Spatial econometrics functions in R: Classes and methods (38)

Theme Track: Methods of Spatial Analysis - Exploratory Spatial Data Analysis

Author:
Bivand, Roger

Developments in the R implementation of the S data analysis language are providing new and effective tools needed for implementing functions for spatial analysis. The release of R packages for constructing and manipulating spatial weights, and for testing for global and local dependence during 2001 has been followed by work on functions for spatial econometrics. The paper gives an introduction to classes in R, to the use of object attributes, and to class-based method despatch. These features are important because they encapsulate information about the data in a predictable way, also when the data is for example a model formula. This permits the flexible handling of subsetting, missing data, dummy variables, and other issues, based on existing classes that are extended to handle spatial econometrics functions. For the user, it is convenient if generic access functions can be applied to spatial analysis classes, such as making a summary or plotting a spatial neighbours structure. The same applies to the use of equivalent model formulae, describing the model to be estimated, for a range of estimating functions. In this setting, a spatial linear model should build on the classes of the arguments of the underlying linear model. There should be no difference in the syntax of shared arguments between the aspatial linear model, spatial econometrics models, or geographically weighted regression models, although of course function-specific arguments should be introduced. It is also of interest to compare spatial econometric formulations with other related model structures, such as those for mixed effects models, and to explore other alternative approaches. These may include extensions to repeated measurements, to spatial time series, and to generalised linear models, although here the spatial case is often currently unresolved in terms of choice of methods. However, the underlying classes are important in that their implementation may make the flexible extension of spatial analysis tools more or less difficult, and consequently that they should admit the quick prototyping of experimental new modelling techniques rather than hinder it.

[Home]    [Sitemap]