Every week, ESPN, NFL.com, and every sports site publishes power rankings. They are content filler, not analysis. Here is why.
The evidence (Source: analysis of ESPN power rankings 2019-2024):
- Preseason number 1 has won the Super Bowl 2 times in the last 20 years.
- The eventual Super Bowl champion was ranked outside the top 5 in preseason power rankings in 12 of the last 20 seasons.
- Week 8 power rankings have a correlation coefficient of only 0.43 with final standings. That is barely better than random.
Why they are bad at prediction:
- Recency bias. A team that wins 45-3 in Week 4 jumps 8 spots. A team that grinds out a 13-10 road win drops 3 spots. The margin of victory is weighted over the quality of the win.
- Brand bias. Cowboys, Packers, and Steelers are consistently ranked higher than their records justify. Small-market teams are ranked lower.
- Narrative over data. Rankings reward storylines, not statistical profiles. A team with a flashy quarterback gets ranked higher than a team with elite defense and average offense.
What actually predicts playoff success (Source: Football Outsiders DVOA):
- DVOA (Defense-adjusted Value Over Average) has a 0.71 correlation with playoff success. Almost double the correlation of subjective power rankings.
- Point differential through Week 8 has a 0.65 correlation with final record. Better than any human ranking.
- Simple expected wins (based on points scored and allowed, Pythagorean expectation) outperforms expert rankings.
Sources:
- ESPN — historical power rankings archive
- Football Outsiders — DVOA methodology and data
- Pro Football Reference — Pythagorean expected wins