I would say that 3 classes is few enough that it should be fairly resistant to data mining. Figure 2b and 2c aren't as obvious to me that there are 3 classes,

I would say that 3 classes is few enough that it should be fairly resistant to data mining. Figure 2b and 2c aren't as obvious to me that there are 3 classes, but 2a and 2d are pretty clear. "All science is either physics or stamp collecting", but sometimes you have to have enough data points to start looking for patterns, and this looks like a pretty reasonable way to go about it. I don't think you can draw any sort of causal patterns yet, but the fact that there's a strong correlation between a 2-factor model based on the data and BMI leads to some testable hypotheses. [1]http://www.nature.com/nature/journal/vaop/ncurrent/full/nature09944.html http://www.nature.com/nature/journal/vaop/ncurrent/full/natu...