Wild Card Monday, Bills-Steelers & Buccaneers-Eagles: Advanced Reviews

The Bills and Eagles continue their late-season form, with one team looking like a strong Super Bowl contender, and the other swept out with questions about its future

The adjusted scores quantify team play quality, with emphasis on stable metrics (success rate) and downplaying higher variance events (turnovers, special team, penalties, fumble luck, etc). Adjusted expected points added (EPA), in conjunction with opportunity-based metrics like total plays and drives, projects adjusted points. Adjusted scores have been tested against actual scores and offer slightly better predictive ability, though their primary benefit is explanatory.

All 2023 & 2022 and historical Adjusted Scores and other site metrics are available in a downloadable format to paid subscribers via Google Sheet.

Find previous advanced reviews here

** Adjusted Scores table:

• “Pass” - Pass rate over expectation (based on context of each play and historical averages

• “Success” - Success rate on offense, a key metric in adjusted score vs actual

• “H & A” - Home or away team

BUF-PIT

The Bills handled business at home against, perhaps, the weakest opponent of the Wild Card round. There were some exciting moments in the second half, at one point the Bills having only a single-score lead, but the game was never close by win probability calculations.

The Bills had an 8% success rate advantage offensively, and their advantages with turnovers (net +10.7 EPA) and late-down conversions (+10.5 EPA) more than mitigated poor luck on special teams (two missed field goals at -8.6 EPA).

In a prime example of how perceptions don’t match win probabilities, especially earlier in games, was the Steelers decision *not* to use a timeout after Josh Allen was sacked and the Bills faced 2nd & 17 on their own side of the field with around 40 seconds remaining. At that point, the Bills were up two scores and would get the ball first in the second half, giving them a 97% chance of winning according to the nflfastR model.