Spatial Data Analysis

Spatial data analysis studies data that carry a geographic location, focusing not only on what happens and how much, but on where it happens and how places relate to one another. Its distinctive feature is that location matters: unlike conventional statistics, it does not treat observations as independent of their position.

Spatial autocorrelation

The concept at its core is spatial autocorrelation: the degree to which the values of nearby places resemble one another. It is summarized in the so-called first law of geography, stated by Waldo Tobler: “everything is related to everything else, but near things are more related than distant things.” Autocorrelation is positive when similar values cluster together, negative when they alternate like a chessboard, and null when the arrangement is random. The most common measure for areal data is Moran's I.

Concepts of its own

Spatial analysis works with geographic units—regions or polygons such as municipalities or census tracts—and with a spatial weights matrix that encodes which units are neighbors, for example whether they share a border. A classic problem is the modifiable areal unit problem (MAUP): results change depending on how the zones are divided or aggregated, both when their size and their shape vary. And it helps to distinguish the data types: points (specific locations), areas (values aggregated by zone) and continuous surfaces estimated by interpolation.

Not the same as a GIS

Spatial analysis should not be confused with a geographic information system (GIS). The GIS is the tool or software platform for storing, managing and visualizing geographic data; spatial analysis is the set of statistical and modeling methods applied to those data to discover patterns and relationships. In a sentence: the GIS is where the data are stored and drawn; spatial analysis is how they are analyzed and what is concluded.