The public domain Spatial Statistics Toolbox for Matlab 1.1, 2.0, and Spacestatpack for Fortran 90 excel at estimating large-scale lattice models. The Matlab Spatial Statistics Toolbox includes code for simultaneous spatial autoregressions (SAR), conditional spatial autoregressions (CAR), and mixed regressive spatially autoregressive (MRSA) models. In addition, it contains code for creating sparse spatial weight matrices and finding the log-determinants (needed for maximum likelihood). Hence, the Matlab Spatial Statistics Toolbox includes the most common estimators employed in spatial econometrics. These products use sparse matrices and other computational techniques to greatly accelerate computations and to expand the size of potential data sets analyzed.
Spatial Statistics Toolbox for Matlab 2.0 (zip)
The package also contains some spatiotemporal data from Baton Rouge. It currently occupies over 8 MB, so you may wish to time your download strategically. Also, you can download this package, other packages, and the articles via anonymous FTP (ftp.spatial-statistics.com/Spatial_Statistics_Toolbox or ftp.spatiotemporal.com). Some FTP clients perform downloading much better than browsers. For example, WS_FTP, CuteFTP, and FTP Explorer allow resumption of interrupted transfers and contain other features that make them ideal for downloading large files over the net.
If you wish to examine the package without downloading the entire file, you can download the documentation separately (pdf format) :
Spatial Statistics Toolbox for Matlab 2.0 - Documentation Only (pdf)
Spatial Statistics Toolbox for Matlab 1.1 (zip)
The package also contains matlab spatial data files as well as pdf versions of some of my relevant articles. In terms of data, it contains a 506 observation spatial data set pertaining to pollution, a 3,107 observation spatial data set pertaining to the US presidential election turnout, and a 20,640 observation data set pertaining to California housing values. These routines are separate from those in version 2.0. I plan to revise these slightly. The fdelw2 routine sometimes fails with Matlab 6.0+, and this has been remedied in version 2.0.
If you wish to examine the package without downloading the entire file, you can download the documentation separately (pdf format) :
Spatial Statistics Toolbox for Matlab 1.1 - Documentation Only (pdf)
I also wrote in Fortran 90 a subset of these functions of the Matlab Spatial Statistics toolbox. The Fortran 90 SpaceStatPack includes PC executable code. It contains the same approximation routine described above for the log-determinant of the variance-covariance matrix. This algorithm that gives it incredible speed relative to conventional techniques. As an example, for the Geographical Analysis article data with 3,107 observations, the routines can find the nearest neighbors, compute the approximation to the log-determinant (which has confidence intervals quantifying the accuracy of the approximation), and calculate the maximum likelihood estimates in under 3 seconds on a Pentium III 500 mhz computer.
SpaceStatPack 1.0 in Fortran 90 (zip)
Again, you can just looking at the documentation (in pdf format).
SpaceStatPack 1.0 in Fortran 90 - Documentation Only (pdf)
Despite having touted the capabilities of SpaceStatPack, the Matlab Spatial Statistics Toolbox offers an easier-to-use, more flexible means of working with spatial statistics. I would urge all those who can obtain Matlab (students can obtain it inexpensively) to go this route. However, SpaceStatPack (including its component modules) offers a route for those who prefer Fortran or cannot obtain Matlab.
I have Fortran 90 software (source code and PC executable code) for computing spatial-temporal weight matrices (nearest neighbors subject to non-zeros only previous observations). This appears on Fortran-2000.com .This implements the computationally difficult part of the spatial-temporal estimation method discussed in:
Pace, R. Kelley and Ronald Barry, O.W. Gilley, C.F.
Sirmans, “A Method for Spatial-temporal Forecasting with an Application to Real
Estate Prices,” International Journal of
Forecasting, Volume 16, Number 2, April-June
2000, p. 229-246.
It requires less than 8 minutes to find all the neighbors for 100,000 observations on a 600 Mhz PC.
Home Articles Software Manuscripts Links Search