It’s pretty easy to show that while on one scatter plot questionnaire results of depressed people may tend to form a cluster of points, typical medication is only working in half of the cases, ergo if we add one more dimension than we could separate it into at least two clusters with probably widely different physical explanation.