Decision Tree: How do I interpret the Organic Tree?

The Organic Tree provides an overview of the statistical relationships between nodes.  In later releases there will be flexibility about how the tree is displayed.

At present the following applies:

  • The width of a branch is proportional to the number of people in that node.

  •  The angle of a branch indicates how significant the difference in Analysis % within the branch is compared to its parent.

    A branch almost in line with its parent, does not differ that significantly in Analysis %

    A branch sticking out at a large angle, shows a significant difference in Analysis %

  • What do all the colours mean? The colours are as explained in the section:

    Grey - node Analysis % similar to that of the root node

    Red - node Analysis % greater than that of the root node

    Blue - node Analysis % less than that of the root node

Some examples of Organic Trees

Significant but low gain

In the above example most of the splits are highly significant (large angles), but are only resulting in small changes to the Analysis % (colours are muted or grey).  This is because the root node has a very high Analysis %, of 82%.  This has two consequences:

  • Changes in Analysis % are never very large.

    The best node, with an Analysis % of 96%, still only has a gain of 1.17 and so is only lightly shaded.  Similarly, the worst node, with an Analysis % of 55%, has a gain of 0.67, which is still close to 1.0 and so very lightly coloured.

  • Changes in Analysis % are, however, very significant.

    Since changes in Analysis % are based on large samples they are typically very significant.  Node 1 (shaded grey) represents a change in Analysis % from 82% to 90%, which has a Z-Score of 142.

High Gain but not Significant

In the above example most of the splits represent large changes to the Analysis % (bold colours), which are not very significant (small angles).  This is because the root node has a very low Analysis %, of 0.0005% (10 out of 2 million!).  This has two consequences:

  • Changes in Analysis % are typically very large.

    The best node, with an Analysis % of 50% (2 people out of 4), has a gain of almost 100,000, and so is very intense red.  Similarly, the worst nodes, have an Analysis % of 0%, have a gain of 0, and so are very intense blue.

  • Changes in Analysis % are, however, not very significant.

    Since changes in Analysis % are based on small counts, and are changes to what is initially a very low percentage, they are typically of very low significance.  Node 1 (shaded light red) represents a change in Analysis % from 0.0005% to 0.001%, has a Z-Score of 2.48.

    In statistical terms this is officially classed as significant, but that is mainly due to the very large samples available in marketing data.  However, dealing with counts of 10 in a few million, although statistically classed as significant, is not likely to have business significance!