I have a mixture of categorical and ordinal variables from a survey that I am trying to use to create "profiles" or segments that differ from one another with respect to a dependent variable (the dependent is also ordinal, 5 point scale). Several of the 'predictor' variables are on a 0-10 scale, a lot of the variables are 0,1 variables.
I have two challenges I was hoping to receive some guidance on that I can't find a specific existing answer to:
(1) The goal of the project isn't to "predict" per se the dependent variable ( in that case I was anticipating using some sort of logistic regression for multiple categories) but instead to create groupings of survey respondents with similar characteristics with respect to the dependent (opinion of a brand, for example) So, for example, -- we would want to know if there is a group that has high opinion of brands that is defined by having high levels of another ordinal variable and two categorical variables versus another group that has also has a high opinion of brands but for different reasons, i.e., high levels on another one of the other ordinal variables) Ideally, I want to be able to produce something similar to the CHAID analysis if I only had categorical data.
(2) The other challenge I have is that all the variables are highly correlated with one another (they are all different funnel metrics for a marketing survey) but there do appear to be (upon some casual data subsetting) particular subsets of the data that relate differently to the dependent variable than others, just on some casual inspection. Unfortunately, trying to then integrate many of the other variables into this casual inspection of the data becomes unmanageable from a descriptive statistics perspective. To what extent should I worry about this high correlation with respect to whatever the suggestion is for question 1?