Spatial Sampling and Sample Size in Hedonic Property Models

Spatial Sampling and Sample Size in Hedonic Property Models

Nancy Lozano-Gracia, GeoDa Center, Arizona State University, nlozano@asu.edu,
Luc Anselin, School of Geographical Sciences and GeoDa Center, Arizona State University, luc.anselin@asu.edu

II World Conference of the Spatial Econometrics Association

Hedonic Property Models have been widely used to elicit an individual’s willingness to
pay for house or community attributes ranging from living area and number of bathrooms
to environmental quality. Although individual data on house sales is used in may cases,
many examples in the literature have used some sort of aggregation at larger scales
(Census tracts or even counties) to define the price equilibrium in a hedonic context.
Often such aggregation is by necessity without the availability of individual data. To
date, there has been very little systematic analysis of the consequences of using sampling
or aggregation on the value of parameter estimates in hedonic models.

In this paper we start by using a spatial hedonic model at the individual house level as our
reference and explore the effects that aggregation at different levels has on the estimate of
the marginal price of environmental quality. We use a large geocoded sample of over
100,000 house sales transactions for 1999 in the Los Angeles basin (Los Angeles, Orange,
Riverside and San Bernadino counties). The data we use contains information on the
house price as well as house-specific and neighborhood attributes. Environmental quality
is computed from data on pollutants observed at monitoring stations and interpolated by
means of a geostatistical kriging procedure. We also explore the effect of using alternative
random sampling schemes and the extent to which spatial autocorrelation remains a
problem in each case. Finally, we assess also the extent to which the economic
interpretation of the model is sensitive to the level of aggregation of the data as well as to
the sampling method.