SPAN is a program to implement Search Partition ANalysis, as described in references one, two and seven. It also implements various other statistical procedures, including ROC curves, tree and simple descriptive statistics. The objective of search partition analysis is to obtain binary rules that partition data into two groups such that each is relatively homogeneous with respect to an outcome variable. The rules are defined in terms of logical Boolean expressions. A search is done by sifting through all Boolean expressions of a certain type.

As partition rules are defined logically, the method has closer similarities to tree growing than to (arithmetic) linear discriminant analysis procedures. However, in contrast to tree-growing, SPAN does a global search for a partition rather than successively splitting data. There is, however, a basic tree-growing facility in the program.

The program runs under WINDOWS systems. There is no Mac version. Features of the program include:

• graphics for exploratory data analysis - bar charts, empirical density estimation, distribution and survival plots, scatter diagrams
• boolean algebraic manipulation
• complexity penalised search partition analysis
• selection of optimal cut-points, ROC and QROC curves
• different criteria for partition effectiveness (mean square error, entropy, Gini, quality indices, log-rank, chi-square)
• construction of classification/regression trees
• online helps and manual Bitmapped graphics files
• cross-validation and training and test samples
• descriptive statistics, scaled rectangle diagrams, mosaic plots
• randomisation tests of partitions.

[Back to table of contents]