This term is used in two quite different ways. One is in relation to subdividing data according to the relationship between a dependent variable and explanatory variables and the other structures the analysis of choosing between various courses of action.
This was called Automatic Interaction Detection for a long time and now also goes under various names used by software vendors, including Regression Tree, Answer Tree, Classification Tree and CART. It is a technique frequently employed in Data Mining and it is a useful exploratory analysis technique prior to regression analysis. It can quickly analyse a large set of candidate explanatory variables to determine the most influential variables on the dependent variable.
The basic idea is to hierarchically segment the population
on the database based on a dependent (categorical) variable such as bought/did
not buy a product. The explanatory
variables are categorical too, such as:
Decision Tree Analysis is useful in choosing between alternatives when there are uncertain consequences. This technique can also work out how much it is worth to collect more information that reduces uncertainty.
Decision Tree Analysis is based on:
The expected monetary value (EMV) is computed for each strategy alternative. The EMV is a weighted average of the possible monetary values weighted by their probabilities.
A decision tree is a convenient way to set out the problem, perform the computations, and undertake sensitivity analysis.