I have 4 categorical variables, one is by state so it has 50 possible values. I also have 3 continuously distributed values. I'm trying to predict a continuous variable. I'm using a random forest for my model.
I'm wondering what the best way is to determine the effect size for each variable. I want to be able to say something like "A person from California causes an x% increase in the dependent variable, holding all other variables constant."
The method I'm considering is brute-forcing the other variables. For example, I hold the state variable at a specific value and brute force possible combinations for the other variables, running each case through the model, then take the mean of the dependent variable across the whole sample. This seems fine but obviously is computationally expensive.
Is there some better way to do something like this?