If goal is to make a predictive model, we would use other tools than CHAID (random forest for example). Chaid is to interactively explore data and relationships. As an exploratory tool, being able to visualize different boxes/split results enables exploration of relationships, such as alternative predictors and split points.
I find Partition difficult to use “interactively” (choosing variables and split points) vs. AnswerTree. They might also consider an exhaustive CHAID model (multiple splits) for partition. For competing “Candidate” variables, it would be nice to simply make choices on the Candidates table. It would be great to hover over a variable and get split information (n, dependent variable measures, p-value information/significant differences) and the ability to choose to split there AND choose the number of splits. Instead of separate sections for choices.
Otherwise, if you still have to go into triangle menus, then it would be good to have the competing variable choice box sorted the same way as the competing variable summary table. We need a way to quickly evaluate all the candidates and their split point information (n, conditional dependent variable scores).
