The locational attributes of spatial data (i.e., for the settlements, households, regions, etc.) are formally expressed by means
of the geometric features of points, lines or areal units (polygons) in a plane, or, less frequently, on a surface. This spatial referencing
of observations is also the salient feature of a Geographic Information System (GIS), which makes it a natural tool to aid in the
analysis of spatial data. I return to this issue in more detail below.
The crucial role of location for spatial data, both in an absolute sense (coordinates) and in a relative sense (spatial
arrangement, distance) has major implications for the way in which they should be treated in statistical analysis, as discussed in detail
in Anselin (1990a). Indeed, location gives rise to two classes of socalled spatial effects: spatial dependence and spatial heterogeneity.
The first, often also referred to as spatial autocorrelation or spatial association, follows directly from Tobler’s (1979) First Law of
Geography, according to which "everything is related to everything else, but near things are more related than distant things." As a
consequence, similar values for a variable will tend to occur in nearby locations, leading to spatial clusters. For example, a high crime
neighborhood in an inner city will often be surrounded by other high crime areas, or a low income county in a remote region may be
neighboring other low income counties. This spatial clustering implies that many samples of geographical data will no longer satisfy
the usual statistical assumption of independence of observations.