No I'm NOT using a "curvilinear relationship" for 2019. That's the whole point. You don't a priori know what that relationship is. And a correlation (if used for calculating "variance explained") assumes linearity so obviously I'm not making any such assumptions. Look.. you want to just fit a function to the data with blinders on and ignore population stats. Try this: There's your "curvilinear" relationship. And it says precisely the opposite of what you're arguing. It says that Tannehill in 2019 gets FAR worse with low-volume than with high-volume and that he's best mid-level volume. Of course.. none of this has any credibility because we have tons of data from other QB seasons and it IS a linear relationship for the population (I showed that!) so there is no justification for fitting any such curve like this. Point is though, even IF you fit such a curve your hypothesis fails. And this type of analysis would be "lying with statistics". No other way to put it.