It’s a word almost guaranteed to be uttered by the Fox announcers during this month’s Super Bowl, probably after a quick score and a defensive stop: momentum. Curmudgeons like myself will scoff and note that statistical inquiries have mostly debunked the notion that individual athletes are truly “clutch” or “hot” and not just vessels of random variation in success rate.
But what about entire teams? UW-Milwaukee math professor Paul Roebber trained a neural network to analyze 10 years of NFL play data, looking for signs that momentum (defined by Roebber as three successive changes in ball possession) changed a team’s win probability beyond the inherent benefits of score and field position.
It’s time to pick your Milwaukee favorites for the year!
He found that, at least for NFL teams in a single game, teams actually can get on a roll that predicts future success – or failure. His research cites as examples of momentum the streaks of positive outcomes (for both teams) in Super Bowl LI, in which the Atlanta Falcons led the New England Patriots 28-9 deficit with just over 17 minutes remaining in the game – a near-lock 99.3% win probability. Football fans know what happened next – and how simultaneously unlikely and certain the Patriots’ comeback felt at the time.
This does have us wondering what’s next, though: Perhaps machine learning will validate Troy Aikman’s pronunciation of Green Bay?
