Library
I was curious about the different applications of Multi Quadrant Analysis in BayesiaLab. In the White Paper on Product Optimization you discuss its application using product type as the selector variable. Extending the example it seems you might also apply the same logic to different consumer or demographic groups. Which is to say, using pre-defined segments as the selector variable to understand the how the importance of different variables in the network (wrt a given target node) vary across these groups.As I understand it the underlying networks for the different states of the selector variable are identical, which implicitly assumes that there would be no difference in the networks learned from data defined by these subsets. Nonetheless, it seems an interesting application. Thoughts?
Quote 0 0
Library
Indeed, you can use any breakout variables in the Multi Quadrant Analysis. Whereas the structures generated for each state of the selector variable will be identical, the parameters will be estimated by using the data subsets defined by each breakout state. The underlying hypothesis is to consider that the structure learned on the entire data set reflects the “Total Market”, and can then be applied to all the segments.If you think this hypothesis does not hold for your problem, you will have to learn one network (structure + parameters) per segment. However, when the data subsets are too small, it can be difficult to find robust models. One additional problem will be the impossibility to use the Multi-Quadrant tool for easily comparing the segments.
Quote 0 0