Please consult this list of frequently asked questions before contacting SpaceStat support or requesting further information.
In order to run SpaceStat, does one need other software (Gauss, ArcView)?
No, SpaceStat is completely self-contained. The SpaceStat Extension requires ArcView 3.x, but SpaceStat itself does not.
Does SpaceStat interface with GIS software other than ArcView (MapInfo, Idrisi)?
Partially. The old interfaces with Idrisi and GisPlus are still in SpaceStat, but they are not current for the latest version of those GIS.
However, SpaceStat report files are flat ASCII files (comma delimited or tab delimited) that can be imported in a wide range of other software, including most GIS, spreadsheet and statistical programs. The difference with the ArcView interface is that you will have to do the work to import, join, and otherwise manipulate the SpaceStat report files, whereas in the ArcView interface this is done for you in a seamless manner.
Also, SpaceStat has functionality to construct a number of spatial characteristics (such as contiguity, centroids, area, etc.) from generic ASCII boundary files (BND files in SpaceStat). To the extent that many GIS can create or export such files, SpaceStat can take this information and manipulate it to carry out spatial data analysis.
Does SpaceStat run in Windows?
Yes, SpaceStat runs as a DOS window in Win95/98, Windows NT, and Windows 2000. Versions prior to 1.91 do not run in Win2000.
Is there a SpaceStat version for the Mac, for Unix?
No, there is not (yet?). So far, SpaceStat is purely a 32 bit DOS product. Also, since the SpaceStat Extension for ArcView relies on dynamically linked libraries, it only works in a MS Windows environment.
Does SpaceStat have geostatistical functionality?
No, there is currently no geostatistical analysis in SpaceStat (in the narrow sense of the word, following the definition in Cressie 1993). The focus of SpaceStat is on lattice data.
Can SpaceStat analyze point pattern data?
Yes, to the extent that the point locations are taken as given and thus for all practical purposes can be treated as lattice data. Spatial weights can be constructed based on the distance between points. The only information needed by SpaceStat is the point coordinates. However, SpaceStat does not contain statistics for the analysis of point clusters or similar "traditional" point pattern analysis.
Can SpaceStat analyze space-time dependence?
Not directly, but with some work a limited form of space-time analysis can be implemented, assuming that the same spatial process applies throughout. This is accomplished by treating the space time data as different "regimes", with their own coefficient values (including different error variances by regime). The estimation of models with both spatial error dependence and spatial lag dependence includes groupwise heteroskedasticity and/or regimes. However, the space-time techniques discussed in Chapter 10 of Anselin 1988 Spatial Econometrics are not included explicitly.
Can SpaceStat analyze limited dependent variables with spatial autocorrelation?
Apart from the join count statistics there is no explicit treatment of limited dependent variables with spatial autocorrelation. There are several data transformation methods included in SpaceStat that may yield acceptable results (particularly for proportion data), but there are no routines for the estimation of regression models with spatial dependence for count data or binary data. Note that this is still very much uncharted terrain and no "out of the box" solutions currently exist. Also, there is no such thing as a logit model with spatial autocorrelation.
What is the largest number of variables a SpaceStat data set can contain?
The number of variables depends on the available memory. Typically, this is not a binding constraint, but the number of observations is.
How does SpaceStat handle large data sets?
The binding constraint in most spatial data analyses is the number of observations, since a spatial weights matrix has a dimension N by N, where N is the number of observations. Since all matrix elements are stored in double precision, this quickly gets out of hand (e.g., with each matrix element taking 8 bytes, a data set with 1,000 observations would require roughly 8 megabytes for the spatial weights matrix). Typically, the available memory is what determines how large a data set can be handled, although numerical precision must be considered as well.
SpaceStat uses sparse formats to store and manipulate spatial weights for all test statistics and for all estimation except maximum likelihood (ML). ML estimation of spatial lag and spatial error models uses an eigenvalue based method to compute the likelihood, which has doubtful numerical precision beyond 1,000 observations or so (depending on the type and structure of the weights matrix). However, estimation based on instrumental variables and generalized moments uses the sparse weights and can be carried out for very large data sets (several thousands of observations). Since these are large sample methods, they are particularly suited to handle this problem.
Note that the minimum size for a data set is 20 observations. Since most spatial data analysis methods are large sample techniques, smaller data sets are not allowed.
Can SpaceStat estimate regression models that contain both a spatial lag and a spatial autoregressive error term?
Maximum likelihood estimation of this model is not currently implemented. However, by specifying lagged explanatory variables (WX) as instruments for the spatial lag in a standard 2SLS estimation, Version 1.90 allows for the estimation of this specification by means of Kelejian and Prucha's generalized spatial two stage least squares method.
Does SpaceStat support regression models with multiple spatial weights?
Maximum likelihood estimation of such models is not supported, but by a judicious use of instrumental variables (WX) such models can be estimated. However, one should be very careful to avoid identification problems when the spatial weights defining the lags show overlap. This should not be a problem when different orders of contiguity are used.