NFL Regression: Identifying Teams That Over or Under Performed by a Statistically Significant Margin in 2020


The principle of regression to the mean holds that when a sample data point in a dataset falls outside of predictable outcomes, a future data point is likely to be closer to the mean. Using the vast amount of data available concerning the NFL and basic Euclidean Geometry we can identify teams likely to experience a positive or negative regression in win totals in 2021 compared to 2020. 

To accomplish this, we can assess well-accepted metrics that demonstrate unlikely to be repeated results. The first metric is taken directly from middle-school geometry, the Pythagoras’ Theorem. In short, A2+B2=C2. We translate this to the following formula for the NFL:

You may notice that the product of the formula above will always be no greater than one. Unless the team in question failed to allow a single point, it would result in a percentage less than 100. Simply multiply that product by the number of games played, and you have an expected number of wins. 

This is an example of the most basic version of the metric. From this point, many bells and whistles are available. One typical example is that many trim the point totals to remove “garbage time” points. The other metrics used are win percentage in one-score games, fumble recovery rate, and net turnover differential. Historically, these metrics are highly indicative of an impending regression.  

The sections below identify over and underperforming teams from 2020. Underperformers expect a positive regression in their win totals from 2020, or an increase in their win totals from the prior season. Overperformers, those that out-kicked their coverage, expect to experience a negative regression in win totals. 

The graph above demonstrates that the Jaguars, 49ers, and Falcons underperformed in 10 of the 12 metrics last season. Conversely, it demonstrates that the Colts, Titans, and Chiefs overperformed in 11 of the 12 metrics. It may help to think of “0” as the mean performance for each team while looking at the graph. 

Underperformers: Jaguars, 49ers, and Falcons

The Jacksonville Jaguars finished 2020 at 1-15. The data suggests the team performed more consistently as a team finishing 5-11. In a 16 game season, this amounts to a 25% deviation. Some obvious data points explain the variance. The Jaguars recovered only 35% of available fumbles. While causing a fumble is a repeatable skill, recovering fumbles is not a skill on the macro level. In addition, the Jaguars finished the season with a record of 1-6 in games decided by a single score: a win rate of 14.2%. This represents a substantial negative deviation from the expected outcome. Therefore, when you access win totals for the current season, apart from personnel changes, the Jaguars were already substantially better than their previous record indicates. A 5-11 team posing as a 1-15 (though that posing got them Trevor Lawrence; going 5-11 and without a trade, they miss Lawrence, Wilson, and Lance in the 2021 Draft.)

The San Francisco 49ers finished the 2020 season with a record of 6-10. The 49ers’ deviation was not as significant as the Jaguars as the data indicates they performed more similarly to a team with 7-8 wins. Nevertheless, this represents a 10% deviation from the expected win total. The 49ers experienced the trifecta of negative regressions last season. First, they only recovered 36.3% of available fumbles. Second, the 49ers prevailed in 42.8% of one-score games. Admittedly, that is nothing compared to the Jaguars’ performance, but it is still a negative regression. Third, the 49ers had a double-digit negative net turnover differential. To compound this notorious “hat-trick,” San Francisco also suffered an unequaled number of losses due to injury. They lost their starting quarterback, Garoppolo, for ten games. They lost Bosa and Ford for thirteen games each. Sherman sat out 11 games, and Samuel sat out nine. Kittle and Mostert sat for eight apiece. The 49ers dealt with these regressions while facing one of the most difficult schedules in the league last season. 

The Atlanta Falcons finished with a 4-12 record in 2020. The data indicates that the Falcons, like the 49ers, performed at the level of a 7-8 win team. This represents a 22% negative deviation from their expected win total. The Falcons went a dismal 2-8 in one-score games: a win rate of just 20%. Atlanta led in 13 of its 16 games last season. They had the lead at halftime in nine games, yet they managed only four wins. Surprisingly, the Falcons narrowly exceeded expectations in fumble recovery, so perhaps the expected wins are slightly closer to seven wins than eight. Atlanta faced a demanding schedule last year. This year, the Falcons’ schedule is in the bottom third of the league. Will you be shocked if the Falcons double their win total from a year ago? You shouldn’t be.

Overperformers: Colts, Titans, and Chiefs

The Indianapolis Colts went 11-5 last season and far exceeded their expected win total. The data indicates the Colts played as a nine-win team. A two-game positive variance results in a 12.5 percent increase. Indianapolis went 5-2 in one-score games last season – a 71.4% win rate. They recovered ten of fourteen fumbles. That is an identical and astonishing 71.4 percent success rate on fumble recoveries. The Colts ended the season at a +10 net turnover differential, which was second-best in the league. The Colts lost in the Wild-Card game. The data expects a significant negative regression in 2021. 

Sticking in what the model sees as a suddenly suspect AFC South, the Tennessee Titans also finished at 11-5 in 2020. The data viewed the Titans as an 8.5 – 9 win team last season. Identical to the Colts, Tennessee enjoyed a two-game positive variance which resulted in a 12.5 percent increase. Outdueling the Colts, the Titans went 7-2 in one-score games. That is a success rate of 77 percent. They recovered a more modest number of fumbles than the Colts, eight of thirteen. However, that is still a success rate of 61 percent. The only team to have a higher net turnover differential than the Colts was the Titans at +11. As was the case with the Colts, the Titans lost in the Wild Card Game. The data expects a significant negative regression in 2021.

The Kansas City Chiefs finished the regular season at 14-2. I want to stress that while the data believes the Colts and Titans are average teams that performed well, it understands that the Chiefs are a different animal. The Chiefs are an elite team that also enjoyed a positive deviation from the mean. The data viewed the Chiefs as a ten-win team—a four-win positive variance results in a 25% increase. The immediate number that leaps off the page is that the Chiefs finished 2020 with a whopping net +7 wins in one-score games. Kansas City lost the fumble recovery battle with a net -3. However, they still finished in the top third of the league in net turnover differential at +6. If a few people want to attribute some of the mean-defying 2020 Chiefs’ performance to the supreme talents of Mahomes and Reid, I will not protest vociferously. However, the numbers are unambiguous. The Chiefs should expect a regression to the mean this season which will negatively impact their win total.    

+ posts