this is correct. conventional statistical functions rely on a whole host of assumptions about the shape and type of data. Typically that its a bell-curve distri

this is correct. conventional statistical functions rely on a whole host of assumptions about the shape and type of data. Typically that its a bell-curve distribution & ratio style data. The more you move away from that parametric style of data, the more conventional statistical functions simply have no meaning. To use an example. I worked at a shop where we had a freemium model. About, let's say, 50% of our customers paid $0.00. About 2% paid over, say, 10K. Between those two points we had a reasonably clean exponential curve, clustering at key product price points. (I've goosed the numbers for confidentiality purposes, but this is close enough for discussion). In this scenario, what was the average ROI per account? Well, truth is, that's not a useful answer, as it was roughly $0.01. I wound up building a small set of non-parametric data tools to assess the customer base. It was a fun journey, but a critical lesson is: the kind of data you have and its shape determine the statistical functions that have meaning, and, they didn't teach that kind of data analysis in your basic calc-based stats class in college. :)