THE BLOG
01/31/2014 07:21 pm ET Updated Apr 02, 2014

Counting on Math for Super Bowl Prediction

Which team is going to win the Super Bowl this year?

While fans (short for fanatics, right?) may have strong subjective feelings about the outcome, it is possible to come up with an objective prediction based in mathematics. One technique, among many -- and not original with me, is the method of "point-spread ratings." The idea is this: Each team receives a rating, and the predicted difference in score between two teams at a neutral site is simply the difference in their ratings. Shown below are the point-spread ratings for the top few teams and the bottom two through the recent Championship games:

Team Point-Spread Rating
Seattle 18.0
Denver 17.8
San Francisco 15.7
New Orleans 13.9
...
Cleveland 1.1
Jacksonville 0.0

So, if the Seattle Seahawks, the top-rated team, were to now play the Jacksonville Jaguars, the lowest-rated team, Seattle's margin of victory at a neutral site would be 18 points.

However, a correction must be added to account for home-field advantage. Historically, over the last several years, the home team has on average scored about three points more than the visitor. So in the example just mentioned, if the Seattle vs. Jacksonville game were played in Seattle, then we would predict Seattle to win by 21 points.

How does the method work? At the risk of over-simplifying, consider just two games have been played (and ignore any home-field corrections). Suppose Minnesota beats Green Bay by 10 points, and Green Bay beats Chicago by 5 points. As a consequence we would predict Minnesota to then beat Chicago by 15 points.

Since 1998 my yearly correct prediction percentages for game outcomes ('straight-up') has varied between about 60 percent and 70 percent. For the current year 63 percent of the regular season games were correctly predicted and 80 percent (8-2) of the postseason games were correctly predicted. Incorrect predictions in the playoffs this year were Cincinnati over San Diego and Carolina over San Francisco. In the Championship games, Seattle was predicted to beat San Francisco by 5.1 points -- the actual margin was 6 points, 23-17; and Denver was predicted to beat New England by 7.7 points, while the actual margin was 10 points, 26-16.

Who is going to win Sunday? I'm personally torn. I enjoy seeing the underdog win (go Seahawks!) but would like Manning to win another Superbowl ring before he retires (go Broncos!). What does my point-spread ratings method predict? It narrowly predicts Seattle over the Broncos by a .2 point edge -- although with that margin perhaps we should expect overtime!

The full current set of ratings for all 32 NFL teams may be found here. It is updated weekly over any NFL season.

Below are a few comments, including limitations, on the point-spread ratings method as shown on the website above, in no particular order:

(1) The Minnesota/Green Bay/Chicago example above is referred to as the transitive property to mathematicians. This property, as many fans already know, does not generally hold true in sports. In general, a given team may 'match-up' better with one team rather than another.
(2) The method as implemented counts each game as equally important. Championship teams, of course, are generally those that 'peak' near the end of the season, so success may be better judged by what happens later in the season.
(3) A better home-field correction would involve corrections for individual fields rather than a single averaged correction over all fields. That is, a correction factor by stadium could be used.
(4) The method as implemented only uses scores for the current season. In fact, it usually takes two or three weeks to have enough comparative data to fully rank all 32 teams.