DATA SCIENCE
Over the first two weeks of the NFL playoffs, I shared my model (V 5.0) predictions (Wild Card, Divisional). After 10 games, model performance has almost exactly matched performance in the test set for predicting winners (actual: 7/10, 70%; test set: 69.6%). Once again, when it comes to predicting score difference, the model performed better than other estimates shared by ESPN (supplied by Ceasars Sportsbook by William Hill; CSWH) and FiveThirtyEight (QB-adjusted Elo) [more details in Divisional Review section]. The model continues to perform well relative to these other predictions and this week, for the first time, predicts a different winner in one of the games, Bills over the Chiefs [more details in Why does V 5.0 favor the Bills? section].
Below are V 5.0 predictions for the Conference Championships, including predictions from CSWH and FiveThirtyEight that were available at time of writing:

In contrast to the previous two weeks, the favorites do not all match as V 5.0 predicts the Bills to win, while the other two favor the Chiefs. All three agree that Packers are favored, but V 5.0 predicts a closer game than the others.
The most likely outcome for V 5.0 picks, if we assume its winner accuracy is 69.6%, is to get both Predictions correct (2 correct: 48%, 1 correct: 42%, 0 correct: 9%).
Why does V 5.0 favor the Bills?
Last week I described how the top contributors to the Random Forest algorithm were mostly power rankings [eatdrinkandsleepfootball], but also efficiency metrics [numberFire]and Elo scores [FiveThirtyEight]. For the Bills and the Chiefs, the power rankings and Elo scores are quite similar, but the efficiency metrics differ a bit, especially in recent weeks.

For almost the entire season, the Chiefs Overall Offense rating was higher than the Bills, but after week 16, the Bills pulled ahead in this metric and currently lead the league. The Bills offense put up respectable performances against two top-10 defenses (Colts and Ravens), which bolstered their rating. On the other side of the ball, the Bills Overall Defense rating been pretty constant since week 13, while this rating for the Chiefs has gradually worsened. I’d be remiss to not mention a crucial factor for this game—the status of MVP Quarterback Patrick Mahomes, who sustained an injury last week. If he does not play, V 5.0 might actually be underestimating the Bills’ chances for victory.
For the Buccaneers and the Packers, while the net efficiency rating favors Tampa Bay, the power rankings and Elo scores favor Green Bay enough to tip the Packers’ direction.
Divisional Review
As mentioned, for the two previous weeks, all the three sources predicted the same winners. Therefore, we can once again compare mean absolute error (MAE, lower values indicate better predictions).

The % correct is guaranteed to be different after this Sunday’s games, and MAE will likely change as well. We’ll find out soon if V 5.0 remain ahead of the others heading into Super Bowl LV.
Some Interesting Story Lines
Fans of the NFL know that interesting stories abound across the league. Here are a couple I’m thinking about going into the Conference Championships:
- Unlike the other three teams playing this Sunday, the Buffalo Bills have never won a Super Bowl. They made four consecutive appearances in the early 90s, but went 0–4. If V 5.0 is right, they’ll have another shot at the Lombardi Trophy in a couple weeks.
- If a few things go well for the Buccaneers and they are able to upset the Packers, they would be the first team to play a Super Bowl in its home stadium, potentially giving them a boost over their AFC opponent. Tom Brady would also pad his (likely insurmountable) record of Super Bowl appearances (currently 9).