Editorial

Abstract

Introduction

Software

Software Evaluations

Assessment

Summary
  Functionality
  Ease of Use
  Implementation

Conclusions

Acknowledgements

References

Case Studies Index

5. Summary and discussion

5.1 Functionality of software

As can be seen he packages can be divided into two groups in terms of the focus of the functionality provided.

Both cdv and MANET have an emphasis on graphical facilities, rather than numerical ones. They are both extremely flexible in the way that the appearance of graphs and maps can be altered. For example, when a point symbol map is produced in cdv, the size of the symbols used to represent the data can be altered by a simple slider bar. Not only is this an extremely simple and easy to use technique, it works quickly enough that it is possible to experiment with the effect of different sized symbols on the appearance of the map, and a suitable size can very easily be determined. This can be compared with a typical Windows mapping package, such as MapInfo, in which the same operation requires the use of a menu panel in which the size is specified numerically. The effect can only be seen by entering the new value and returning to the map, so that adjustments to the size requires a constant switching between these two windows.

MANET has a similar facility with histograms, where the number and size of the histogram bins can be altered dynamically. Again, this is superb as a tool for demonstrating the effect of changing these parameters on the histogram and would make a good teaching tool. MANET also has an excellent missing data facility and provides state of the art visualization tools such as weighted histograms, trellis displays, mosiac plots and the use of brushing and tracing.

There are situations however, where this emphasis on the graphical is more open to debate. For example, neither package appears to put labels on the axis of scatterplots, dotplots or histograms. In this sense, these graphs are unsuited to table look-up, one of Cleveland's two uses for graphical displays. It is possible to discover the value of any particular data point, in cdv at least, by clicking on that point in the graph. However, it would seem that one of the uses of graphs such as dotplots is to indicate the numerical range in the values of a variable, as well as illustrating the shape of the distribution and the presence of outliers.

Table 5: Summary of types of ESDA functionality in the four packages

This emphasis on graphical display in cdv and MANET leads these packages to provide very little in the way of numerical analysis measures. In contrast, SAGE and SpaceSat provide a much richer range of numerical methods, but with correspondingly weaker graphical facilities. SpaceStat itself has no graphics of course, but uses ArcView as a tool for exploring the data and results of analysis. In SAGE the graphical tools are integrated into the system, but ARC/INFO is used for map drawing. The contrast with cdv and MANET illustrates the difference between the traditional world of the paper map and the new one of interactive visualization. Whereas the colour symbolism of a choropleth map can be changed easily and quickly in both cdv and MANET, it is a very cumbersome process in SAGE. What is more it is impossible to get SAGE to automatically generate a greyscale which has the correct number of shades for the dataset being mapped (which happens in cdv and MANET).

All the packages provide facilities for the exploration of non-spatial elements of the data but again the two groups of packages differ in their provision of tools for exploring the spatial properties of the data - the identification of spatial trends (representing the smooth element of the data) or the presence of spatial outliers (the rough). SAGE and SpaceStat provide specialised graphs (such as the Moran scatterplot and lagged boxplot) and numerical measures (such as local autocorrelation statistics) for this purpose, whereas cdv and MANET rely on the use of the map and its link to other graphs

Finally, SAGE is unique in providing facilities for the manipulation of the areal units themselves, through the use of a regionalisation facility.

5.2 Ease of Use

It was suggested at the beginning of this report that graphical tools might be intuitively easier to use for those unfamiliar with spatial data. While this may be true, it is clear that none of these software packages can be used without the investment of considerable time. The problems lie in several areas:

There is an inherent problem in any menu-based system, that the menus are rarely self-explanatory, and must be learnt. Although the use of Windows style menu interfaces has provided a uniform look to software interfaces, so that the mechanics of making selections, setting parameters, opening files etc can be learnt once and applied to several packages, the logic of what the menus mean still has to be learnt. A feature of software for spatial data (including mapping packages and GIS), which sets it apart from many other software packages is that there is a definite logic to the sequence in which menus should be used. The clearest example in the case of ESDA is that you must create a W matrix before some other functions can be used - SAGE does this automatically, but in SpaceStat it must be done explicitly.

Some of the visualization techniques used would require some explanation to those unfamiliar with computer graphics. For example, cdv's plotting of multivariate maps by assigning each variable to one of Red, Green or Blue is a technique which has been tried by various cartographers (it is used in Dorling's census atlas for example (Dorling 1995) but which can produce a map which is hard to interpret. There are several difficulties here:

• The idea of a single `colour' being used to represent two variables simultaneously will certainly be a new concept to many people familiar with viewing traditional choropleth maps. In paper maps this is tackled by the careful design of a legend, and something similar is attempted in cdv, with an optional, pop-up legend.
• RGB colour space is known to be non-intuitive - for example people more familiar with colour mixing using paints would not expect Red and Green to combine to produce yellow.
• Even when the viewer has grasped the previous two concepts, the map may be hard to read. The extreme ends of the colour spectrum will be easy to interpret - vivid red clearly means one variable is at a maximum, while the other is at a minimum. However, it is far harder to assign a clear meaning to shades in the middle of the spectrum.

There were other examples in the various packages where the meaning of some of the graphs and charts was not clear. Overall, we had the strong feeling that work on these packages had concentrated on exploring what was technically possible. Now that working systems have been developed, we feel there is a need for research into just how effective these tools are for general users.

5.3 Implementation

The packages varied somewhat in the ease with which they could be implemented. Given that all four are the product of academic researchers, rather than professional software vendors, it would be unfair to be over-critical in this area, but if JISC decided that there was a case for encouraging the use of this type of software, then this is an area which would require attention.