The Multi-Quadrant Analysis allows analyzing a model on different subsets of data. The states of the Breakout variable are used for automatically splitting the dataset and creating one network per subset (breakout variables are usually products or geographical areas).
Since version 5.4, the local networks can be designed by:
- Keeping the structure of the original network and learning the parameters on the local datasets, or
- Learning new networks, i.e. structure plus parameters, on the local datasets.
Each network is thus analyzed with the metric that has been selected: Mutual Information, Binary Mutual Information, Total Effects, Standardized Total Effects, Direct Effects, or Standardized Direct Effects. The results on each subset can be compared via the Multi-Quadrant's graphic interface.
New Feature: Adapt Structural Coefficient
This new option allows computing a Structural Coefficient for all the sub-datasets that is equivalent to the one defined for the entire dataset. For example, if the structural coefficient is set to 1 in the original network, and if the breakout variable splits the dataset into 2 equal parts, the structural coefficient will be set to 0.5 for both sub-models.
New Feature: Overall
Prior to version 6.0, the Multi-Quadrant came with a quadrant per state of the Breakout variable. It comes now with an additional quadrant "Overall" that corresponds to the model analyzed before breaking out the dataset.