Photo Credit: Erik Dorst at Flickr Creative Commons
When the NFL schedule for the upcoming season is announced in April, it is common practice for people to engage in the “schedule game.” That’s where one goes through a team’s game-by-game schedule and makes predictions about each game. In the end, one adds up the wins and losses and arrives at a forecast for that team’s record.
Most of the time, it’s just a harmless parlor trick. For all of us fans, it’s a simple way of evaluating a schedule. For NFL writers, it’s an easy way to generate page views. And while I understand the incentives of writers and editors to draw site traffic, I cannot let ESPN Cleveland’s inflammatory headline go unchallenged:
NFL NATION REPORTERS PREDICT BROWNS WILL GO 0-16 IN 2017 (link)
To be fair, Pat McManamon did his own Browns’ schedule game and predicted that the Browns would go 2-14. The winless story comes from analyzing the schedule game played by the writers covering other NFL teams at ESPN.
Still, I suspect that ESPN Cleveland didn’t mind the site traffic generated by that headline and the corresponding story.
But while the NFL schedule game may be a fun parlor game, the idea that it constitutes real “analysis” by those at ESPN—consider the seriousness implied in this SportsCenter spot—and other media outlets is a sad commentary given that it contains a logical fallacy and ignores simple probability theory.
Why is the NFL schedule game flawed? And, even if it is discredited, how seriously should Browns fans take the prediction that the team will go 0-16 in 2017?
To demonstrate the logical fallacy of the NFL schedule game, suppose that two teams play 16 games against one another on a neutral field. Suppose that Team A is ever-so-slightly better than Team B such that the odds of Team A winning any one game in the series is 51%. Team B, correspondingly, would have a 49% probability of winning any one contest.
Now, if someone had to bet money on one of these two teams, it would be rational to always bet that Team A would win in any single contest; they have better odds of winning. It would therefore follow that, if one played the schedule game in this rivalry, it would be rational to bet that Team A would win every game if we assume that each contest is an independent event (i.e., the outcome of one game does not affect the outcome of a future game).
In other words, this approach would lead one to predict that Team A would finish the series 16-0 while Team B would finish 0-16.
This is obviously not a realistic perspective on how a 16-game season between these two teams is likely to play out. Sometimes, the underdog will win. In fact, with a distribution of odds of 51%/49%, the most probable season-long outcome is that the two teams split the season series and end with 8-8 records.
The same perspective can be applied to the 2017 Cleveland Browns. Even if one assumes that the Browns will be underdogs in every game—which is a big “if” with a 2017 schedule featuring the Jets, Jaguars and Bears—the likelihood that the team will lose every game is remarkably small.
To answer that, one would need to know the expected win probability for the Browns in each of their 16 games for the 2017 season. Using Vegas betting lines to establish these probabilities would be a good option–since Vegas incorporates off-season acquisitions–but I am not aware of any sportsbook that has released betting lines beyond Week 1.
As a lesser option, one can use FiveThirtyEight’s NFL Elo ratings. As outlined here, Elo ratings can be used to estimate the expected win probability for any team in an NFL game. If one wants to know the Browns’ win probability for each of their 2017 games, these can be calculated using the following equation as offered by FiveThirtyEight:
Prob(Browns win) = 1 / (10 ^ (-(ELOBrowns – ELOOpponent) / 400) + 1)
FiveThirtyEight also suggests that home-field advantage should add 65 additional Elo points to a team’s score. How does this play out? Consider the Browns’ season opener against Pittsburgh. According to Elo, the Browns have an Elo rating of 1251 while the Steelers have a rating of 1646 (link; change the date to pre-Super Bowl). Adding 65 points to the Browns total to account for the game being played at FirstEnergy Stadium, the calculation is such that:
Prob(Browns win) = 1 / (10 ^ (-((1251+65) – 1646) / 400) + 1) = 13.0%
Repeating the estimation for the rest of the schedule, Elo suggests that the Browns—as things stand today–would be underdogs in every game this season. Their best chance of winning would come at home against Jacksonville (49.4% probability) or the New York Jets (34.8%).
If these probabilities seem low, that’s because Elo does not account for off-season acquisitions or expected development by a team’s young players. This simply represents a snapshot of the talent on the Browns roster as of the end of the 2016 season. Considering that the Browns will likely field a much improved club in 2017 with the additions of Kevin Zeitler, J.C. Tretter and the 2017 draft class—and other clubs (e.g., the Jets) are likely to decline—the Elo system offers a very conservative forecasting method for the Browns’ 2017 season.
By estimating the expected win probabilities of the Cleveland Browns for each game on their 2017 schedule, one can derive the expected probability distribution of the club’s record. Based on end-of-season Elo ratings—again, this does NOT account for off-season acquisitions or expected off-season player development—Table 1 provides the Browns’ expected record based on their 2017 schedule of opponents. For simplicity’s sake, I assume that each game is an independent event.
As demonstrated in Table 1, using FiveThirtyEight’s NFL Elo ratings would suggest that the Browns would have a 1.73% probability of going winless over the course of their 2017 schedule. However, Cleveland’s off-season acquisitions, coaching changes and expected player development should produce a more competitive Browns team in 2017 than was put on the field at the end of 2016. As a result, even this small expected probability likely overshoots the Browns’ probability of going 0-16. Once Vegas posts its betting lines for all 17 weeks of the season, it would be possible to generate an updated forecast that corrects for these off-season developments; I strongly suspect that number will decline from 1.73%.
The NFL schedule game is a simple way to evaluate a team’s schedule. When the schedule is released every April, many of us go down the Browns’ schedule and try to figure out where the wins will come from in the upcoming year. I do it too. It’s a fun and harmless way to think about the upcoming season. And if it serves as click bait for media outlets, that’s perfectly reasonable for its entertainment value.
But the NFL schedule game presented as real “analysis”? The outcomes espoused as valid predictions? Come on. The shortcomings of this approach are obvious.
And the idea that the Browns will go 0-16? Please. The ESPN Cleveland article and headline are hocking a prediction that would seem to have near a 1% likelihood of occurring. Yes, 1-in-100 probabilities do occur and the estimated probability isn’t yet close enough to zero for us long-suffering Browns fans. But the ESPN Cleveland article clearly should not be interpreted as a valid—or realistic—prediction about the upcoming NFL season.