Geostatistics - A Primer of GIS: Fundamental Geographic and Cartographic Concepts - page 272

everything else, but near things are more related than distant things”
(Tobler, 1970). A basic principle of “classical” statistics is that data are ran-
domly distributed—for example, the ages of people in a city have nothing to
do with the city (“classical” distinguishes the statistical concepts and tech-
niques discussed here from Bayesian statistics). Obviously, geographical data
are affected, sometimes even determined, by their location. Amphibians
f lourish near water bodies, schools are located near where children live,
stores are accessible to shoppers, and traffic jams occur on busy roads. The
concept of Tobler's first law may seem blatantly obvious to geographically
minded people, yet it is completely at odds with the principle of “classical”
statistics. (“Classical” distinguishes these from Bayesian statistics.) Geo-
graphical things and events can sometimes be more complicated. For exam-
ple, infectious diseases can spread by plane travel across oceans. Geogra-
phers call this “jumping scale” because the infection, which normally
spreads at a local scale, “jumps” globally and becomes active at the local
scale again.
Following the concept of spatial autocorrelation, it is easy to grasp that
samples—for example, measurements of temperature or anything else—
cannot reflect a completely uninfluenced area, where the measured charac-
teristics follow a random distribution. For example, let's assume that most
frogs prefer a partly wet habitat. If a researcher is determining the impact of
household pesticide use on the local frog population, it makes the most
sense to collect frogs near water bodies in low-lying wet areas. However, that
nationalatlas.gov offers an interactive mapping application—here, the per capita
number of jobs for a portion of the country is shown.
From www.nationalatlas.gov/natlas/Natlasstart.asp