Modelling Environmental Influences on Property Prices in an Urban Environment

Iain R. Lake, Andrew A. Lovett, Ian J. Bateman and Ian H. Langford
Centre for Social and Economic Research on the Global Environment (CSERGE; University of East Anglia & University College London) and Centre for Environmental Risk (CER) School of Environmental Sciences University of East Anglia Norwich NR4 7TJ.
Email: I.Lake@uea.ac.uk



Abstract

The aim of this research was to assign money values to the negative impacts associated with the noise and visual intrusion associated with new road developments. These impacts do not have observable prices and so have to be calculated indirectly. One way of doing this is to examine their effect upon house prices. The valuations such a method produces can then be included alongside other costs and benefits in the appraisal of a road development. However in order to calculate these prices, one also has to control for the many other factors that affect house prices, in addition to specifying the two road variables. In previous research this has required much time and effort which has consequently limited the scope of such studies. The aim of this project was to use a GIS and large scale digital data to derive all the required variables in a quick and efficient manner.  However, GIS allows a large number of possible explanatory variables to be calculated, leading to a large and complex dataset. This paper describes how such a dataset was modelled and presents price estimates for road noise and visual intrusion. It concludes by commenting upon the benefits of using GIS in this type of study and considers the main limitations to their wider adoption.

1. Background

1.1 Cost benefit analysis

Investment decisions concerning UK highways have traditionally been assessed using cost benefit analysis (CBA). This technique assigns monetary values to all the impacts of a proposed road. While the valuation of construction costs is relatively straightforward, the estimation of money values for road benefits such as travel time savings or accident reduction is more problematic, although agreed values for both now exist (Bateman et al., 1993). However despite a long standing and ongoing research programme undertaken by the Department of Transport and its agencies, two items which remain currently undervalued in highway CBA's are those of road noise and the visual impact of roads.

The economic theory underpinning the evaluation of noise is the same as that for any other item which in some way affects an individuals enjoyment of life or 'utility'. Economists argue that we can measure the value of a desirable item by looking at how much an individual is willing to pay for it (Turner et al., 1994). Individuals do not purchase lower levels of road noise or views without roads. Therefore economists have sought to value such 'goods' by looking at individual's purchases of other items which secure lower noise levels or reduced views of roads. Such a technique is known as 'hedonic pricing' (HP) (Freeman 1993; Hufschmidt et al., 1983; and Bateman., 1993) and has frequently been applied via the property market. Here, controlling for known determinants of property prices (see below), the remaining variation in prices can be related to focus variables (here road noise, and road visibility) thus providing information on the value of these variables. Such values can then be entered alongside (and in common units to) the other costs and benefits of a proposed road thus facilitating an improved CBA of the projected investment. This paper describes such a HP study and demonstrates the substantial advantages which application of a geographical information system (GIS) can bring to such a study.

1.2 The hedonic pricing method

In applying the HP method to the property market the determinants of house prices can be divided into four groups:

All these have to be adequately controlled for before the effect of an environmental variable can be elicited. One problem with such studies in the past has been the time and effort required to collect data on all these variables for a large number of properties. This research aimed to circumvent these difficulties by using large scale Ordnance Survey (OS) digital map data and a GIS to generate possible explanatory variables in a fast, efficient and readily standardised manner. Due to the advantages of a GIS the authors envisaged three main ways in which this study would improve upon previous work. A GIS should:

The use of GIS in HP studies is not new (e.g. Metz and Clark, 1997 and Powe et al., 1997). However, both these studies use relatively simple GIS techniques and it was felt that neither exploited the full potential of a GIS. This project aimed to do the latter, and specifically concentrate upon the environmental variables of visual impact and road traffic noise.

The paper begins by describing the study area before explaining how a GIS can be used to calculate possible explanatory variables. It then shows how the property price data can be modelled before illustrating how the effect of an individual variable can be calculated.

2. The study site

A 50 km2 urban area in Glasgow, Scotland was chosen as the study site for several reasons. Scotland was chosen because it is the only part of Great Britain where property prices are publicly available on a register; the Register of Sasines (Lloyd, 1996). This lists the address, price and registration date of each property transaction and can be obtained in digital format from the Land Value Information Unit at the University of Paisley.

It was decided to use an urban study site, as the high density of properties would allow a small study area thus reducing digital data storage requirements and, more importantly, processing time. Glasgow was chosen as it is a socially heterogeneous city with large variations in environmental quality. A sample of 4000 properties was extracted from the study area for the year of 1986. These addresses were grid referenced using OS ADDRESS-POINT, which gives a unique grid reference for each postal address in the UK (Martin et al., 1994).

3. Defining and extracting explanatory variables

In order to estimate the effect of an individual variable upon property prices, the structural, neighbourhood, accessibility as well as the environmental attributes of a property need to be calculated. The procedure used to define each of these will be considered in turn.

3.1 Structural variables

Structural variables define the fabric of each building and the plot upon which it is built. OS Land-Line.Plus records the location of ground features such as building outlines or fences with a spatial accuracy of 40 cm (OS, 1996). By extracting elected line features from these tiles, the floor area and plot area (property floor area + garden area) were calculated for each property. The type of property (e.g. terraced, detached) was automatically identified by examining the connectivity between individual properties. This procedure was enhanced through use of OS ADDRESS-POINT, which allowed the number of addresses in each property to be determined. As an example, two properties connected only to each other are likely to be semi-detached. However, if there are two individual addresses in each of these properties then this is likely to be a "four in a block" arrangement. A measure of the shape of each property was also determined by dividing the square root of the floor area of each property by its perimeter (Blamire and Barnsley, 1996).

However, if this was the only way of obtaining structural variables other important factors such as the number of storeys, construction material, or age would not be controlled for. Therefore the GIS analysis was supplemented with a simple external inspection of each property during which these variables were recorded. One limitation with this method is that variables relating to a property's interior cannot be recorded. This means that potentially important factors such as type of heating, double glazing, and the property's internal state of repair were not included in the analysis. A total of 53 structural variables were ultimately created and these included classifying each property into one of 14 categories depending upon its type (detached etc.), data from the Sasines register, and the number of ADDRESS-POINT seeds in each property. A number of variables were also combined and used to create measures such as "pre war stone faced properties" or "post 1945 flats".

3.2 Neighbourhood and accessibility variables

Neighbourhood variables describe the characteristics of the local area in which the property is located and census data are a good indicator of these attributes. Therefore each property was matched to census derived variables at three spatial scales varying from output area (the smallest area for which census data are available in Scotland), to all output areas within 200m of the property, and the postcode sector level. These allowed the consideration of local as well as wider neighbourhood effects. The census variables used were chosen on the criterion that previous studies had shown them to be related to the level of social deprivation in an area (Carstairs and Morris, 1992).

Accessibility variables define the ease with which local amenities can be reached from the property and for this study schools, bus routes, railway stations, shops, parks, and the Central Business District were all considered. Having located these facilities using various local directories, they were grid referenced using ADDRESS-POINT and Land-Line.Plus. Three separate measure of accessibility, car travel times, walking distance and straight line distance were then calculated from each amenity to every property. Car travel times were calculated using the road network and estimates of road speeds along various road types (DOT, 1993). Walking distances were calculated using the same network although this was amended to exclude routes unsuitable for pedestrians (e.g. motorways) and supplemented with pedestrian only routes such as paths and footbridges. The latter were identified from Land-Line.Plus.

3.3 Environmental variables

Having defined variables relating to the structure, neighbourhood and accessibility of each property, environmental variables relating of each property were specified. Although several different environmental variables were researched, this paper will concentrate specifically upon the visual impact and noise associated with roads.

3.3.1 Environmental variables: Visual impact assessment

The assessment of visual impact for any feature in the landscape is a good example of how a HP study can be improved through use of a GIS. In traditional HP studies visual impact has been assessed by laboriously visiting all the sample properties and, for instance, recording whether a river can be seen from the property (McLeod, 1984). Using GIS allows us not only to determine whether the river can be seen in a fraction of the time, but also permits more sophisticated measures of visual impact to be considered. For example, how much of the river can be seen and how far away it is (Fisher, 1996).

Obtaining measures of visual impact for each property was performed in two steps. The first involved calculating what could be seen from each property. Measures of impact were derived by weighting each visible cell by its distance from the observer (Howes and Gatrell, 1993).

To calculate what can be seen from each property a Digital Terrain Model (DTM) of the study area was created. This was created from several components, the first of which is land elevation. Contours and spot heights were extracted from Land-Form PROFILE tiles for the study area. This is a dataset produced by the OS and contains contours and spot heights from 1:10000 scale maps (OS, 1996). River and tidal limits were extracted from Land-Line.Plus and used to represent sharp changes in slope (breaklines). These were modelled into a triangulated irregular network (TIN) and then interpolation techniques were used to extract a 1 square metre grid of land heights. However, as the study area is predominantly urban, the impact of buildings upon what can be seen from each property also needed to be taken into account. This was achieved by extracting every building's location from Land-Line.Plus. Each building was then visually inspected via a site survey, and the height recorded in terms of the number of storeys. A grid was subsequently created of building heights by assuming that each storey was 3 metres high.

The land surface in cities is often disturbed by cuttings and embankments. Locations of such features are shown on Land-Line.Plus and then were taken into account by assuming each slope to be 30 degrees steep. This information was used to create a TIN of slopes in the study area. Although this method is fairly basic, the results that it produced were felt to be superior to any approach which simply ignored slopes. This TIN of cuttings and embankments was converted into a 1 metre grid. The grids of building heights, slopes and land elevations were combined to create a final DTM of the study area. This process is illustrated in Figure 1.

Figure 1: The procedure used to define and urban DTM

A viewshed was then calculated from a central point at the front or back of each property and was designed to simulate the view from a property's window. The viewsheds were thus calculated using an observer height of 2 metres and a viewing angle of 45 degrees either side of the perpendicular to each building. The resolution of the final DTM (1 m) was essential to preserve the accuracy of the original data. However, this meant that the processing requirements for each viewshed were large with the consequence that each viewshed had to be constrained to a horizontal distance of 500m.

The resulting viewsheds were overlaid with a land use map to show which land uses could be seen from each property. The land use map was created by extracting the lines and polygons representing various land uses from Land-Line.Plus. These included features such as roads, buildings and water bodies. The resulting map did not comprehensively classify all land uses, but identified those whose location could be determined with a large degree of certainty. These were also land uses which it was thought might affect property prices.

The impact of any land use on a property's view can be expected to decrease with distance from the property. Separate visual impact measures were recorded for both the front and back of each property. Each land use was extracted in turn from the viewsheds and one of three distance decay functions applied to each visible cell (no distance weighting, an inverse linear distance weighting and an inverse squared distance weighting). These grids were summed to derive a visual impact score for each land use at each property. Separate measures of visual impact were recorded from both the front and back of each property. This process is illustrated in Figure 2. The maximum weighted cell value was also recorded as a measure of the greatest visual intrusion. This procedure produced five measures of visual impact for each land use from the front of each property and further five from the back of each property. These are shown in Figure 3.

Figure 2: Deriving measure of visual impact

Figure 3: Measures of visual impact calculated from the front and back of each property

3.3.2 Environmental variables: Noise levels

The road traffic noise level at each property was calculated by using a simplified version of the procedure recommended by the Department of Transport (DOT, 1988). Simplification was necessary due to the high cost of purchasing a dedicated piece of noise modelling software that could be operated within a GIS. It was also appropriate as many assumptions had to be made in assigning traffic flows and compositions to each road. This means that a more sophisticated modelling procedure would not necessarily have improved the accuracy of the results. The semi-automatic procedure involved calculating variables such as the road gradient and the distance between the road and the property within a GIS. These were exported to a PC based programme (Microsoft Excel) and combined with other information to calculate the noise level.

Data on the volume and composition of traffic along a number of roads in the study area were obtained from Glasgow City Council. These were used in conjunction with various interpolation techniques to produce volume and composition estimates for all the roads in the study area. Estimates of road speeds were derived from national roadspeed estimates (DOT, 1993). The gradient of each road was calculated from the GIS and all these data were used to calculated the noise level being emitted from each road. Each sample property was matched to its nearest road and the GIS used to calculate the distance between the road and property. This in combination with data on the road noise level were exported into a programme, and the noise level at each property calculated. These noise levels were corrected for reflections from other buildings and ground absorption of the sound by making global assumptions about the nature of the study area.

This procedure will work well unless other roads, in addition to the nearest, have a large impact upon noise levels. This was accounted for in two ways. If the nearest road to a property was a multi-carriageway road, then the noise contribution from the other carriageways were calculated, and the noise levels combined. The second situation was if the nearest road to each property was a minor road but there was a motorway, A or B class road within 100m of the property. In these cases the contribution from this major road was calculated and the noise levels combined. This procedure is illustrated in Figure 4.

Figure 4: Calculation of noise levels at each property

4. Assessing the influence of different variable groupings

The methodology described above generated around 300 possible explanatory variables for each of the 4000 sample properties. However, before the influence of an individual variable was calculated it was decided to examine the explanatory power of the different variable groupings (structural, neighbourhood, accessibility or environmental).

This stage involved entering all variables of a specific group into an ordinary least squares (OLS) regression, and examining the adjusted R2. The natural logarithm of selling price was found to be the best functional form for the dependent variable. The regressions were performed using a stepwise regression technique which eliminated all variables which were not significant at the 0.1 level (0.1 or 10% was chosen to increase the number of variables and hence provide a broad estimate of the explanatory power associated with each grouping). This was achieved by re-running the model and removing the least significant variable at each stage, until no variables with a significance level of less than 0.1 remained in the model. As we are only concerned with the adjusted R2 no account needed to be taken of the correlations between explanatory variables. An overall model was also produced by including all possible explanatory variables in the regression. The results of this analysis are shown in Table 1.

Table 1: The explanatory power of different variable groups

Several aspects of these results are notable. The first is that the combined explanatory power of the variables is over 80%. This shows that the methodology has matched, and in many cases surpassed, the degree of explanation obtained in previous studies (e.g. Nelson, 1982; Garrod and Willis, 1992).

The results show that as a group, neighbourhood variables accounted for most of the variation in property prices followed by structural variables, environmental variables and accessibility variables. All groups showed high adjusted R2 values, reflecting a degree of inter-correlation. indicating that they are probably highly correlated with one another. It is interesting that neighbourhood variables showed the highest adjusted R2 values. This is not a result mentioned in previous studies and may be because some neighbourhood variables, such as the percentage of households with a toilet, are in essence structural variables giving the probability that a property contains a certain feature.

When all possible explanatory variables were entered into the OLS regression, a large number (125) were significant at the 0.1 level or better. Many of these were highly inter-correlated. The large number of explanatory variables and the high correlations give an indication of the likely problems in extracting individual variables from the OLS regression.

5. Assessing the influence of an individual variable

Principal Components Analysis (PCA) was used to address problems of collinearity and the large number of possible explanatory variables (326). This is a statistical technique which can transform a dataset with many intercorrelated variables into one with a smaller number of uncorrelated variables known as principal components (PCs). The technique works by constructing linear composites of the variables that capture most of the original information (Dunteman, 1989). For this reason it seemed ideal for this study. However, one disadvantage of PCA is that because the new variables (PCs) are composites of the original data, their interpretation in relation to the original variables may not be straightforward. When PCs are created they are often rotated using what is known as varimax rotation (Kaiser, 1958). This moves the co-ordinate axes so that the PCs are still uncorrelated, but ensures that variables have either high or low loadings on the PC. Such a procedure usually makes interpretation of the components easier.

The objective of PCA is to simplify the dataset, but unless we ignore some of the PCs we obtain the same number of PCs as original variables. However, some PCs explain more of the variation in the data than others. Therefore using the Kaiser rule (Dunteman, 1989), PCs explaining only a small amount of the variability in the dataset were eliminated by only selecting PCs with eigenvalues greater than 1. 72 PCs met this criterion and it was encouraging to find that most of them were readily interpretable.

An OLS regression was then performed with the natural logarithm of property price as the dependent variable and all the PCs as possible explanatory variables. The regression was implemented as previously, using a backward stepwise regression technique to eliminate all variables which were not significant at the 0.1 level. This produced a model with an adjusted r-squared of 75.3% and 50 significant PCs. Most significant PCs had effects upon property prices which were in accordance with prior expectations.

However, the aim of the analysis is to quantify the impact of individual variables. This cannot be done directly because the coefficients in the regression do not refer to individual variables. An appropriate procedure thus had to be developed and is illustrated in Figure 5. This involved substituting PCs related to a particular factor (such as noise) with the original variables in the regression. The new regression thus contained a mixture of PCs and variables. The variables were excluded from the analysis of they were correlated with any of the PCs by more than 0.3. If two variables were correlated with each other by more than 0.3 the variables producing the highest adjusted R2 in the regression were selected for inclusion. By excluding correlations greater than 0.3 it was felt that collinearity would be kept within acceptable bounds. Although the study examined the impact of a variety of individual variables this paper will concentrate upon the noise and visual impact of roads.

Figure 5: Extracting individual coefficients from the OLS regression

5.1 Road noise

Estimates of road noise were calculated for each property in the study area. However, in quiet areas many of these were below the background noise level (that is typically produced by sources such as people and the wind). This was not a problem in previous studies as noise was physically measured and so these sources would have been recorded. Therefore it was decided to focus the analysis upon properties with noise levels likely to cause annoyance and hence an impact upon property prices. In the UK compensation for noise nuisance, under the noise insulation requirements, is only given for noise levels above 68 dB(A) (DOT 1988). Noise below this level is assumed to be near to the background level. Therefore all properties with noise levels greater than 68 dB(A) were selected for this analysis.

New PCs were created from the variables in this subsample and regressed against the natural logarithm of property price. One PC related to road noise was significant in the model. This was substituted with the original noise variable and the result in Table 2 was obtained. This shows that each decibel increase in road noise depresses property prices by an average of 1.07%. This result is slightly higher than in previous studies. Nelson (1982) found an average noise depreciation over several studies of 0.40%. However, as previously mentioned, most of these studies used actual noise measurements and so did not need to select a specific noise range to study. Consequently they had a larger range of noise values. It is also important to realise that most previous studies were carried out in the US over ten years ago. The impacts of noise may differ between countries and have changed over time.

Table 2: The effect of noise upon property prices

5.2 Visual impact of road

When PCs relating to the visual impact of roads were removed from the model and substituted with the original variables, the results in Table 3 were obtained. These estimates have been produced for the average value of each variable in the dataset. This was necessary because many of the variables had a small range of values in the dataset. If the results were framed in terms of a unit increase in the variable then this might have interpreted beyond the range of values in the dataset.

Table 3: The effect of visible road upon property prices

The results show that the visual impact of roads depressed the average property price by about 2.5%. This is the first study where such estimates have been produced and so it is not possible to compare the results to previous studies. The estimates for the individual variables show that road visible from the back of a property has a positive impact upon prices whereas road visible from the front has a negative impact. This is an unexpected result. It can be hypothesised that because most roads are far from the rear of a property, then this indicates a property whose back is not congested with many buildings. This may be seen as a favourable asset and hence increase property price. Road is visible from the front of nearly all properties and so this effect is not repeated there.

6. How well have property prices been modelled?

The results showed that over 80% of the property price variation was accounted for by the explanatory variables. This result is similar to, and in many cases better than those reported in previous studies (Powe et al., 1997, Pennington et al., 1988, Garrod and Willis, 1992). Such a result is especially pleasing for two reasons. The first is that it was produced with no information about the interior of sample properties. This indicates that our other variables had a higher explanatory power that in previous studies. It also indicates that because Glasgow is a very heterogeneous area our explanatory variables have succeeded in accounting for much of the property price variation.

However, although the explanatory power of the model is encouraging, the ability to extract estimates for individual variables is equally important. The results showed that in the model most PCs were readily interpretable and tended to concentrate upon specific variables. This indicates that our study area was heterogeneous enough to produce a large variability in the datasets.

The procedure adopted to extract an estimate for individual variables also appeared to work well. The main difficulty would have occurred if the PCs related to the specific factor were also strongly correlated with other variables. This meant that we could not account for this other variable. However, most of our PCs were very strongly associated with one factor and so this was not thought to be a major problem.

The procedure adopted to extract individual parameter estimates has not been applied to the modelling of property prices in the past. It is our judgement, nevertheless, that we have produced a method though which individual parameter estimates can be extracted from such a large and complex dataset.

7. Has GIS proved useful in improving the modelling of prices?

GIS has allowed a wide range of variables to be calculated for individual properties. These were nearly always of a range and sophistication that exceeded all previous studies. GIS performed least well in defining structural variables because of the obvious inability to describe the interior of a property or variables such as property age. However, the main limitation to the use of GIS in property price modelling was the time and effort required to collect and input data into the system. As an example the location of shops had to be copied from telephone books and grid referenced using ADDRESS-POINT, before being input into a GIS. If more of these data sources were grid referenced and readily available, they would significantly decrease the time required for such a study.

This may become less of a problem as more organisations store data in digital format. A good example is the Yellow Point database (Yellow Pages, 1997), which is a joint venture between the OS and British Telecom. This combines data from the Yellow Pages business directory with grid references from ADDRESS-POINT. This would make the identification of amenities such as shops a straightforward process.

The grid referencing of property sales information was also time consuming. However, Scotland is developing a GIS based land register (Lloyd, 1996), which will make the extraction of grid referenced property price information relatively easily. This has the potential to revolutionise studies such as this if the register is integrated with other databases.

The National Land Information Service is an example of how, if such a register was extended, it could further aid property price modelling. This is a system being tested in Bristol which combines data from the land registry with digital data from the Ordnance Survey, as well as data held by Bristol City Council and the Valuation office (Land Registry, 1998; Chappalaz, 1996). Organisations such as the valuation office hold information such as the age and size of each property which if included would allow the user to extract grid-referenced property prices, with information on the structure of each property.

This may seem a revolutionary idea but a similar system, the Multiple Listings Service (Anderson and Cordell, 1988), is readily available in the US. Such a system would save time and allow GIS to focus upon accessibility, neighbourhood and environmental variables which cannot usually be specified from any existing databases.

8. Summary and Conclusions

The study has shown how through the integration of GIS and large scale digital data it is possible to derive structural, accessibility, neighbourhood and environmental relating to individual properties. These variables were found to match, and in many cases better, the explanatory power found in previous studies. The use of GIS produced a data set with many possible explanatory variables which were simplified through the use of Principal Components Analysis. A method to extract an individual variable coefficient was developed and it was found that, for all properties with noise levels greater than 68 dB(A), each decibel increase in road noise depressed property prices by an average of 1.07%. It was also shown that road visible from the front of a property depressed the average property price by 2.5%. Road visible from the back of the property increased the price, although this was hypothesised to be due to the presence of enclosed rear courtyards in buildings such as tenements. The main limitation in this study was found to be the difficulty collecting input data for the GIS. It was hypothesised that as more organisations start to hold information within a GIS, this should become less of a problem in the future.

Acknowledgements

This research is funded by the Economic and Social Research Council. (ESRC) We would also like to acknowledge the help that the following organisations have given to this research.

References

Anderson, L.M.., & Cordell, H.K. (1988). Influence of trees on residential property values in Athens, Georgia (U.S.A.): a survey based on actual sales prices. Landscape and Urban Planning, 15, 153-164.

Bateman, I.J., Turner, R.K., & Bateman, I.J. (1993). Extending cost-benefit analysis of UK highway proposals: environmental evaluation and equity. Project Appraisal, 8(4), 213- 224.

Blamire, P.A., & Barnsley, M.J. (1996). Inferring urban land use from an analysis of the spatial and morphological characteristics of discrete scene objects. Proceedings of the 1996 Remote Sensing Society Annual Conference.

Carstairs, V., & Morris, R. (1991). Deprivation and health in Scotland. Aberdeen: Aberdeen University Press.

Chapallaz, N. (1996). A National Land Information Service for Great Britain. Paper Presented at the European ArcInfo user group conference, London.

Department of Transport (Welsh Office) (1988). Calculation of Road Traffic Noise. London: HMSO.

Department of Transport (1993). Vehicle Speeds in Great Britain 1992; Transport Statistics Report. London: HMSO.

Dunteman, G.H. (1989). Principal components analysis. Series: Quantitative Applications in the Social Sciences. Sage University Paper 07-069. London: Sage Publications.

Fisher, P.F. (1996). Extending the applicability of viewsheds in landscape planning. Photogrammetric Engineering and Remote Sensing, 62(11), 1297-1302.

Freeman, A.M. (1993). The measurement of environmental resource values: Theory and methods. Washington: Resources for the Future.

Garrod, G., & Willis, K.G. (1992). Valuing goods' characteristics: an application of the hedonic price method to environmental attributes, Journal of Environmental Management, 34, 59-76.

Howes, D., & Gatrell, A. (1993). Visibility analysis in GIS: Issues in the environmental impact assessment of windfarm developments. Proceeding of the Fourth European Conference and Exhibition on Geographical Information Systems. Genoa, Italy.

Hufschmidt, M.M., James, D.E., Meister, A.S., Bowen, B.T., & Dixon, J.A. (1983). Environment, Natural Systems and Development: An Economic Valuation Guide. Baltimore and London: John Hopkins University Press.

Kaiser, H.F. (1958). The varimax criterion for analytic rotation in factor analysis. Psychometrika, 23, 187-200.

Land Registry (1998). National Land Information Service. http://www.open.gov.uk/landreg/nlis.htm

Lloyd, C. (1996) Keepers of the house: a new land-register system for Scotland. Mapping Awareness, 10(5), 40-43.

Martin, D., Longley, P., & Higgs, D. (1994). The use of GIS in the analysis of diverse urban datasets. Computers, Environment and Urban Systems, 18(1), 55-66.

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

Metz, W.C. & Clark, D.E. (1997). The effect of decisions about spent nuclear fuel storage on residential property prices. Risk Analysis, 17(5), 571-582.

Nelson, J.P. (1982). Highway noise and property values: A survey of recent evidence. Journal of Transport Economics and Policy, (1982) pp. 117 - 138.

Ordnance Survey (1996), Digital Map Data and Customised Services. Catalogue produced by the Ordnance Survey. Southampton.

Pennington, G., Topham, N., & Ward, R. (1988) Aircraft noise and residential property values adjacent to Manchester International Airport. Salford Papers in Economics, 88-7. University of Salford: UK.

Powe, N.A., Garrod, G.D., Brundson, C.F., & Willis, K.G. (1997). Using a geographical information system to estimate an hedonic price model of the benefits of woodland access. Forestry, 70(2), 139-149.

Price, C., (1991) Landscape valuation and decision making. Mimeo, School of Agricultural and Forest Sciences, Bangor University.

Turner, R.K., Pearce, D.W., & Bateman, I.J. (1994). Environmental Economics: An elementary introduction. Hemel Hempstead: Harvester Wheatsheaf.

Yellow Pages, (1997). A Yellow Point in Space and Time. Brochure published by the Business database, Slough.