Source: Wall Street Journal
It’s the last week of the NFL Regular Season, and there is so much that needs to be decided especially in the AFC. I thought about doing a post on the different playoff and seeding scenarios, but it honestly is headache-inducing. People who are way more versed in that sort of thing have better graphics for it, and that’s what I have presented below with the proper credits.
I also spent the last week on a few more modeling techniques that I’ll get into below as well. The last model is one of those “hey, I simulated each game in Week 17 10,000 times, and this is who I project will win based on those simulations.” Yeah, I finally got that to work!! Not only does it predict who will win, but I can also simulate and get a ranking of each team which I have done below.
So let’s get going, shall we?
Week 17 Playoff Scenarios
The NFC is pretty much nearly set. New Orleans has the #1 seed locked in and a bye in the first week. Dallas is set as the #4 seed and are basically waiting to find out who they will host next week at the #5 spot. Chicago and the Los Angeles Rams are basically playing today for either a bye as the #2 seed or a #3 seed and host to a Wildcard game next week. Seattle is in and just need to know where they’ll fall as either a #5 or #6 seed. Minnesota and Philadelphia are pretty much the teams that have various scenarios on whether they make it into the playoffs at all and where.
Below is a graphic I found from @Ivan_Urena1 from Twitter that shows the various scenarios and what the final seedings would be based on the results of those games. Not a coincidence that all four games are being played at the same time this afternoon.
Quite honestly, the AFC playoff possibilities are a hot mess. What’s really fascinating about what’s going on in the AFC is that there are teams fighting for a Bye Week next weekend that may very well not even make it into the playoffs! Of course, one of the more important games is the Sunday night game on NBC between Indianapolis and Tennessee. It’s basically a play-in game: winner moves on, loser goes home.
Once again from @Ivan_Urena1 are two graphics that go through all the various scenarios. Convoluted isn’t it?
Final Projected “Power” Rankings
The last model I worked on is a Bayesian model using PyMC3 in Python. I ended up using it with Google Colab since it already had all the necessary libraries installed and ready to go. One of the advantages of using a Bayesian model is that you can simulate however many games you want for games being played in the future based on data from the past. For this example, of course, I ran 10,000 simulations of all the Week 17 games. From this information, I can get the probability that the home team will win. I can also use those simulations to get ranking of how the teams will end up at the end of the season. The following table shows how that worked out for my model:
| Rank | NFC | AFC |
|---|---|---|
| 1 | New Orleans Saints | Kansas City Chiefs |
| 2 | Los Angeles Rams | Los Angeles Chargers |
| 3 | Chicago Bears | New England Patriots |
| 4 | Seattle Seahawks | Houston Texans |
| 5 | Dallas Cowboys | Baltimore Ravens |
| 6 | Minnesota Vikings | Tennessee Titans |
New Orleans came out to be the clear overall #1 seed, using a phrase that comes up during the NCAA Tournament in March every year. The Rams were a very close #2 overall. They already played each other in New Orleans back on November 4th, and it was definitely an offensive showdown with New Orleans coming out on top 45-35. I think it would be fun to see those two play in the Conference Championship in three weeks for the right to go to the Super Bowl. In the AFC, the Chiefs were the #1 seed (#3 overall). I imagine most people remember the the Chiefs-Rams game back on November 19th where 105 total points were scored! I think a Super Bowl game between those two would be fun to watch…well, unless you’re just REALLY into defense.
Week 17 Win Predictions
All right, admittedly my “new” table down below is very busy. I really went nuts trying out different models this week with the hardest one to master being the Bayesian model. Let me explain what each model includes:
- Model 1: Original based on a Poisson distribution between points scored at Home vs. Away for each team
- Model 2: Based on Nate Silver’s/FiveThirtyEight’s Elo model with one little tweak by me. They use a Home-Field Advantage (HFA) metric of 65 which is based on home teams having a 2.6 point advantage over away teams. I use a 2.3 point advantage based on strictly 2018 data which results in an HFA of 57.5.
- Model 3: Uses data from Model 1, Model 2, and four statistics proprietary to Football Outsiders that measure team efficiency: Weighted DVOA, Offense DVOA, Defense DVOA, and Special Teams DVOA.
- Model 4: The Bayesian model described above based on 10,000 simulations of each game played in Week 17 based on points scored at Home and Away.
I also included a column that takes the average of the third and fourth model as well as a column with the predicted score based on the first model and all the usual columns I’ve included before.
| # | Game | Model 1 (Original) | ELO Model 2 (HFA=57.5) | Model 3 (FO+ELO) | Model 4 (Bayesian) | Avg of Model 3 & 4 | Vegas | Poisson Predicted Final Score | Expected Tot. Points Scored | Vegas Over/Under |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | MIA | 46.57% | 41.14% | 15.28% | 56.74% | 36.01% | 32.79% | 21 | ||
| BUF | 46.92% | 58.86% | 80.58% | 43.26% | 61.92% | 69.74% | 17 | 38 | 40 | |
| 2 | ATL | 43.12% | 54.15% | 64.67% | 46.00% | 55.34% | 54.98% | 32 | ||
| TB | 51.64% | 45.85% | 30.44% | 54.00% | 42.22% | 47.39% | 26 | 58 | 52 | |
| 3 | DAL | 50.75% | 62.06% | 40.30% | 69.88% | 55.09% | 31.25% | 23 | ||
| NYG | 43.04% | 37.94% | 53.65% | 30.12% | 41.89% | 71.23% | 18 | 41 | 41 | |
| 4 | CAR | 1.15% | 13.04% | 1.09% | 1.30% | 1.20% | 26.67% | 21 | ||
| NO | 98.35% | 86.96% | 98.43% | 98.70% | 98.57% | 75.82% | 31 | 52 | 43 | |
| 5 | NYJ | 1.85% | 10.83% | 0.27% | 3.57% | 1.92% | 15.15% | 22 | ||
| NE | 97.43% | 89.17% | 99.57% | 96.43% | 98.00% | 88.26% | 31 | 53 | 46 | |
| 6 | JAC | 6.99% | 29.20% | 3.50% | 7.15% | 5.33% | 28.33% | 16 | ||
| HOU | 90.51% | 70.80% | 95.00% | 92.85% | 93.93% | 74.18% | 21 | 37 | 40.5 | |
| 7 | DET | 10.81% | 35.24% | 20.44% | 21.00% | 20.72% | 25.32% | 23 | ||
| GB | 86.26% | 64.76% | 75.28% | 79.00% | 77.14% | 77.22% | 25 | 48 | 45.5 | |
| 8 | PHI | 58.71% | 65.04% | 76.58% | 81.07% | 78.83% | 73.48% | 24 | ||
| WAS | 35.33% | 34.96% | 18.81% | 18.93% | 18.87% | 28.99% | 18 | 42 | 41.5 | |
| 9 | LAC | 61.14% | 64.12% | 66.08% | 76.50% | 71.29% | 74.24% | 24 | ||
| DEN | 32.98% | 35.88% | 28.48% | 23.50% | 25.99% | 28.25% | 17 | 41 | 43 | |
| 10 | OAK | 0.10% | 11.73% | 0.46% | 0.37% | 0.42% | 14.04% | 25 | ||
| KC | 97.32% | 88.27% | 99.18% | 99.63% | 99.41% | 89.37% | 42 | 67 | 52.5 | |
| 11 | CHI | 57.20% | 44.03% | 45.70% | 50.44% | 48.07% | 30.12% | 24 | ||
| MIN | 36.74% | 55.97% | 48.09% | 49.56% | 48.83% | 72.31% | 18 | 42 | 40.5 | |
| 12 | CLE | 5.10% | 18.92% | 16.05% | 10.93% | 13.49% | 30.21% | 16 | ||
| BAL | 92.93% | 81.08% | 79.81% | 89.07% | 84.44% | 72.25% | 21 | 37 | 40.5 | |
| 13 | SF | 0.68% | 14.93% | 4.96% | 2.12% | 3.54% | 20.92% | 21 | ||
| LAR | 99.00% | 85.07% | 93.36% | 97.88% | 95.62% | 81.57% | 33 | 54 | 48.5 | |
| 14 | CIN | 2.90% | 17.34% | 0.49% | 8.01% | 4.25% | 14.37% | 24 | ||
| PIT | 96.08% | 82.66% | 99.26% | 91.99% | 95.63% | 89.05% | 31 | 55 | 45.5 | |
| 15 | ARI | 0.10% | 12.32% | 0.01% | 0.65% | 0.33% | 15.87% | 13 | ||
| SEA | 99.82% | 87.68% | 99.98% | 99.35% | 99.67% | 87.57% | 27 | 40 | 39.5 | |
| 16 | IND | 36.79% | 37.90% | 39.20% | 27.85% | 33.53% | 63.13% | 23 | ||
| TEN | 57.27% | 62.10% | 54.95% | 72.15% | 63.55% | 39.37% | 20 | 43 | 44.5 |
Not too many differences between my “average” model and money line odds from Pinnacle. I highlighted the three games that were flipped. I certainly hope that Dallas wins this weekend so that they go into the playoffs on a win streak, but with nothing being decided for them playoff-wise, it wouldn’t surprise if they bench their starters early in the game.
I have the Chicago-Minnesota game as basically a coin flip, but Vegas seems to be more confident in a Minnesota win.
And finally, the Sunday night game is almost equally flipped between my model and Vegas’s predictions. I double checked my numbers, but that’s definitely how my predictions shake out. That should definitely make for an entertaining game hopefully.
Enjoy the end of the Regular Season! Hopefully your favorite teams makes it into the playoffs so that you can enjoy more football, and if they’re not, well, better luck next year, I guess.