Editorial

Abstract

Introduction

The Projection Pursuit

RADVIZ

Parallel Coordinates

User Interaction

Finding Outliers

Conclusions

References

Appendix A - Datasets

Appendix B

Case Studies Index

5. Making Use of User Interaction

All of the above methods may be enhanced by the introduction of user interaction. In particular, the use of linked plots, where the output from any of these techniques could be linked to another view of the data, using the technique of data brushing - see for example Tierney (1990). Two particularly useful techniques are those of linked maps and slicing. The first of these is documented in Brunsdon and Charlton (1996), and in short involves highlighting zones on a map corresponding to selected points on a scatterplot, or vice versa. This is particularly useful if one of the projection-based techniques is used. For example, one can check whether the spur in the minimising MNND projection pursuit corresponds to any particular geographical pattern, as in the screenshot in figure 12.

Figure 12: A Map-Linking Screenshot

Another useful interactive approach is slicing. In this case, points in a scatterplot are selected according to the value of an auxiliary variable. This value is controlled by a slider button, as in the second screenshot, figure 13. The value shown in the slider is the central point of a decile ´slice' of the data, based on the values of the variable LLTI - clearly any other variable could also be used. Moving the slider causes the highlighted points in the scatterplot to change - so one can see which regions of the projection correspond to high and low values of the slicing variable. This method helps to interpret the patterns seen in projection-based methods such as projection pursuit and RADVIZ.

Figure 13: A Slicing Screenshot