Winning the turnover battle has a strong correlation to winning the game. This is not news. However, the ways in which other sites have gone about predicting this historically has been flawed (FPS included). Here’s what we know about turnovers:
- Fumble recovery is a function almost entirely of luck. We know that regression sways recovery rates over the course of a season, however using regression within the context of a single game is a fool’s errand. Hence, outside of possibly finding correlation between QB hits and fumble rates (to be investigated soon), on a week-to-week basis we will not include any attempt to predict fumble/fumble recovery rates for a team.
- Relatively speaking, interceptions are slightly more predictive, but the means to analyze this on a week-to-week basis has typically come up flat in the analytics community. The fact of the matter is, all the conditions that come in to play during an actual IN-GAME interception don’t correlate with season-long traditional data. In other words, just because a QB has thrown x interceptions, or a defense has forced y interceptions has very little to no merit on if a QB will throw a pick in THIS GAME.
With this being said, here are two advanced data points that do trend with game-to-game interception rates:
- How long a QB holds on to the ball. (Yes there’s much more to this, but especially for QBs with low TTT rates, we have found lower interception rates all other things equal)
- How aggressive a QB is (how frequently he throws into tight coverage on average), relative to how accurate he is (Completion rate +/- expected).
*Both data points we thank Next Gen Stats for lending to us.
Using this data, along with Pro Football Focus’ “Comparative Gradings” between Team Pass Protection and Opponent Pass Rush/Pass Coverage Grade we have found a model that predicts turnovers much stronger than any other before. Here is how it plays out for week 12:
Note that the final ratings are OPPONENT TO Likihood, meaning this is setup to be interpretted as “Team A will benefit from Turnovers at this rate” with the higher the number meaning the higher liklihood of a TO from your opponent.
Next up we will share our full model for the week.