ptdotme

ptdotme's NFL Elo Power Rankings After Week 12

(By popular demand, see the rankings without the influence of last year's season-ending ratings in my comment below, or here.)

This is OC. I’ve written code to calculate NFL team Elo ratings^1 on a week to week basis. The goal is to use simple Elo ratings to create power rankings without the influence of human bias and emotion. The ratings are based on each team's rating from the previous week, with a "parity reset" applied every offseason. The model has been tested for accuracy against game outcomes since week 1 of the 2012^2 season. The ratings are derived only from each game's score, venue, and date. There are a number of variables/weights in my secret sauce but otherwise they're fairly conservative, basic, Elo ratings.

This is all a work in progress and feedback is appreciated! See more stats and info on my 2023 NFL Elo Power Rankings page.

Note: You may feel teams are ranked too high or low based on recent games, and you may be right. However, the model has been tested and is more accurate when it doesn't overreact to individual games. The goal is model accuracy, not rankings that "look right." ^(2)  

^(1)See Wikipedia. Elo ratings are numeric and assigned to each contestant. The ratings are used to estimate performance. After each game, the difference between a contestant's estimated and actual performance is used to update their rating. For the NFL, this can all be summarized as "Who did you beat/lose to and by how much? And who have they beaten/lost to and by how much? And who have they beaten/lost to ...

^(2)Moving backward through the years, each preceding season has a much smaller impact on the current ratings. Team ratings from 2010 have zero impact on today's ratings. Ratings from only a couple years ago, say 2021, have almost no effect on today's ratings. Seasons going back to 2010 are only used by the model to ensure its accuracy over thousands of NFL games.  

ptdotme's NFL Elo Power Rankings After Week 11

(By popular demand, see the rankings without the influence of last year's season-ending ratings in my comment below, or here.)

For this week, the model has been tweaked to slightly increase the value of wins and slightly decrease the importance of margin of victory. As always, this change was only implemented because it increased the accuracy of the model. Again, this was only a slight change so you won't see a big change in the power rankings -- teams went up or down a few Elo rating points and a few pairs of similarly-rated teams swapped rankings.

This is OC. I’ve written code to calculate NFL team Elo ratings^1 on a week to week basis. The goal is to use simple Elo ratings to create power rankings without the influence of human bias and emotion. The ratings are based on each team's rating from the previous week, with a "parity reset" applied every offseason. The model has been tested for accuracy against game outcomes since week 1 of the 2012^2 season. The ratings are derived only from each game's score, venue, and date. There are a number of variables/weights in my secret sauce but otherwise they're fairly conservative, basic, Elo ratings.

This is all a work in progress and feedback is appreciated! See more stats and info on my 2023 NFL Elo Power Rankings page.

Note: You may feel teams are ranked too high or low based on recent games, and you may be right. However, the model has been tested and is more accurate when it doesn't overreact to individual games. The goal is model accuracy, not rankings that "look right." ^(2)  

The top 4 teams won handily. The Ravens' Elo rating got a boost from a big win after Burrow went down, and the Bengals have nearly fallen out of the top 10. The Eagles were slight Elo underdogs at the Chiefs, and picked up a couple Elo points from their win last night. The Browns beat a weaker Steelers team by a smaller than expected margin, and thus lost an Elo rating point and didn't return to the top 10. The Jaguars had a nice win and skipped over a few closely bunched teams to get back into the top 10.

The Buccs, Steelers, Broncos, and Texans are all rated nearly the same, so their ordering from 17th-20th isn't meaningful.

The Broncos are on a league-leading 4-game, +33 Elo hot streak (the Raiders, Bears, and Steelers are on 3-game Elo hot streaks). The Jets are on a 4-game -29 Elo cold streak (the Titans, Panthers, Seahawks, and Patriots are on 3-game cold streak).

^(1)See Wikipedia. Elo ratings are numeric and assigned to each contestant. The ratings are used to estimate performance. After each game, the difference between a contestant's estimated and actual performance is used to update their rating. For the NFL, this can all be summarized as "Who did you beat/lose to and by how much? And who have they beaten/lost to and by how much? And who have they beaten/lost to ...

^(2)Moving backward through the years, each preceding season has a much smaller impact on the current ratings. Team ratings from 2010 have zero impact on today's ratings. Ratings from only a couple years ago, say 2021, have almost no effect on today's ratings. Seasons going back to 2010 are only used by the model to ensure its accuracy over thousands of NFL games.  

ptdotme's NFL Elo Power Rankings After Week 10

(By popular demand, see the rankings without the influence of last year's season-ending ratings in my comment below, or here.)

This is OC. I’ve written code to calculate NFL team Elo ratings^1 on a week to week basis. The goal is to use simple Elo ratings to create power rankings without the influence of human bias and emotion. The ratings are based on each team's rating from the previous week, with a "parity reset" applied every offseason. The model has been tested for accuracy against game outcomes since week 1 of the 2012^2 season. The ratings are derived only from each game's score, venue, and date. There are a number of variables/weights in my secret sauce but otherwise they're fairly conservative, basic, Elo ratings.

This is all a work in progress and feedback is appreciated! See more stats and info on my 2023 NFL Elo Power Rankings page.

Note: You may feel teams are ranked too high or low based on recent games, and you may be right. However, the model has been tested and is more accurate when it doesn't overreact to individual games. The goal is model accuracy, not rankings that "look right." ^(2)  

The 49ers once again played like a #1 ranked team, soundly beating a strong Jaguars team. The Cowboys dominated the Giants, and climbed up to 2nd place. The Ravens/Browns game was close, so only 4 Elo rating points were exchanged following that game. The Bills/Broncos game was also a 2-point game, but the Bills lost 7 Elo rating points because that was a bigger upset loss.

The Bills "should" be ranked lower due to their poor play for many weeks now, but haven't had a blowout loss yet to really punish their Elo rating. They are on by far the longest Elo cold streak at 6 straight games, losing 39 points over that span. The next longest Elo cold streak is 3 games (Rams, Jets, Seahawks, Bears, Patriots, and Giants).

The Steelers are outliers at 19th with a 6-3 record, but they haven't beaten any strong teams by a wide enough margin to be ranked any higher. The Chargers are another outlier at 9th place with a 4-5 record. Three of their losses were close, costing them only 1 Elo point each, and they had defense and special teams scores in the Jets game that really boosted their Elo rating.

^(1)See Wikipedia. Elo ratings are numeric and assigned to each contestant. The ratings are used to estimate performance. After each game, the difference between a contestant's estimated and actual performance is used to update their rating. For the NFL, this can all be summarized as "Who did you beat/lose to and by how much? And who have they beaten/lost to and by how much? And who have they beaten/lost to ...

^(2)Moving backward through the years, each preceding season has a much smaller impact on the current ratings. Team ratings from 2010 have zero impact on today's ratings. Ratings from only a couple years ago, say 2021, have almost no effect on today's ratings. Seasons going back to 2010 are only used by the model to ensure its accuracy over thousands of NFL games.  

ptdotme's NFL Elo Power Rankings After Week 8

(By popular demand, see the rankings without the influence of last year's season-ending ratings in my comment below, or here.)

This is OC. I’ve written code to calculate and format NFL team Elo ratings^1 on a week to week basis. The goal is to use simple Elo ratings to create power rankings without the influence of human bias and emotion. The ratings are based on each team's rating from the previous week, with a "parity reset" applied every offseason. The model has been tested for accuracy against game outcomes since week 1 of the 2013^2 season. The ratings are derived only from each game's score, venue, and date. There are a number of variables/weights in my secret sauce but otherwise they're fairly conservative, basic, Elo ratings.

This is all a work in progress and feedback is appreciated! See more stats and info on my 2023 NFL Elo Power Rankings page.

Note: You may feel teams are ranked too high or low based on recent games, and you may be right. However, the model has been tested and is more accurate when it doesn't overreact to individual games. The goal is model accuracy, not rankings that "look right." ^(2)  

The Eagles and Ravens have a legitimate claim at 1st place in human-written rankings, but in week 8 neither won by a wide enough margin to improve their Elo rating. As expected, the Bills' Elo rating continues to deteriorate. They are on a 4-game, -18 Elo slide. But other top teams are struggling even more, and the Bills have fallen backwards into 1st place after the 49ers' latest loss. With another big win the Cowboys, who have been bouncing between 3rd and 5th place, move back into 3rd. The Chiefs fell to 5th after losing by 15 points to the Broncos. The Dolphins, Lions, and Jaguars all move up a spot at Seattle's expense. The Jaguars finally crack the top 10: they're on a 5-game, +28 Elo hot streak.

Big games to look forward to in week 9 are Cowboys/Eagles, Chiefs/Dolphins, Bills/Bengals, and Ravens/Seahawks.

^(1)See Wikipedia. Elo ratings are numeric and assigned to each contestant. The ratings are used to estimate performance. After each game, the difference between a contestant's estimated and actual performance is used to update their rating. For the NFL, this can all be summarized as "Who did you beat/lose to and by how much? And who have they beaten/lost to and by how much? And who have they beaten/lost to ...

^(2)Moving backward through the years, each preceding season has a much smaller impact on the current ratings. Team ratings from 2011 have zero impact on today's ratings. Ratings from only a couple years ago, say 2021, have almost no effect on today's ratings. Seasons going back to 2011 are only used by the model to ensure its accuracy over thousands of NFL games.