Integrating Land Use in a Hedonic Price Model Using GIS URISA 2001 Yan Kestens Marius Thériault François Des Rosiers Centre de Recherche en Aménagement et Développement Laval University, Québec, Canada
Presentation Outline Introduction Objective Method Results Conclusion
Introduction - Achieving a better understanding of the spatial and temporal dynamics of the Quebec City Region - Hedonic Modeling using an important database describing over 30,000 transactions covering the period - Integrating land use characteristics using GIS
Introduction What is Hedonic Modelling? Calculate the specific contribution of an attribute Explanatory power Predictive power Method based on multiple regression analysis (MRA) Sale price = 0 + 1 Var 1 + 2 Var 2 +…+ n Var n + Additive form: Multiplicative form:
Introduction What is Hedonic Modelling? Databases Statistical Techniques GIS A largely used method in property assessment: CAMA Which has taken advantage from the development of computer technology and the GIS domain
Introduction Model specifications: the explanatory variables Accessibility (car travel time/distance to services, etc.) Property-specific attributes (living area, lot size, nb of bathrooms, etc.) Socio-economic attributes (census data) Location (Euclidean distance to externalities) Environmental attributes (noise, vegetation, view, etc.)
Previous Work Numerous hedonic models but only few of which integrate the environmental dimension Sight is the most important sense in our sensitive experience with our environment Those few hedonic models which do integrate environmental dimensions use variables resulting from ground surveys which are money- and time-consuming The environmental dimension plays a significant role in the determination of house prices Vegetation has an overall positive impact on preference
Previous Work Morales 76: Manchester, Connecticut: +6 to 9% for houses with a good tree cover Seila et al. 82: +7% for new built houses with trees Anderson and Cordell, 88, Athen, Georgia, +3 to +5% for houses with trees Luttik 2000; 8 cities in the Netherlands, positive impact of green areas and presence of trees significant in 2 cities out of 8 : +7 and +8% on sale price. Dombrow et al. 2000; positive contribution of the presence of mature trees of +2%.
Previous Work Criticism: method of data collection use of front view photographs on-field surveying of the properties Consequence: biases bias related to fraction of vegetation from only one point of view bias related to subjectivity of surveyors
Previous work Quebec City Region Impact of high-voltage power lines, schools, shopping centers, landscaping attributes Explorations with GIS tools and statistical methods: PCA, Trend Surface Analysis, Kriging techniques, Spatial Autocorrelation measures, interactive variables. However, significant spatial structure in residuals.
Objective Improving the hedonic price models by adding environmental data Improving the explicative and predictive power of the models Reducing heteroskedasticity and spatial autocorrelation in the residuals Using an inexpensive and efficient method to obtain data for the whole area of study: using GIS tools
Method n=1,392 DataModelling procedure Sub-sample of 10% Range: $50,000-$250,000 Mean price: $112,000
Method 75 physical attributes DataModelling procedure Over 1,400 potential interactive variables 48 environmental variables 14 location variables 36 census variables 40 accessibility estimators
Method Extraction of land use information using color areal photographs DataModelling procedure continuity availability low price
Method 124 areal color photographs covering the area of study DataModelling procedure
Method Scanning: raster images DataModelling procedure Spatial referencing using road network, hydrology and buildings from topographic map (Geographic Transformer) Building of a mosaic (Arc View)
Method Categorization using ISODATA technique DataModelling procedure Tree coverage Lawn surfaces Barren land Mineral surfaces
Method DataModelling procedure Computing of land use information using buffer functions
Method Multiplicative form, explained variable Ln of selling price DataModelling procedure Three steps*: -Model A: Property-specific attributes -Model B: location, accessibility and census variables added -Model C: land use data added * Regression specification: OLS, stepwise procedure Controlling for multicolinearity (VIF), spatial autocorrelation (I of Moran), heteroskedasticity (Goldfeld-Quandt test).
Results Spatial Autocorrelation Model AModel BModel C
Results Spatial Autocorrelation Model AModel BModel C
Results Model AModel BModel C
Results Model C
Results Spatial Autocorrelation Model AModel BModel C
Validation Results Validation of the final model with the sub-sample of 10% Calculated predicted error of 10.9% Adj. R-square: Std. Error of estimate: 0.159
Results Effect of Barren land cover in a 100 m radius Lawn in poor condition increases the feeling of insecurity (Kuo 1998)
Results Negative effect of trees at a local scale for aged population In accordance with previous findings by Des Rosiers et al. (2001) Effect of Tree vs Mineral cover in a 80 m radius Positive effect of trees at a local scale for younger population
Results Effect of an attached garage considering lawn area in a 500 m radius
Conclusion Model performance: slight increase in explained variance, but important drop of spatial autocorrelation Use of interactive variables: the effect of an environmental attribute is not constant over space Use of areal photographs integrated in a GIS proved to be efficient and low cost
Integrating Land Use in a Hedonic Price Model Using GIS URISA 2001 Yan Kestens Marius Thériault François Des Rosiers Urban and Regional Planning Research Centre Laval University, Québec, Canada
Results Effect of Time Distance considering lawn areas in a 300 m radius and vegetation cover in a 60 m radius