Skip to main content
Menu
Madden’s Player Speed Scores Are Surprisingly Good. But Whose Ratings Did The Game Get Wrong?

Football fans born after “Dirty Dancing” left theaters have never known an NFL without John Madden Football. The venerable video game franchise now known as Madden NFL released its first title in 1988 — for MS-DOS and the Apple II — the season Joe Montana and the San Francisco 49ers beat the Cincinnati Bengals for their third Super Bowl title. 

The brainchild of football nerd and Electronic Arts founder Trip Hawkins, the genre-defining title almost didn’t get made. Hardware constraints of the era limited the original game prototype to seven-on-seven play, but Coach John Madden demanded “real” 11-on-11 football before he would lend the use of his name. Madden wanted a game that coaches and football analysts could use, not a toy. After more than three years of development — at a time when most games took half that long to produce — Madden and others associated with the project expected it to be canceled. “All my memories are of pain,” game producer Joe Ybarra said of the game’s development. Ybarra left the Madden project EA soon after it was released and reportedly didn’t watch football again for a full year.

Today, Madden NFL is one of the most successful games of all time, and its release each year helps mark the beginning of a new football season — and not just for fans. Generations of players and coaches have grown up playing Madden, which pivoted to a more fast-paced arcade style compared with the original vision of Hawkins and Madden. The game even made it to Canton before the coach: The Hall of Fame opened an exhibit for Madden NFL in 2003 — three years before Madden himself was inducted

As the game has grown in popularity, so have its simulation systems. Madden ratings — the game’s way of assigning skill points to each player — began with just 10 categories. Today’s version ranks players across 53 separate dimensions, from acceleration to zone coverage rating. And it’s a living, breathing rubric. The system is updated after each week of the NFL season, and fans of the game pay close attention to the scores that are assigned.

Given its cultural impact, we wondered just how good Madden’s ratings were. Are fans and players who obsess over scores just using Madden as another proxy for ritual warfare and personal vanity, or do the scores capture important information about player performance that analysts should take seriously? In other words, are ratings tethered to anything objective, or are they just the opinions of a handful of nerds in Silicon Valley?

The answer we found is that Madden ratings — at least the ones we looked at — are quite good. Madden’s speed rating is perhaps the best example. At the majority of positions, the most important trait in Madden NFL is how fast a player can move, so we should probably expect the Madden rating team to put quite a bit of effort into getting their speed scores correct. And that’s just what we found.

We took data including player career maximum speed, previous year’s average max speed per game, current year average max speed per game and combine 40-yard dash times, and fed them into a multilevel model1 to predict Week 9 Madden Speed Scores and had reasonable success. Based on these results, Madden game designers aren’t merely using 40 times (which don’t exist for many players in the league) to score players; they’re also leveraging the NFL’s Next Gen Stats tracking data to some degree to generate their ratings. 

They’re being smart about it, too. Judging from the model and from conversations with a source at EA, the Madden ratings team is weighting current-year speeds more heavily than the previous year, and the model suggests that tight ends get the biggest boost in score for being faster than their peers, followed closely by quarterbacks.

There appear to be opinions baked into the scores as well, as a purely quantitative approach can’t explain all the ratings doled out by EA. This shouldn’t be surprising. The Madden team is also known for changing ratings if public outcry is loud enough: Last year, Madden revised Keenan Allen’s ratings after the Chargers receiver announced he was boycotting the game over his poor numbers. 

Using our model and the latest set of data2 published on the Madden website, we identified a few players over which Madden and a purely model-based approach appear to differ, and the results are interesting.

Madden ratings aren’t completely numbers-driven

Selected over- and underrated players by Madden NFL 22 based on difference in predicted speed ratings and Madden speed ratings

Max SPEED (mph) RATINGS
Player Pos Career 2020 2021 predict Madden Diff.
Daniel Sorensen FS 22.0 19.1 18.4 89 82 -7
Michael Pittman WR 21.3 19.0 19.1 91 87 -4
Jakobi Meyers WR 20.3 17.7 18.7 88 85 -3
Keenan Allen WR 21.1 18.3 18.8 90 88 -2
Patrick Queen MLB 19.6 18.5 17.1 84 89 +5
Tyreek Hill WR 22.8 20.5 20.6 92 99 +7
Kyle Pitts TE 19.2 18.1 80 91 +11
Lamar Jackson QB 21.0 19.4 18.5 85 96 +11

Max speeds from a player’s career are his fastest recorded speed. Speeds from 2020 and 2021 are a player’s average per-game maximum speed, through Week 8 of the 2021 season. Madden ratings are entering Week 9 of the 2021 season.

Sources: EA Sports, NFL Next Gen Stats

Injecting human judgment into a grading system isn’t necessarily a bad thing. Daniel Sorensen is a good example. He’s having a terrible year for the Kansas City Chiefs, and he’s playing a role in their defense’s surrender of 69.5 points over expected this year, 31st in the league. Sorensen’s performance has earned a coverage grade of 42.2 out of 100 from Pro Football Focus and an overall grade of 35.5. But despite his struggles, he still appears to be fast! Based upon his on-field speed as measured by player tracking data, we’d expect a purely model-based speed score for Sorensen of around 89, over 7 points more than Madden’s Week 9 grade of 82. Perhaps the folks at EA are loath to reward a player who is performing as poorly as Sorensen with a high score in an important category like speed rating, since it has such an outsized impact on gameplay. And If that’s the case, it’s hard to fault them.

Other notable outliers include Chiefs wide receiver Tyreek Hill, who has the highest speed score in all of Madden at 99. Hill is certainly fast, and perhaps we’re splitting hairs here, but his career max speed of 22.8 mph is good for just fourth in the league since 2017.3 Moreover, his current-year weekly average max speed, while very respectable, doesn’t even crack the top 50.

NFL predictions

Related: Our 2021 NFL Predictions Read more. »

Two Baltimore Ravens teammates also make the list as slightly overrated speedsters. Lamar Jackson appears to be running more slowly this year, perhaps due to nagging injuries to his back. His average max speed per game of 19.4 mph last season was nearly a full mile per hour faster than his speed this season. Madden is unbothered, though. The 96 speed score Madden gives Lamar halfway through the year is almost 11 points higher than his model-based score.4

Middle linebacker Patrick Queen is a stronger case of overrating. The 2020 first-round pick’s speed data is lower than what EA typically requires of a player with something approaching a 90 speed score, and Queen has slowed down substantially in his sophomore season, averaging a per-game max speed of just 17.1 mph in 2021.

Finally, Keenan Allen is still underrated, if only slightly. He should ask for a speed rating bump of 2 points or so — he’s earned it. 

After 33 years, the Madden franchise is still going strong. In 2020, EA paid the NFL $1.5 billion to continue its exclusive license with the league through 2026. The game’s shadow stretches to encompass football’s past and its near future, and Madden NFL’s developers have earned the right to be cited as a source of valuable football information. Madden’s original vision for the franchise as a tool for serious NFL analysis wasn’t ever blocked — it was just contained for a few decades. 

Check out our latest NFL predictions.


Who’s the most chaotic fictional football coach? | FiveThirtyEight

Footnotes

  1. Multilevel models allow the slope and intercept of linear models to vary by subgroup. In this case, we allowed the slope of career max speed to vary by position group and the intercept to vary by 40 time and position group. We limited the model sample to players with at least 100 total NFL snaps.

  2. Player ratings after Week 8 of this season.

  3. Among players at all positions with at least 100 career snaps since 2017.

  4. Some of Lamar’s lower predicted score is due to a lack of a 40 time in the combine. If we assume a 4.3-second 40-yard dash, his model score rises to 88, which is probably a better current estimate of his speed.

Josh Hermsmeyer is a football writer and analyst.

Comments