Understanding interactions of explanatory variables
for some variables of interest
Decision Tree Analysis is an exploratory method which explains some variable of interest without using a strict regression model.
It falls half-way between the categories of Segmentation and Key Drivers Analysis.
It is designed primarily for situations where:
- The variable of interest is a two category (Yes / No) or multi-category (choose one option from 3+) variable.
- Most of the potential predictors are categorical, rather than numeric scales.
- We are interested in understanding the interactions between predictors, rather than their overall effects.
- A simple model is required based on observed data, rather than a regression model.
- We want to identify small groups in the population, which contain a high proportion of cases, belonging to some category of interest.
The most well-known form of decision tree analysis is known as CHAID (Chi-squared Automatic Interaction Detector). Predictors are built up in layers in a tree-like fashion with a root, nodes and branches. For multi-category predictors, CHAID creates optimally merged categories in order to simplify the resulting model and reduce the number of branches.
These models are often useful in their own right. For example, “What combination and interaction of demographic category typifies someone who buys brand X?”. They are also useful as an exploratory lead-in for other methods to help identify which categories of potential predictor should be merged and combined in, say, a Drivers Analysis.
Case Studies
-
CHAID Public Opinion Pollster
One example is an ongoing CHAID analyses for an public opinion pollster to understand what…
-
Diabetes CHAID Analysis
A published study among Type 1 diabetes patients was based on our CHAID analaysis. We…
-
Employment Tribunals
We have worked on several CHAID analyses to determine the combination of features of employment…
-
Sports Betting Ordinal CHAID
We undertook an Ordinal CHAID Analysis for a leading sports betting chain to determine what…
-
Maternity Leave Discrimination
By performing a CHAID analysis, we were able to determine which combination of features, profiles…
-
Ethnic Minority Voting
By looking at the CHAID analysis of ethnic minority BAME voters, we were able to…
Key Factors
- Identify small groups in the Population which account for high proportions of behaviour or interest
- Intuitive segments defined by combinations of existing variables
- Optimal merges and combinations
- Ideal for categorical data (e.g. demographics)
