I should add that the word "predict" means something different in science/math than in colloquial speech. It means "logically derive from assumptions" rather than "claim something will occur in the future". So increasing the predictive power of a model also means you can explain what occurred in the past better, not just predict the future better. The limitations of stats really lie in not having detailed enough stats rather than the methods of statistical analysis. We just don't have detailed enough football stats like how far away player X was from player Y, or what the formation was, etc. NFL Next Gen Stats is taking one small step in that direction, but it's not publicly available and it's still not ideal. Gather that type of data and the Tannehill debate would look a lot different too.