The effects of land uses on residential property values are crucial when ev
aluating costs and benefits of land projects for the purpose of public poli
cy prescription or business decision making. It is widely recognized that a
nuclear plant or a prison, for example, may often have an adverse effect o
n the property values of the nearby houses, while a park, a museum, or a un
iversity usually has a beneficial effect. The effect of a land use defined
as a function of distance between the locations of the land use factor and
a particular house is, however, inherently nonlinear (in an unknown form) a
nd the use of a simple linear regression method could lead to a misleading
conclusion. The purpose of this paper is to estimate the land use effect fu
nction by using the recently developed techniques of the nonparametric regr
ession method. There are three important features of our statistical model.
First, it is a semiparametric model, which keeps a conventional linear for
m with respect to the dwelling attributes of the house just like in the pop
ular hedonic model, but treats its location characteristics in a nonparamet
ric fashion using the kernel method (Robinson 1988). Second, unlike the usu
al nonparametric regression, it keeps additive structure in the nonparametr
ic component (Hastie and Tibshirani 1986, 1990), so that it retains much of
the interpretative features of the linear models. Third, it uses the local
linear smoother developed by Fan (1992, 1996), which is superior to other
smoothers in terms of avoiding the boundary effect and other features. We e
stimate the effects of three land use factors: (1) golf courses, (2) a univ
ersity, and (3) a nitrogen plant on the neighborhood home values in Lawrenc
e, Kansas. The data on the sales price and other attributes of the house wi
th 6,400 observations over the period from 1986 to 1995 are obtained from t
he Douglas County Appraisal Office and the data on distance to the three si
tes above are constructed using the Geographic Information System (GIS).