u5ge91ma@WATDCS.UWATERLOO.CA (Graham Dudley) (02/07/90)
Does anyone know of any GIS which has the capability of performing multivariate cluster analysis or, contrariwise, any clustering software which has the capability of presenting results in a mapped form? Or, are there any references in the literature which discuss the use of cluster analysis in conjunction with GIS technology? Any assistance would be most appreciated. ..graham -------- Graham Dudley BITNET: U5GE91MA@WATDCS.UWaterloo.ca Department of Geography CONNECT: DUDLEYG University of Waterloo Waterloo, Ontario, Canada N2L 3G1
MARBLE@MAPVXA.CFM.OHIO-STATE.EDU (Duane Marble) (02/07/90)
Nothing in this category that I have heard of. What operations are you doing? What is being clustered? Points, Lines, Polygons ??? Duane Marble Ohio State
u5ge91ma@WATDCS.UWATERLOO.CA (Graham Dudley) (02/07/90)
At this point, the object is to use an agglomerative cluster analysis as a possible way to address the modifiable areal unit problem. As such, the objects to be clustered would be either points or small areas (e.g., census enumeration areas). Currently, I use SAS to perform the cluster analysis and then use a micro-based mapping package to present the results (this is being done in an exploratory way using census tracts as the basic statistical unit). I realize that the mapping of areal information can be done using SASGRAPH but I find this rather inflexible. I also would like to tried SYSTAT on the MAC but my version (3.2) does not allow for the importing of boundary files (contrary to what the manual says). Ideally, I would like to be able to view interactively what is happening spatially as the observations are agglomerated into fewer and fewer clusters (i.e., as the level of resolution decreases). I would also, at some point, like to apply this idea with a raster-based system as an alternative approach to performing overlay operations. ..graham -------- Graham Dudley BITNET: U5GE91MA@WATDCS.UWaterloo.ca Department of Geography CONNECT: DUDLEYG University of Waterloo Waterloo, Ontario, Canada N2L 3G1
MARBLE@MAPVXA.CFM.OHIO-STATE.EDU (Duane Marble) (02/07/90)
Although it does not do formal cluster analysis, the graphic operations you describe may be found in the GIS/Spreadsheet package offered by Don Cooke and GDT in Lyme, NH. As far as I can tell this comes as close as anything to what you describe. The incorporation of formal clustering algorithms would be interesting. Why not talk to Don about it? Duane Marble - OSU
V107REF9@UBVMS.BITNET (02/07/90)
I don't quite see the utility of using ordinary cluster analysis (in SAS) in studying modifiable areal unit problem (MAUP) because the aggregation ( or clustering ) is subject to the contiguity principle while the SAS clustering (or I think any clustering procedures) will not take into account of this principle. Maybe you have found a way to modify the clustering procedure to incorporate this constraint. If so, I am interested in knowing how to do it. I have been working on the MAUP for more than half a year and I have been following the literature quite closely. Would you mind indicating what aspects of the MAUP (e.g. scale problem or the aggregation/zoning problem) you are investigating ? I know Don Cooke, who used to working for Bureau of Census, has 'invented' something called 'Geo-spreadsheet', which allows you to aggregate areal units (both dissolving polygons and aggregating polygon attributes) interactively. However, the analytical capabilities of the software is still quite limited. David Wong Dept. of Geography SUNY Buffalo Bitnet:v107ref9@ubvms P.S. Recently, G.Arbia wrote a book on the MAUP and I think the book has significantly breakthrough in this long-lasting geographical problem.
MARBLE@MAPVXA.CFM.OHIO-STATE.EDU (Duane Marble) (02/07/90)
Cluster analysis with explicit contiguity tables has been used in geography for over twenty years! Check some of the early regionalization work by Brian Berry. This is also reported upon in several reviews of the literature at subsequent dates. Duane Marble
N050032@UNIVSCVM.BITNET (stan green) (02/09/90)
one source is Steven Savage's monograph on late archaic landscapes available from the South Carolina Institute of Archaeology and Anthropology (Studies in Anthropology) Columbia, SC 29208 approx 10 american dollars He uses MAP in conjunction with some cluster analysis
u5ge91ma@WATDCS.UWATERLOO.CA (Graham Dudley) (02/09/90)
In response to David Wong's question, the idea of using an agglomerative clustering approach is that if one can start with "non-modifiable" units (or as close as is practically possible), one can allow the data to "aggregate" themselves rather than imposing an a priori aggregation upon the data. Picking an appropriate level(s) of aggregation and mapping the clusters, one can see what, if any, spatial pattern appears at a particular scale/level of resolution. The objective is to overcome the aggregation effects and make explicit use of the scale effects (i.e., at different levels of aggregation, differing spatial arrangements may reflect processes operating at different scales). Although my own research in to this (or, more specifically, techniques which can overcome the aggregation effects and make use of the scale effects) is still at a very preliminary stage, the idea of using cluster analysis came to me as a result of looking at the retail mix of Toronto's underground pedestrian mall system (something many of you will get a chance to experience first hand in April :-) ). Using the proportion of particular types of retail establishments in each of the eight malls studied as the variables, a cluster analysis was performed (no spatial information was included in this part of the analysis). Three distinct clusters were evident as a result of the analysis and, when these clusters were mapped, a very definite spatial pattern emerged (a northern, central and southern cluster were apparent). The clusters appeared to reflect the particular market each of the individual malls served (for example, the central cluster involves some of the large of office towers in the CBD and, as such, has a higher proportion of coffee shops, restaurants, and other food establishments). Obviously this is a rather simplistic example but I found it interesting that in the absence of any explicit spatial information being included in the analysis, a very definite spatial pattern emerged. It is this sort of idea that I would like to pursue a little further. ..graham PS Thanks for the Arbia reference. It looks interesting (I hope to be able to examine it a bit more detail in the next few days). -------- Graham Dudley BITNET: U5GE91MA@WATDCS.UWaterloo.ca Department of Geography CONNECT: DUDLEYG University of Waterloo Waterloo, Ontario, Canada N2L 3G1
FRANK@MECAN1.MAINE.EDU (02/10/90)
i (sort of) remember a paper in the internat. journal for gis on a system in australia (CSIRO?) which had such a capability. andrew