datasets at various levels of disaggregation. Such data can be modelled using the multi-levelled hedonic specification to produce more realistic and conceptually more sound estimates of locational externalities.

A GIS is an ideal medium to approach hedonic house price research for several reasons. It is capable of organising and managing large spatial datasets, such as those used in hedonic house price studies. Moreover, a GIS can handle these data at various spatial resolutions, such as at the level of the individual property and neighbourhood, which is important in the context of this research. A GIS also provides a valuable platform for spatial analysis, particularly with respect to the distance and proximity measures that have caused controversy in previous work. Finally, a GIS can aid the visualisation of the spatial data and map the results of the modelling.

Although GIS technology has been generally available for well over a decade, its use within hedonic house price studies has been rare, with the few recent exceptions including Waddell & Berry (1993), Waddell et al, (1993), Sanchez (1993), Cooley et al, (1995), Kennedy, et al (1996) and Lake et al. (1998). The use of GIS in these studies has been mainly to calculate distance and proximity measures in terms of both physical distance and time using tool-boxes such as NETWORK in ARC / INFO. Other uses of GIS has been to calculate lot-size from digitised boundary coverage's of landuse (Waddell & Berry, 1993), the mapping of the error terms to determine the existence of spatial autocorrelation (Waddell et al, 1993), and the construction of viewsheds to model the visual impact of landuse upon house prices (Lake et al., 1998). Hence, it would appear that GIS has been under represented in this field of research.

4. The study area

The chosen study area is the Welsh capital, Cardiff, and its predominately nineteenth century urban core, the so called Inner Area. This area has previously been used for work concerning the changing geographies of revenue raising (Martin et al, (1992), Longley et al, (1994)), the results of which form the basis of this study. The Inner Area is characterised by nineteenth century terraced housing, with twentieth century semi-detached and detached property located around its periphery. It covers approximately 24 square kilometres, and contains around 46 000 individual properties, and 920 streets. It has been identified by Cardiff City Council as a convenient unit for various aspects of urban policy. It is an interesting area to study the effects of location since it is more heterogeneous than suburban locations, with property prices varying notably across smaller areas. Hence it could be argued that locational externalities play a more significant part in price determination than in the suburbs of Cardiff.

The data for this study came from a survey of estate agents, a Housing Condition Survey undertaken by Cardiff City Council and the 1991 Census. Estate Agents supplied information relating to the asking prices and structural attributes of around 700 properties in the Inner Area. The structural attribute data acquired was typical of those used in most studies and included property type, and number of rooms, but also included data pertaining to room size and to the structural quality of the property. The Cardiff Housing Condition Survey (CHCS) provides a recent detailed picture of housing and environmental conditions, as well as socio-economic characteristics of occupying households, of a one in five sample of privately owned domestic dwellings within the Inner Area. To aid this survey, the Inner Area was disaggregated into eighty-one Housing Condition Survey (HCS) areas, which were chosen by the City Council for their high degree of internal homogeneity of dwelling stock. The HCS areas aggregate precisely into wards, with the mean size of each area being approximately 450 dwellings. Several variables concerned with socio-economic class, family status/life-cycle and ethnicity were extracted from the 1991 Census at ward level in an attempt to model factors considered to influence housing supply and demand and thus residential differentiation (e.g. Hirchfiled et al., 1995; Blake & Openshaw, 1996). Principal Component Analysis was then performed on the variables and two components associated with socio-economic class and housing quality were synthesised. The socio-economic class is perhaps the most important category since this can provide proxy information relating to general affluence and deprivation.

5. Building a context - sensitive GIS

The Cardiff Arc / Info GIS that has been developed is based around four spatial levels of resolution; the property, the street the HCS area and the ward. Each of these levels are represented by a coverage, and each of the coverages are linked together by a set of unique identification numbers. The property level coverage is composed of point data referring to the location of each property. The basis of this coverage is the Ordnance Survey ADDRESS-POINT (OSAPR) product. This is a point coverage that contains individual postal addresses geo-referenced at the resolution of 0.1 metres, and coded by a unique identification number. The street coverage is a digitised line coverage of the street network in the Inner Area, covering approximately 920 individual streets. Each street was assigned a unique identification number. For the purposes of calculating accessibility measures, the line coverage was enlarged with the addition of main roads that connect the Inner Area to the M4 motorway at the northern edge of Cardiff. The HCS areas and wards are represented by polygon coverages, which both have unique identification numbers. It is the system of unique identification numbers that link the four coverages together. The GIS was used to link individual properties in the point coverage to streets in the line coverage and HCS and wards in the polygon coverages. In addition, since a requirement of the multi-level hedonic specification is that each of the four levels (properties, street, HCS areas and wards) nest perfectly into one another, some of the longer streets that traversed the HCS area and ward boundaries had to be divided into sub-street sections. An illustration of this can be seen in Figure 1.

Figure 1: Each of the four levels (properties, street, HCS areas and wards) nest perfectly into one another.

 In addition to these coverages, other coverages were constructed in an attempt to quantify the major locational externalities that would operate within the Inner Area. These were digitised coverages representing parks and open space, primary and secondary schools, hospitals, churches, major shopping / commercial areas, major sports / leisure facilities, railway lines and railway stations, the River Taff, industrial landuses and institutional landuses. The parks and open space category was further sub-divided to make a distinction of Bute Park, a major open space and recreational area in the centre of the city. In a similar manner, industrial landuses were sub-divided into 'heavy' and 'light' industrial areas. The former corresponds principally to the traditional manufacturing and extractive industries associated with the docks. The latter principally corresponds to modern trading estates that are generally devoid of these traditional types of industries. Institutional landuses were taken as non-residential buildings and activities relating to the Government and the University. The location of schools were represented by a point coverage and also a polygon coverage of their hypothetical catchment areas. Information upon the size and examination results of pupils was also added to the secondary school catchment area coverage. Figure 2 shows the resulting composite GIS, illustrating the complex nature of the built environment of the Inner Area.

Figure 2: Composite GIS illustrating the complex nature of the built environment of the Inner Area.

5.1 Using the GIS to generate locational Externality measures.

The construction of the four level GIS will allow the exploration and evaluation of the two implicit concepts of locational externalities. Firstly, that their effect on property values diminishes with distance, and secondly that they will operate across the four different spatial scales of the Cardiff Inner Area. Measures of all the externalities were generated at the different spatial scales using a variety of ARC / INFO commands.

5.2 Property level externalities

5.2.1 Measuring Accessibility using the GIS

A basic function of any GIS is to measure the Euclidean distance between two points. However, for a more accurate measurement of accessibility, the underlying typology of the road network should be taken into account. In this case, accessibility between two points is the shortest route on the network connecting them. Within ARC / INFO, accessibility along the street network can be calculated using the NETWORK module. An additional advantage of using a network to measure accessibility is that the shortest route need not be measured solely in terms of distance. Other costs, called impedance costs, can be used when calculating accessibility. Impedance costs take into account factors that may affect accessibility between two locations in an urban area, such as speed limits. Impedance costs were assigned as estimated travel times along each street. These were calculated by dividing the distance of each street by its speed limit. The speed limit was calculated using the information on the class of street (primary, secondary, residential, cul-de-sac) that was recorded in the CHCS. Roads that have a lower travel time and thus impedance cost will be favoured when calculating accessibility.

Accessibility to the city centre was calculated using the ALLOCATION command in the NETWORK module, working from the city centre outwards along the street network. This allowed the minimum travel time from the city centre to each node on the street network to be calculated simultaneously taking into account the impedance costs. Moreover, the ALLOCATION command allows more than one destination to be selected at any one time. Hence, accessibility to the nearest point of interest, such as the nearest railway station, can be calculated simultaneously for each property. Therefore, using this method, access to the CBD, the nearest M4 motorway junction and the nearest railway station in Cardiff were calculated for each property. A separate accessibility measure was also calculated for travel time to Cardiff Central railway station, the main rail terminus in the city.

5.2.2 Measuring proximity to non-residential landuses

The externality effects of non-residential landuse will depend upon their distance from the property and their relative attractiveness as an amenity / disamenity. Both must be modelled simultaneously. This can be achieved in ARC / INFO by using the ACCESSIBILITY command, which calculates a measure of proximity as directly proportional to the supply of attributes at a given location, and inversely proportional to the distance away from it. This can be summarised thus:

Pi = S Wj dij -b i=1,.., n j=1, .., m


Pi is the proximity of property i
Wj is the attractiveness of externality j
dij is the distance between property i and externality j
b is the exponent for distance decay
n is the number of properties
m is the number of externalities being measured

The attributes of the externality are used to compute an attractiveness index, whilst the effects of distance are scaled using a distance decay function. Two distance decay functions are provided within ACCESSIBILITY: a power function that provides a gentle cut off to destinations and an exponential function that provides a steeper cut off. The latter is typically used for computing interactions over small distances, such as within a city. The process of finding the value of b in computing ACCESSIBILITY is called calibration, and is an important aspect of measuring the externality effect. If the exponent is small, the externality effect increases and vice versa. Since this value is not known a priori, different values were used to generate different sets of externality measures for each landuse. These can then be modelled using the hedonic approach as a means of estimating the optimal exponent for each landuse. The attractiveness index used in the computation was taken as the area of land squared, and this was calculated within ARC / INFO using the AREA command. For those landuses represented by a point coverage, such as schools, the attractiveness index was set to unity. Hence, using the ACCESSIBILITY command, separate measures of the externality effects of non-residential landuses were calculated.

Proximity measures to the River Taff and the railway lines were calculated in a slightly different way. By using previous studies as a yard stick (Lansford & Jones, 1995; McLeod, 1984), it was hypothesized that the externality effects associated with these features would be quite small. To quantify this, four sets of buffer zones were generated at intervals of fifty metres from each feature . The POINT-IN-POLYGON command was then used with the property coverage to determine which properties fell within fifty, one hundred, one hundred and fifty and two hundred metres of the river and railway lines respectively.

5.3 Street level externalities

5.3.1 Measuring street quality externalities

Environmental quality data recorded for individual properties in the CHCS were aggregated to street level and then used to quantify two attributes of street environment: overall street quality divided into four categories (poor, below average, above average, good) and the impact of small scale non-residential activity located in the street. It is anticipated that these two attributes will affect all properties located within the immediate street. However, since it is not known a priori how a locational externality diminishes with distance, the environmental quality of one street may also influence properties in adjacent streets, depending upon the perceptions of the buyer. To account for this, a range of street quality externality measures were calculated, using buffer zones of various sizes generated around each property.

Two methods were considered for generating these buffer zones. The first method used a simple circular buffer zone around a property. Unfortunately, this assumes that the urban area is continuous, and does not take into consideration physical boundaries such as railway lines, rivers, main roads and the orientation of the street network, which may influence the buyers perception of the surrounding street environment. This is illustrated in the upper panel of Figure 3, in which a circular buffer cuts across a railway line and captures the attributes of the street environment on the opposite side, even though a buyer is likely to perceive these to be inconsequential and thus to have no influence upon property price.

Figure 3: Upper panel - a circular buffer cuts across a railway line and captures the attributes of the street environment on the opposite side. Lower panel - routes can then be used to generate buffer zones which lie exactly along the street network.

Instead, a more sophisticated method of extracting the street quality data was used, which took into account the underlying topology of the urban area. This involved using the street network as the basis for the data extraction procedure. The ALLOCATE command in the NETWORK module allows routes to be calculated along the street network, using the length of the street as an impedance cost to limit its extent. These routes can then be used to generate buffer zones which lie exactly along the street network, and is illustrated in the lower panel of Figure 3. This ameliorates the problem of the buffer zone crossing boundaries, such as railways. These buffer zones can then be used to extract street quality data held in the street coverage. However, a problem arises when specifying the maximum extent of each buffer zone. Theoretically, this would depend upon the influence of street quality on the surrounding properties. Because the extent of this is not known a priori, it was decided to restrict the extent of each buffer to 100 metres and 200 metres respectively. These values represent the range of the majority of street lengths in the Inner Area. A further consideration was the influence of road junctions and turns in the road, since these could diminish the perception of the quality of adjacent streets. This can be accounted for in NETWORK by specifying an extra impedance cost for a turn in the network, which would limit the size of the buffer zone and thus the extent of the externality effect. Such an impedance cost would necessarily be arbitrary, and a value of 50 metres was chosen since this would have the effect of diminishing the perception of street quality by a third of the overall length. The result, illustrated in Figure 4, is a more realistic view to how street quality would be perceived from any particular house, as opposed to using circular buffers that does not take the typology into account. A POINT-IN-POLYGON analysis was then undertaken using each of the buffers to extract the two attributes pertaining to street quality. Hence, each property had a measure of street quality based on two attributes at three spatial resolutions (0 - 50m, 50 - 100m and 100 - 200m).

Figure 4: A more realistic view to how street quality would be perceived from any particular house, as opposed to using circular buffers that does not take the typology into account.

Secondary school catchment areas had to be approximated using the THIESSEN command. This converts the secondary school point coverage into a Thiessen polygon coverage, representing each individual schools catchment area, and each property was placed into a catchment area using the POINT-IN-POLYGON command.

5.4 HCS Area level externalities

These externalities are basically blanket measures that will have an absolute effect upon all property prices within a HCS Area. The variables measuring the quality of local amenities were constructed using information from the CHCS. HCS Area level landuse externalities were calculated within ARC / INFO using the INTERSECT command to calculate the proportion of non-residential landuses in each HCS Area. The operation computes the geometric intersection of the landuse coverages and the HCS Area coverage, and only preserves the areas common to both. This allowed the total area of open space and non-residential landuse in each HCS Area to be extracted, and the proportion that this represented to be calculated using the AREA command. Housing density was calculated by similar means using the ADDRESS-POINT property coverage as a means of determining the number of properties in a particular HCS Area.

5.5 Ward level externalities

The social economic class variable computed using principal components analysis was used as a measure of social composition of each ward, whilst prestige and desirability were captured using the ward boundaries. Table 2 is a summary of all the locational variables constructed for the Inner Area study. This illustrates the importance of GIS and spatially disaggregated data in generating locational externality measures, and indicates a possible reason to why the previous studies that did not have access to such data failed to capture the complexity of locational attributes.

Table 2
A Summary of the Locational Attributes Generated by the GIS

Property Level

Street Level (cont.)

Accessibility to CBD
Street quality 0-50m: Above Average
Accessibility to M4 motorway
Street quality 0-50m: Good
Accessibility to railway stations
Street quality 50-100m: Poor
Proximity to hospitals
Street quality 50-100m: Below Average
Proximity to sports centres
Street quality 50-100m: Above Average
Proximity to community centres
Street quality 50-100m: Good
Proximity to institutional centres
Street quality 100-200m: Poor
Proximity to local shops
Street quality 100-200m: Below Average
Proximity to primary schools
Street quality 100-200m: Above Average
Proximity to secondary schools
Street quality 100-200m: Good
Proximity to Bute Park
Street non-residential landuse.
Proximity to parks / open space
Sch Catchment: Willows High School
Proximity to light industrial land-use
Sch Catchment: Fitzalan High School
Proximity to heavy industrial land-use
Sch Catchment: Cantonia High School
Rail 0 -50m
Sch Catchment: Cathays High School
Rail 50 - 100m
Sch Catchment: St Teilo's High School
Rail 100 - 150m
HCS Area Level

Rail 150 - 200m
Percentage Local Authority tenure
River 0 - 50m
Percentage of open space
River 50 - 100m
Percentage of non-residential land-use
River 100 - 150m
Housing density
River 150 - 200m
Quality of local shops
Street Level

Quality of local public transport
Road Type: Primary
Quality of local sport facilities
Road Type: Secondary
Quality of local parks
Road Type: Residential
Quality of local community facilities
Road Type: Cul-de-sac / Close
Neighbourhood Level

Street quality 0-50m: Poor
Social economic class
Street quality 0-50m: Below Average

6. Visualising locational externalities

The structural and locational attribute data were modelled using the multi-level hedonic specification to replicate the spatial dynamics of the Cardiff housing market. An important feature of the multi-level hedonic specification is that it allows the estimated parameters of the housing attributes to vary at each of the four levels of resolution, reflecting the influence of local context upon price. Hence, the spatial variation of externalities effects can modelled. Moreover, the estimated parameters can be used to calculate the effects of these externalities on each property in monetary terms, and these values can be visualised within the GRID module of the ARC / INFO GIS. The paper will conclude with a brief review of three externality price surface models in the Inner Area that have been generated using parameters estimated from a multi-level hedonic model that allows the parameters of the externalities to vary across different spatial scales (Orford, 1997).

Figure 5 is a price surface model of the impact of parks and open spaces on house prices in the Inner Area. The results of the hedonic model suggested that parks have a greater impact on property prices in streets which have smaller, cheaper property, than in streets which have larger, more expensive property. This is illustrated in the surface model, which shows that the price of parks and open space is relatively more expensive in higher density housing around Victoria Park, Lansdowne Park and Roath Park, and relatively less expensive around Tremorfa Park. Smaller parks have much smaller externality effects, with only properties with a view of the parks benefiting

Figure 5: A price surface model of the impact of parks and open spaces on house prices in the Inner Area.

Figure 6 shows the price surface of heavy industries. The sites which have the greatest influence on surrounding property prices, Seawall Road Industrial Estate and Butetown Works, are those immediately adjacent to residential property. Moreover, the price surface suggests that only those properties within visible or audio distance of the sites are significantly affected. There is also an area adjacent to Seawall Road Industrial Estate where properties are not significantly affected by their proximity to the negative externality. This can be explained by reference to the parks and open space price surface which shows that Tremorfa Park has a compensatory effect on the adjacent negative externality.

Figure 6: Price surface of heavy industries.

Figure 7 shows the importance of Bute Park on property prices in the immediate vicinity, with the distance decay of the effect diminishing by half within a few streets distance. The price surface also shows how the extent of Bute parks influence is affected by the road network, the housing stock and also by ward (or neighbourhood) boundaries. For instance, the externality effect is greatest south of Bute Park, where the housing stock is characterised by large properties. The northern effect is not as great, reflecting both the influence of a main road running parallel to the parks northern boundary, and also the smaller size of properties located here.

Figure 7: Importance of Bute Park on property prices in the immediate vicinity.

One interesting feature of these price surfaces is that they do not represent a simple mirror image of the externality, with prices simply decreasing with distance. Instead, the spatial variation has created a mosaic of prices, resulting in a complex interaction of structural and locational attributes, with areas of positive and negative externalities in juxtaposition. This is especially significant since it illustrates how locational externalities operate over very localised areas.

7. Conclusion

This paper has described how it is possible to use ARC / INFO GIS to calculate and extract measures of the locational externalities that affect urban property prices at a local level. This has been achieved partly due to the increase in the availability of spatially disaggregated socio-economic data, geo-referenced to a high resolution, such as the CHCS and the ordnance Survey's ADDRESS-POINT product. The GIS reflects a concept of location that combines the multi-level modelling approach to space and the concepts of locational externalities. Once these measures have been extracted, they can be modelled using a suitable hedonic house price specification, such as the multi-level hedonic specification (Orford, 1997). The results from these models have suggested that locational attributes influence house prices across different spatial scales, and that there exists a complex interaction between structural and locational attributes, with an intricate geography of positive and negative externality effects operating across very localised areas. This has important implication for hedonic house price research, since much previous work has been undertaken at much lower resolutions. Both the ability to model spatial data and the resolution of the data are important if efficient estimates of hosing attributes are to be made.


Anselin, L and Griffith, D.A. (1988): Do spatial effects really matter in regression analysis? Papers of Regional Science Association. Vol 65. pp. 11-34

Ball, M. (1973): Recent empirical work on the determinants of relative house prices. Urban Studies. Vol 10. 1973. pp. 213 - 233

Blake, M and Openshaw, S. (1996): School of Geography, Leeds University, Leeds. LS2 9JT

Can, A. (1990): The measurement of neighbourhood dynamics in urban house prices. Economic Geography. Vol. 66. 1990. pp. 254-272.

Cans, A (1992): Specification and estimation of hedonic house price models. Regional Science and Urban Economics. Vol. 22. 1992. pp. 453 - 474

Cheshire, P. & Sheppard, S. (1995): Evaluating the impact of neighbourhood effects on house prices and land rents: results from an extended model. Econometrica.

Cooley, R., Hobbs, M., and Clewer, A. (1995): Location and residential property values in M.Fischer, T. Sikos and L. Bassa (eds): Recent Developments in Spatial Information, Modelling and Processing, Geomarket Co.

Dubin, R.A. & Sung, C, H (1990): Specification of Hedonic Regressions: Non-nested tests on Measures of Neighbourhood Quality. Journal of Urban Economics. Vol. 27. 1990. pp. 97 - 110.

Follain, J. & Jimenez, E. (1985): Estimating the demand for housing characteristics: a critique and survey. Regional Science and Urban Economics. Vol. 15. 1985. pp 77 - 107.

Garrod, G. and Willis, K. (1992): Valuing the goods characteristics - an application of the hedonic price method to environmental attributes. Journal of Environmental Management. Vol. 34. No. 1 pp. 59-76

Goldstein, H. (1995): Multi-level Statistical Models, Kendals Library of Statistics 3. Second Edition

Graves, P., Murdoch, J.C., Thayer, M.A. and Waldman, D. (1988): The robustness of the hedonic price function - urban air quality. Land Economics. Vol 64. No. 3. pp. 220-233

Heikkila E. Gordon, P. Kim, J. (1989): What happened to the CBD-distance gradient? Land values in a poliocentric city. Environment & Planning A. 1989. Vol 21. pp. 221 - 232

Hirschfield, A. , Brown, P. and Todd, P. (1995): GIS and the analysis of spatially referenced crime data: experiences in Merseyside, UK. International Journal of Geographic Information Systems. Vol. No 2. pp 191-210

Jones, K. (1991): Multi-level Models for Geographical Research. Concepts and Techniques in Modern Geography 54, Norwich: Geo Books

Jones, K & Bullen, N (1993): A multi-level analysis of the variations in domestic property prices: Southern England 1980-87. Urban Studies. Vol. 30.8. 1993. pp. 1409 - 1426.

Jones, K. & Bullen, N. (1994): Contextual models of urban house prices: A comparison of fixed- and random- coefficient models developed by expansion. Economic Geography. 1994.

Kain, J.F. and Quigley, J.M. (1970a): Measuring the value of housing quality. Journal of the American Statistical Association.

Kennedy, G.A., Dai, M., Henning, S.A., and Vendeveer, L.R. (1996): A GIS-based approach for including topographic and locational attributes in the hedonic analysis of rural land values. American Journal of Agricultural Economics. Vol. 78. No. 5. pp. 1419-1437

Krumm , R. (1980): Neighbourhood amenities: an economic analysis. Journal of Urban Economics. Vol 7. 1980. pp 208 - 224

Lake, I. R., Lovett, A. A., Bateman, I. J., & Langford, I. H., (1998): Modelling Environmental Influences on Property Prices in an Urban Environment. Computers, Environment and Urban Systems. Vol. 21, No. 5.

Lansford, N.H. and Jones, L.L. (1995): Recreational and aesthetic value of water using hedonic price analysis. Journal of Agricultural and Resource Economics. Vol. 20. No.2. pp. 341-355

Longley, P. & Higgs, G. & Martin, D. (1994): The predictive use of GIS to model property valuations. International Journal of Geographical Information Systems. Vol. 8.2. 1994. pp 217-235.

Martin, D. & Longley, P. & Higgs, G. (1992): The geographical incidence of local government revenues: an intra-urban case study. Environment and Planning C: Vol. 10. 1992. 253 - 265.

Mcloed, P.B. (1984): The demand for local amenity: An hedonic price analysis. Environment and Planning A. Vol 16. pp 389 - 400

Mingche, M.L. and Brown, J.H. (1980): Micro-neighbourhood externalities and hedonic housing price. Land Economics. Vol 56. 1980. pp. 125-141.

Ohsfeldt, R.L. (1988): Implicit markets and the demand for housing characteristics. Regional Science and Urban Economics. Vol. 18. 1988. pp. 321-343.

Orford, S. (1997): Valuing the built environment: A GIS approach to the hedonic modelling of housing markets. University of Bristol. UK (Unpublished Ph.D. Thesis).

Ozanne & Malpezzi (1985): The efficacy of hedonic estimation with the annual housing survey. Journal of Economic and Social Measurement. Vol. 13. pp 153-172

Pinch, S. (1985): Cities and Services: The Geography of Collective Consumption. Routledge & Kegan Paul.

Powe, N.A., Garrod, D. and Willis, K.G. (1995): Valuation of urban amenities using an hedonic price model. Journal of Property Research. Vol. 12. pp. 137-147

Rosen, S (1974): Hedonic prices and implicit markets: Product differentiation in pure competitions. Journal of Political Economy. Vol. 72. pp. 34-55

Sanchez, T.W. (1993): Measuring the impact of highways and roads on residential property values with GIS: An empirical analysis of Atlanta in Klosterman, R.E and French, S.P (eds). Third International Conference on Computers in Urban Planning and Management. Vol. 2.

Schnare, A. & Struyk, R. (1976): Segmentation in urban housing markets. Journal of Urban Economics. Vol3. 1976. pp. 146 - 166.

Waddel, P. Berry, B. & Hoch, I. (1993): Residential property values in a multi-nodal urban area: New evidence on the implicit price of location. Journal of Real Estate Finance and Economics. Vol 7. 1993. pp 117 - 141