Skip to content

When are the next splits updated?

Changes to the Dimension Settings, Stopping Conditions and the Algorithm Options affect the splits displayed in the Next Splits panel.

For more details on how the splits are updated, see:

Splits for all leaf nodes need to be updated whenever any of the following options change. You have the choice to update them automatically at this time.

If you say "No" then they will be updated next time you perform an action on the tree.

Changes to dimensions

Include / Exclude

Splits are automatically calculated for leaf nodes based on all the dimensions on the Dimensions panel. The update occurs when the user changes to another tab other than the Dimensions panel.

If a dimension is not included, a split will still be calculated but its status will be set to "Excluded By User". This dimension will not be chosen by the automatic build process, but can still be chosen by the user.

As dimensions are added or removed from the Dimensions panel, splits will be calculated or removed from the Next Splits panel of all leaf nodes.

Multiple leaf nodes can be processed together and if system memory allows, all leaf nodes can be updated in a single pass through the data. The batch settings are controlled in the global options. See How do I change the default Options?.

Branch nodes are not automatically updated. To update the splits of a branch node based on the latest Dimension Settings, you will need to select to "Refresh PWE Splits" by right clicking on a node in any of the Tree Views.

Unclassified Handling

Any changes to the Unclassified Handling option will cause a new split to be automatically calculated for all leaf nodes. This happens as soon as you move away from the Dimensions Panel.

For example:

  • You may initially create splits with unclassifieds Free-Floating. In the example below, people with unclassified income have been put in to branch 1 along with the middle income bands.

  • You may then decide to omit unclassified altogether. The split will be recreated. In the example below, the grouping of the other categories has not been affected. This is not always the case.

Branch nodes are not automatically updated. To update the splits of a branch node based on the latest Dimension Settings, you will need to select to "Refresh PWE Splits" by right clicking on a node in any of the Tree Views.

Selector Branches

Any changes to the Selector Branches option will cause a new split to be automatically calculated for all leaf nodes. This happens as soon as you move away from the Dimensions Panel.

For example:

  • If you change the settings for the Income dimension to use Ordered, a split is created with incomes below £20k together and those above £20k together.

  • The colours of the individual categories indicate that this is not a good idea. The top incomes are coloured blue (as they have a low analysis %) but they have been allocated to branch (1) which is coloured red (indicating that the average analysis % is high for this branch).

Branch nodes are not automatically updated. To update the splits of a branch node based on the latest Dimension Settings, you will need to select to "Refresh PWE Splits" by right clicking on a node in any of the Tree Views.

There is a Known Issue, that the split used to create the child nodes will also be updated and the new split will be displayed in the Next Split panel. However, the structure of the displayed tree is still based on the old split.

Changes to stopping conditions

The Status and Reason if Invalid columns in the Next Splits panel are affected by the Split Validity Stopping Conditions. Changes are applied to all leaf nodes as soon as the user clicks OK or Apply.

In the example below, the Significance has been increased to 4.00. As a result the split on Occupation has been made invalid.

However, any splits for nodes that have already been split will not be affected. If they were valid at the time of being split, they will still display as valid.

Changes to algorithm options

Changes to the Algorithm Options are applied to all leaf nodes as soon as the user clicks OK or Apply.

The Split Selection Measure affects the ranking of the splits. For example:

  • If the Split Selection Measure is set to "Overall Significance", the rank order of the splits corresponds to decreasing Significance values.

  • If the Split Selection Measure is set to "Best Variable", the rank order of the splits corresponds to decreasing Cramer's V values.

Changes to the Default Split Creation Strategy will result in a new set of splits being created for all leaf nodes.

  • These first splits were created using the PWE method (set to Binary)

  • Changing the default method to Chaid, results in a new set of Chaid splits being created

Note that any split which has been changed by the user from the default will not be updated. These splits are shown with the How Created column in bold.

  • Here the Year-Of-Birth split has been amended manually so that it is now valid

  • If you now change the default strategy back to PWE, the Year-Of-Birth split is unaffected