Part 2. MM4XL Tools > 2. Analytical Tools > Segmentation Tree > Technicalities

Segmentation Tree


The easiest way to figure out how Segmentation Tree works is to think about regression analysis. Say we were interested in finding which factors have the strongest relationship on the dependent variable Purchase of ice cream (called Criteria in Segmentation Tree) and we used 3 independent variables in the model: air temperature, availability of money, sense of craving. With multiple regression we can measure the degree and form of relationship among variables. With Segmentation Tree we identify groups of similar cases in terms of relationship to the dependent variable, and they are shown graphically in a tree diagram.

Segmentation Tree applies Belsons segmentation method, which is the one that drove the development of AID (automatic interaction detection) methods. The method iteratively splits sample data in two branches for each variable and it finds the highest discriminant value, which stands for the strongest relationship between the chosen criterion and the sampled data. The procedure loops until all cases have been assigned to one branch.

Segmentation trees are heuristic models designed for finding homogeneous subgroups in sample data. The starting point of the analysis is the selection of the independent (or passive) variables and the criterion of the segmentation. In the previous example the criterion was Buyer of vitamins and the passive variables were Gender, Area, Kids, and Age. There is no one standard method for selecting the relevant variables of a model. Some researchers use regression and correlation, but these are not always applicable techniques. Experience and taste play an important role when defining segmentation models.

Segmentation techniques require large samples for reaching useful conclusions. When the segmentation is run, what makes a variable important is the strength of its relationship (level in the tree) and the number of cases it covers.

Lifetime license:

Price: euro

Vote this tool
We proudly serve
Your vote
vote1 vote2 vote3 vote4 vote5