Sorting out access and neighbourhood factors in hedonic price modelling
Published date | 01 June 2000 |
Date | 01 June 2000 |
DOI | https://doi.org/10.1108/14635780010338245 |
Pages | 291-315 |
Author | François Des Rosiers,Marius Thériault,Paul‐Y Villeneuve |
Academic
papers: Hedonic
price modelling
291
Journal of Property Investment &
Finance, Vol. 18 No. 3, 2000,
pp. 291-315. #MCB University
Press, 1463-578X
Received June 1999
ACADEMIC PAPERS
Sorting out access and
neighbourhood factors in
hedonic price modelling
FrancËois Des Rosiers
Faculty of Business Administration, Laval University, Canada
Marius TheÂriault
Director ± Planning and Research Centre, Laval University, Canada, and
Paul-Y Villeneuve
Department of Planning, Laval University, Canada
Keywords Modelling, Property markets, Geographical information systems, Location
Abstract This paper investigates the analytical potential of factor analysis for sorting out
neighbourhood and access factors in hedonic modelling using a simulation procedure that
combines GIS technology and spatial statistics. An application to the housing market of the
Quebec Urban Community (575,000 in population; study based on some 2,400 cottages
transacted from 1993 to 1997) illustrates the relevance of this approach. In the first place,
accessibility from each home to selected activity places is computed on the basis of minimum
travelling time using the TransCAD transportation-oriented GIS software. The spatial
autocorrelation issue is then addressed and a general modelling procedure developed.
Following a five-step approach, property specifics are first introduced in the model; proximity
and neighbourhood attributes are then successively added on. Finally, factor analyses are
performed on each set of access and census variables, thereby reducing to six principal
components an array of 49 individual attributes. Substituting the resulting factors for the initial
descriptors leads to high model performances, controlled collinearity and stable hedonic prices,
although remaining spatial autocorrelation is still detected in the residuals.
1. Introduction: context and objective of research
This paper deals with the integration of neighbourhood and access attributes to
hedonic modelling, with a focus on how to sort out cross-influences between
both series of factors so as to achieve an optimal model design while
minimizing information loss. The hedonic approach aims at explaining
property prices on the basis of their physical and neighbourhood-related
characteristics. Its purpose is to evaluate the respective contribution of each
attribute of the residential bundle to market value (Can, 1990, 1993; Dubin,
The research register for this journal is available at
http://www.mcbup.com/research_registers/jpif.asp
The current issue and full text archive of this journal is available at
http://www.emerald-library.com
The authors gratefully acknowledge Martin Lee-Gosselin, Corinne Thomas, JoseÂe Bouchard,
Isabelle Plamondon, Pierre Lemieux, Raynald Sirois and Yanick Aube for their valuable help at
various stages of this research. This project was funded by the Quebec Province's FCAR
program, the Canadian SSHRC, the Canadian CMHC and the Canadian NSERC. It was realised
in close co-operation with the Quebec Urban Community Appraisal Division, the STCUQ
(Quebec Urban Community Transit Society) and the Quebec Ministry of Transport.
Winner of the European Real Estate Society 1999 Award
JPIF
18,3
292
1998), using multiple regression analysis. While hedonic models have long
proved their usefulness as an analytical device, previous research has shown
that a substantial portion of price variability remains unexplained (Anselin and
Can, 1986; Dubin and Sung, 1987; Can, 1993; Dubin 1998). Moreover, the
appropriate neighbourhood factors needed to improve hedonic models may
change among locations and market segments (Adair et al., 1996), making it
difficult to integrate all significant factors. Finally, multicollinearity of model
attributes, as well as structural heteroskedasticity and spatial autocorrelation
among residuals is detrimental to the stability of regression coefficients (Dubin,
1988; Anselin and Rey, 1991; Can and Megbolugbe, 1997; Basu and Thibodeau,
1998; Pace et al. 1998; Des Rosiers and TheÂriault, 1999). These are issues that
need be addressed and that deserve substantial research efforts.
Geographic Information Systems (GIS) provide resources to enhance real
estate analysis and hedonic modelling (Can, 1992; Des Rosiers and TheÂriault,
1992; Thrall, 1993; Thrall and Marks, 1993; Rodriguez et al., 1995). While GIS
can improve the measurement of location and access variables ± namely by
resorting to time, rather than mere Euclidean distances ± their analytical
capabilities are greatly enhanced where spatial statistics methods are
integrated (Anselin and Getis, 1992; Griffith, 1993; Zhang and Griffith, 1993;
TheÂriault and Des Rosiers, 1995; Levine, 1996). Indeed, procedures such as
centrographic analysis, trend surface analysis, spatial pattern analysis and
autocorrelation analysis (Odland, 1988; Cressie, 1993; Ord and Getis, 1995;
Tiefelsdorf and Boots, 1997) as well as variography and Kriging techniques
(Dubin, 1992; Panatier, 1996) can help detecting additional neighbourhood
factors that must be considered to explain market variability.
All these methods greatly improve the analysis and modelling of the
geographical structure of housing markets. However, they do not overcome the
problem of sorting out adequately access and neighbourhood attributes in the
first place. Depending on data availability, these can be quite numerous and
induce severe collinearity in the model. To reduce its extent, one may limit the
number of descriptors to a minimum, thereby causing a partial loss of
information. Another option developed here ± is to resort to factor analysis
(Thurstone, 1947; Rummel, 1970) in order to generate independent complex
variables used as substitutes for initial attributes. While cross-influences could
possibly be detected between access and neighbourhood attributes, the two
series of house price determinants are considered separately in this paper due
to the data aggregation issue they raise: whereas accessibility may be
computed at the level of street corners, census data used to define
neighbourhood attributes are aggregated at the level of enumeration areas. In
spite of their usefulness for modelling purposes, they provide an averaged out,
and therefore less acute picture of local characteristics.
2. Measuring accessibility: time vs. distance
The accessibility and mobility issues have been addressed in a recent paper by
TheÂriault et al. (1999). Accessibility relates to the ability of individuals to travel
To continue reading
Request your trial