Panel Data Econometrics with R

Chapter 10Spatial Panels

10.1 Spatial Correlation

If the cross‐sectional dimension of a dataset has any form of ordering, or if a distance is defined over each pair of observations (here: spatial units), one can use spatial methods to account for the possibility that correlation be stronger between “nearby” ones. The most commonly used definitions of proximity are either distance‐ or neighborhood‐related. Neighborhood depends on the spatial units being arranged in a topological space on a regular or irregular grid, an example of the latter being state or regional borders in geography.¹ On the subject, see Anselin (1988, Ch. 3).

This subject is most relevant in nonrandom samples such as countries within a geographical region, or regions within one country; but spatial methods can also be employed wherever some kind of distance between observations is defined, be it in a geographic space or perhaps in an economic, demographic, or psychological one. Hence spatial methods, although more common in the former context, can be relevant in random samples too, such as, e.g., in household surveys.

10.1.1 Visual Assessment

Correlation in bidimensional space can be multifaceted, and in some ways more complicated to assess than correlation in time, which has a single dimension and often an obvious direction. Therefore, preliminary data analysis based on visual assessments, while always important and perhaps underutilized in econometric practice (Kleiber and Zeileis, 2008), is all the more ...