## 2014 Season Predictions for ESPN the Magazine

ESPN asked me to predict the 2014  season for their NFL Preview edition of their magazine. I was very hesitant because predicting the season to any degree is extremely difficult. I'm even on record as proclaiming that all pre-season predictions are "worthless." (More on that below). "You want me to predict which teams make the playoffs?" "Yes," they said, "in fact, we want you to predict the winner of all 267 games."

Then it got worse. "We want you to predict every score of every game."

I started doing some math in my head. There's 267 games in the season, including the playoffs, which means there's 2^267 different possible combinations of game outcomes in the season. While that might sound like a lot of different possibilities, it's even more than a human being could possibly fathom. Physicists and astronomers estimate there are about 10^80 atoms in the universe (that's 100 quinvigintillion to you and me). And the NFL season's 2^267 possible outcomes comes to 2.4x10^80, or about 240 quinvigintillion. Put simply, there more than twice as many possible outcomes to the NFL season than there are atoms in the universe. And that just refers to wins and losses, and doesn't even consider scores.

So how hard could it be?

It was hard to resist the challenge, even one as crazy as this. So I said ok. It was a good opportunity to experiment with a season simulation model. I thought I could reuse the season sim model for other purposes like our live playoff probabilities widget, which was a ton of fun last year.

Here's how I built all the predictions in the ESPN The Magazine:

A team strength rating was assigned to each team. This metric was in terms of expected net points per game. In other words, a team that is expected to beat an average team by a net difference of 3 points is a “+3” team. So when a +3 team faces a -3 team, the expected point spread would be 6 points in favor of the +3 team.

This metric was estimated based on a blended model of several factors, including:

-implied team strength generated by reverse-engineering opening Vegas spreads,which were already available for nearly every game in 2014 (This was a concept I first learned from Michael Beuoy who now runs the site Inpredictable.com),

-a regression of factors from previous year performance (some factors like passing efficiency tend to be consistent, while others such as special teams performance or defensive scoring are especially inconsistent and random), and

-minor qualitative adjustments for new information available regarding team composition not already accounted for (Josh Gordon’s suspension or Kiko Alonso’s injury, for example).

Each game match-up was converted into a win probability for each team, which considered home field advantage. This conversion was based on a model of historical win rates as a function of point spread.

The 2014 season was then played out as a Monte Carlo simulation model using those win probabilities 400,000 times. Prior to each simulation, each team’s strength rating was randomly varied to account for uncertainty in team strength assessment. For example, DEN could be a +5 team as predicted, but they could realistically be anywhere between +2 and +7 based on injuries and assessment errors. This element of uncertainty is appropriate because it honestly reflects the limitations of any system’s ability to predict. The uncertainty adjustments gravitated toward average to appropriately account for mean-reversion tendencies.

The teams that qualified for the various playoff seeds most often during the simulations were chosen to be the projected playoff teams.

I created a representative and highly plausible single season of game win/loss outcomes that resulted in the most likely playoff seeds. Care was taken so that there are the “right” number of 13-game winners, 12-game winners…4-game winners, etc you typically see in a season. There are the right number of upsets, home victories, winning streaks, losing streaks, etc. Win totals generally reflect each team’s overall rating.

Each particular game’s score was based on the offensive and defensive ratings of each team. These ratings were created in a manner similar to the overall team ratings. As you would expect, match-ups of strong offenses vs weak defenses tend to result in higher scores, and match-ups of weak offenses vs strong defenses tend to result in lower scores.

Scores for each game were drawn from a sampling of the actual score distributions for all games since 2008. This way, there are the “right” number of close games, blowouts, field goals, touchdowns, and the right number of scoring totals. For example, scores of 14, 17, 24, 31, etc. are very common, and scores of 12, 15, 19, 22, etc. are relatively rare.

Additionally, the winner and loser scores correlate in the same way they do in the regular season. In other words, some games tend to be high-scoring battles and some tend to be low-scoring standoffs.

Total team scoring for the regular season was tallied to ensure each team had appropriate scoring, scoring allowed, and net score season totals.

Playoff game winners were selected based on how often they reached each level of the playoffs in the simulated seasons (division, conference, and super bowl), and not necessary based on the better team in that particular match-up.

Ultimately, the end result looks a lot like last season. 2013 was a very unusual NFL season in that the truly best teams made the playoffs and then survived through the playoffs. Unlike previous seasons that featured lucky underdogs like the 2011 Giants or 2012 Ravens, the 2013 playoff field featured almost all very strong teams. Barring any highly notable and unexpected personnel changes, most of these same teams can be expected to be nearly as strong in 2014.

My purpose with the scores was not to try to outguess the odds makers or pretend to have any scientific, analytic way of projecting scores of every game. If I wanted to do that, team win totals would all cluster around 8 wins, and game scores would all be projected around 24-20. My goal was to present the most plausible, NFL season based on the information we have now. Do I really believe the Jets will beat the Chargers in week 1 by 24-3? I have no idea. But I do believe San Diego will most likely be 9-7 (but there's an 80% chance they'll finish with some other record). They'll  win a couple games they were supposed to lose and lose a couple games they were favored to win. And they'll have a game or two where their offense sputters.

Regarding my prior analysis that preseason predictions are worthless... It would be more than a little hypocritical for me to now claim any authority on such predictions. Please keep in mind that ESPN doesn't stand for Extra-Scientific Prediction Network. Note that the E, actually, stands for entertainment. Still, I made the best projections I could despite my prior remarks. While I stand by the analysis that showed how impossibly difficult preseason predictions are, I sincerely regret the snarky tone I used, particularly directed at Football Outsiders. The NFL's 16-game season is uniquely random among major sports, and any prediction system that can add value must be doing it well. And that's one of the many things FO does well.

Stay tuned for a cool season projection visualization dashboard.

### 11 Responses to “2014 Season Predictions for ESPN the Magazine”

1. Topher Doll says:

Looking forward to it!

2. bones says:

Would 2013 predictions have looked a lot like 2012? If so, given the roughly 50% turnover of playoff teams in recent years, is there a better way to simulate the parity / randomness / injuries / etc. than rolling weighted dice that have a tendency to repeat the previous year?

3. Trevor says:

"Do I really believe the Jets will beat the Chargers in week 1 by 24-3?"

That seems like something of a long shot.

4. JMM says:

So my method of throwing six red die and six white die with the total red count the home team score and the total white die count being the away team score isn't the most sophisticated method out there?

I guess we learn something everyday.

5. Michael Beuoy says:

JMM - Add a seventh red die and you would probably do no worse than your average pundit.

6. Anonymous says:

Hmmm...closer to 3^267. Every regular season game could end in a tie.
How many atoms are there in parallel universes ?

7. Dan says:

Extra-Scientific Prediction Network works, as long as you don't take "extra-" to mean "very" but instead use it as in extraordinary, extrajudicial, extraterrestrial, etc.

8. David Balcom says:

Very cool. What tools did you use to do this? I'd love to see nerd out over those spreadsheets.

9. Jack says:

Hey Brian. This must have been a beast to tackle. Great work. Could you provide us with the data for the odds of each game for the season? It may be on the ESPN site but the links there don't seem to be working for me, and the Tableau data doesn't break it down by game.

Thanks!

10. Nate says:

Jack:
I don't where Brian got his, but...

11. Brian Burke says:

Dan-you win!

David-I coded my own application to handle everything.

Jack-I don't plan to publish all the game probabilities prior to the season. They'll go up each week like always.