Lessons Learned from Review of Predictive Model’s Win Totals

46

The process of projecting an NFL season requires an attempt to consider the wide range of outcomes for each player and team that could play out in the coming season, with the ultimate aim of coming as close as possible to projecting the current version of a player or team.

A recent consideration of a dilemma on the quarterback front was whether or not to remove Cleveland quarterback Baker Mayfield’s 2019 season from a turnover rate projection. Mayfield had a 3.9% interception rate in a disjointed offense in 2019, but that number dropped to 1.6% in a well-structured offense under new head coach/offensive coordinator Kevin Stefanski in 2020. There’s a reasonable argument that Mayfield’s play in 2019 is not relevant to his 2021 projection. In that case, the 2019 data should be removed to create a more accurate projection. These are all subjective judgments, so it can be useful to consider projections against alternatives to consider what those numbers suggest about the NFL season and whether adjustments should be made.

On the team front, a key comparison point is the projected win totals against the win totals listed with major sportsbooks. This piece will highlight three debates that came up in this process, including the teams with the largest positive and negative differences relative to the model.

The win total numbers used for comparison coming from DraftKings Sportsbook. My projected win total numbers are produced by inserting projected points for and against from the full 2021 season into the Pythagorean Expectation formula. The model also tracks the number of games a team is favored in, but projecting a team to be favored in every game and projecting an undefeated season are very different things, which is why I use Pythagorean Expectation to consider win totals. This run of the model had Jordan Love and Tyrod Taylor at quarterback for the Packers and Texans.

Example One: Washington’s Positive Win Total Differences

There were 31 teams available for comparison on DraftKings, with the Green Bay Packers off the board due to the Aaron Rodgers situation. The model had 22 of 31 projections within plus or minus one game of the win total.

The largest discrepancy came with Washington, who came in at 10.65 projected wins, 2.65 above the win total of eight. That number would come down by roughly half a win with Rodgers at quarterback for the Green Bay matchup, but even 2.15 above the win total would be the largest absolute difference of any team and merits review.

The review starts with Washington’s unit ratings. The defensive rating used has Washington as one of the three best units entering the season, tied with Tampa Bay and the Los Angeles Rams. Given the career arcs of defensive end Chase Young and the rest of the defensive line, the improvement the unit made throughout 2020, and offseason additions of first-round linebacker Jamin Davis and free-agent corner William Jackson III, that rating seems reasonable.

On offense, Washington came in 16th in these rankings. The key positives are the additions of quarterback Ryan Fitzpatrick and offensive weapon Curtis Samuel and the overall solidification of the offensive line. The unit already featured playmakers in wide receiver Terry McLaurin and running back Antonio Gibson, so the group has potential. The drawbacks are that Fitzpatrick has never directed an upper-tier offense and that the offensive line is unlikely to be a top-ten group.

After more consideration of the potentially limited ceiling, I downgraded Washington’s rating to move them to 20th. The positive win total difference will still be positive relative to the DraftKings win total will still be positive, but the gap will close. If I’ve overrated Fitzpatrick, this projection won’t work out, but his play in Miami the past two seasons has been consistently solid and Washington has talent around him.

Example Two: Carolina’s Negative Win Total Differences

Carolina came in as the team with the largest discrepancy on the negative side, with my model projecting them at 5.49 wins, 2.01 beneath the win total of 7.5.

A review of the ratings showed Carolina ranks 23rd on offensive and 22nd on defense in the current version of the model. The offensive rating is limited by quarterback Sam Darnold, who has demonstrated the physical tools to play at a high level and remains a viable prospect at his age but needs to improve in several areas relative to his time with the Jets. Offensive coordinator Joe Brady’s system may help him do that, but until he gets on the field in Week One, Darnold’s decision-making, consistency, and command of the offense will remain in question.

On defense, Carolina drafted corner Jaycee Horn 8th overall and added four notable veterans in defensive tackle DaQuan Jones, edge rusher Haason Reddick, inside linebacker Denzel Perryman, and corner A.J. Bouye. The staff is also making schematic changes and will move Jeremy Chinn out of the hybrid linebacker/safety role he played as a rookie to fill the strong safety role that unsigned veteran Tre Boston handled last season.

The critical question is at cornerback, where the Panthers need an instant impact from Horn and a rebound season from at least one veteran. Donte Jackson struggled with foot injuries throughout 2020 and has been inconsistent in coverage despite his excellent speed. Bouye has not played to his peak level over the past two seasons and is suspended for the first two games. The defense has standout players in edge rusher Brian Burns and inside linebacker Shaq Thompson, but if the back-end coverage doesn’t improve, the defensive issues will persist.

Even if Darnold shows notable improvement, getting into the playoffs at 9-8 seems like the high end of the spectrum of outcomes for the season. This team could also be terrible if Darnold flops, as the backup quarterbacks, P.J. Walker and Will Grier, do not inspire confidence. The bust potential for this team, coupled with the limited ceiling and the fact that they won five games last season (with a Pythagorean Expectation of 6.70 wins) may mean the win total of 7.5 is one game too high.

Example Three:  A Lesson from Houston

Houston is ranked dead last in this model’s offensive and defensive ratings. They are projected to score the fourth-fewest points and to allow the most points. The model projected them as the favorite in one game, at home against the Carolina Panthers, with an edge of 0.61 points.

These projected results are not a unique take on Houston’s current state, which has left some questioning why Houston’s win total remains at 4 or 4.5, rather than dropping even further, but the Pythagorean Expectation formula offers some perspective on this dilemma. It is not a perfect formula (evidenced by the fact that almost every team’s record finishes either above or below their Pythagorean Expectation each season), but a reminder of the average record for a team with a given differential. In the case of the Texans, who have this model’s worst point differential projection by a notable amount, the formula still comes out to 5.03 expected wins. For reference, last year’s 1-15 Jacksonville Jaguars team finished with a Pythagorean Expectation of 3.92 wins in a 16 game season.

The 2019 Miami Dolphins are an interesting case study to consider as a comparison. In the “Tank for Tua” campaign that saw the Dolphins unload left tackle Laremy Tunsil to the Texans shortly before the season, Miami was as bad as any football team has ever been for the first few weeks of the season, but rallied and finished 5-11. It’s not inconceivable that the same could happen in Houston. Tyrod Taylor was the starting quarterback for a Buffalo Bills team that went 9-7 in 2017. The roster currently lacks established talent relative to other NFL teams, but it’s a roster of professional football players working to improve, and the inherent competition with so many jobs available will produce some surprise contributors.

Don’t interpret this as an argument to bet the over on Houston. There are many ways for Houston’s situation to unravel and they could become the first team to go 0-17, but it’s important to keep the positive possibilities in mind as well, rather than dismissing the 4 or 4.5 win total as a mistake.

Conclusion

These are only a few examples of a process I apply to each team as I prepare my model for the upcoming season. I won’t be able to hit every debate but will get into some player projections and other considerations in future articles this summer. A final suggestion for others creating projections is to complete your process before comparing your results to others. It’s important to review alternative results and consider the different assumptions other projections are making, but if you haven’t established an initial opinion, you can fall victim to chasing results.

Leave a Reply