I have a parenting measure that I would like to construct a multinomial outcome from (4 categories). I am constructing it as follows from two scales measuring two dimensions of parenting.

1. Calculating the mean score on each scale.

2. Determining a cut point at which to make the scales binary

3. Combining the 2 binary variables (i.e. high on both variables, low on both

variables) to come up with a multinomial outcome.

The problem I am running into is at the 2 step, determining the cutoff point. I can't find anything in the literature about where a cutoff point should be for this measure and population. The distributions for both variables are skewed otherwise I would consider a median split etc. I considered transforming, but I'm not sure I want to proceed that route anymore. I would rather figure out another way to determine a good cutoff point. Is there any specific statistical approach to this that might work? Another idea I had was to use this parenting variable, which is my outcome, as my predictor in a model where I predict several adolescent outcomes. If I can establish that particular groups (in particular parenting categories) do better than others, then this might help me determine where to put that cutoff point? Does that sound reasonable?

any thoughts...

many thanks!

s