17. Options

This menu sets a number of SPAN options. The options are retained and written as a logical sequence in the registration file register.txt.

Boolean expressions in SPAN are automatically simplified according to attributes in a Boolean expression. The simplification does not take account of the nature of variables from which attributes are constructed unless this item is checked. For example the expressionX₁ÇX₁ÇX₁ÇX₂ reduces to X_1∩X₂ regardless of the nature of the way the attributes are constructed. If, however, it is known that X₁ and X₂ are attributes derived from some underlying nominal variable, x, such that X₁ = {x = 1} and X₂ = {x = 2} then X₁ÇX₁ÇX₁ÇX₂ is the null set.

Full Boolean reductions are those simplifications to Boolean expressions that arise because of the way attributes are constructed, as in the above example. These reductions may arise if several attributes are created from a variable either by specifying a series of cuts or by repeat lines in the control file. This option will simplify all such Boolean expressions. More details are in Appendix A5.4.

If this option is in place no simplification of Boolean expressions is done.

The option would not normally be used, but it is useful, for example, to investigate the relationships between attributes using subgroup rectangle diagrams. For example if  you wanted to display rectangles at different cutoffs of systolic blood pressure:

A=(sbp>140) or (sbp>160) or (sbp>180)

which would otherwise simplify to (sbp>40).

Partitions are tied if they have the same complexity and effectiveness. With this option the last found tied partition during a search supplants the existing one.

Missing attributes arise from missing values of variables used to create the attributes (see 8.5). They can be handled in three ways when evaluating a partition:

1.    Skip if any missing in attribute set 

If any attribute is missing in the defined attribute set, the entire observation is skipped (ignored). With this option the sample size for a partition is the same as for any other partition derived from the attribute set.

2.    Skip if any missing in combination 

If any attribute is missing in a defined Boolean combination, the observation is skipped. With this option sample sizes for one partition may differ from that of another, depending on the pattern of missing observations.

3.    Skip if combination not evaluable 

Sometimes it is possible to evaluate a Boolean expression even if some of its attributes are missing. For example, suppose you need to determine whether an observation lies inA = X₁∪X₂. If X₂ is not missing then knowledge of the value of X₁ is immaterial when X₂is present. 

Whether a Boolean expression can be evaluated when some of its constituent attributes are missing, will thus depend on the value of the attributes which are not missing. Another example is A = (X_1∩X₂)∪(X_3∩X₄). A would be present if X_1∩X₂ occurred, regardless of whether X₃ and X₄ were missing, but its value would be indeterminate if X₁was present and X₂ absent with X₃ and X₄ missing. 

The third strategy has the advantage that it may utilise more information, but should be used warily if missing values do not occur at random and are closely associated with the dependent variable(s).

When a partition is accessed from the Process menu, a message is sent informing of the number of missing observations due to missing attributes.

This option sends fuller details of a SPAN session to the output log. Usually you would not want this option, since the output log file can become very large.

If this option is checked points on the complexity hull are added as they are generated by the search. The option may slow down the search and is best not used.

[Back to table of contents]