1234 Spatial Hedonic Models
insignificant or positive signs for distance decay, a finding not compatible with
theory (Dubin, 1992).
Dubin (1988) introduced the concept of spatial autocorrelation into the treat-
ment of hedonic house price models. Her approach was based on geostatistical
principles, in which the structure of spatial autocorrelation follows from an esti-
mated theoretical semi-variogram.^2 As argued in section 26.4.1.2, the geostatistical
approach is conceptually most suited to the analysis of a sample of house sales
transactions. In spite of this, it has seen relatively few applications in applied spa-
tial hedonic work. Other than the work of Dubin and co-authors (e.g., Dubin, 1992,
1998; Caseet al., 2004), some notable examples include articles by Thibodeau (Basu
and Thibodeau, 1998; Gillenet al., 2001), Miltinoet al. (2004), and Bourassaet al.
(2007), as listed in in Table 26.1. An overview of the major methodological issues
is given in Dubinet al. (1999).
The bulk of applied work in spatial hedonic house price analysis takes a lattice
data perspective and employs the standard spatial lag and spatial error models.
An overview of several illustrative studies is given in Tables 26.2 and 26.3, for
cases where, respectively, the spatial lag and spatial error specifications were the
primary focus of attention. Topics covered range from the simple definition of
the hedonic equilibrium and explanation of price differentials to the valuation of
environmental benefits, accessibility to transportation systems, wildfire risk, and
the impact of preservation policies.
Around the same time as Dubin’s article appeared, Can (1990) was one of the
first to consider the implications of spatially autocorrelated errors in the estimation
of spatial regression models using the lattice perspective. Specifically, she allowed
coefficients of the structural characteristics to vary across observations in a spatial
lag specification. Using a sample of 577 house sales for 1980 in Columbus, Ohio,
she concluded that a linear neighborhood quality drift expansion model is the most
appropriate hedonic price specification. Later on, Can (1992) used data from 563
single-family house sales in 1980 for Franklin County to obtain heteroskedastic
consistent estimators for a spatial autoregressive model based on bootstrapping
techniques.
In recent years, the application of spatial econometric techniques in empiri-
cal hedonic studies has become more widespread. Most analyses still rely on ML
Table 26.1 Spatial dependence: geostatisticsArticle SourceDubin (1988) Review of Economics and Statistics
Dubin (1992) Regional Science and Urban Economics
Basu and Thibodeau (1998) Journal of Real Estate Finance and Economics
Dubin (1998) Journal of Real Estate Finance and Economics
Dubinet al. (1999) Journal of Real Estate Literature
Caseet al. (2004) Journal of Real Estate Finance and Economics
Miltinoet al. (2004) Journal of Real Estate Finance and Economics
Bourassaet al. (2007) Journal of Real Estate Finance and Economics