Our New Preseason Ratings: The Methodology

• 2 comments - > Site Help, College Football, NFL - posted by Austin Link on August 5, 2011

Two days ago we unveiled a new and exciting tool on the site, projected end of season records for college football and the NFL.  These include not only expected standings, but also odds for each team to make certain milestones (Win the conference, reach the playoffs, go undefeated, etc.).

Once the season gets going our calculated power ratings will drive the game forecast, but before any games have happened we need something else.  To solve that problem we have developed a system to project what a team’s most likely power rating will be.  This system will assist in early season projections and will also factor slightly into our predictive power ratings.    Here’s a brief look at how it works.

To create the ratings, we compiled statistics across multiple years in dozens of different categories.  We then compared a team’s final rating from one season with data from previous years using regression analysis.  While many adjustments are small, or expected, there are a few factors which provide some interesting insight.

Previous Season Ratings

Power ratings from last year obviously help predict how a team will do in the upcoming season.  Teams that are good tend to stay good for a while.  Surprisingly though, strong ratings from up to 4 years ago also indicate that a team will do well in the future.  This impact means that teams that suddenly get better often regress back the next season.  St. Louis Rams fans beware, last year’s improvement may be more of an exception than a trend.

Previous Season Statistics

Certain statistics from the previous season can also bode well (or poorly) for a team the next year.  The most striking example that we’ve found so far is the impact of turnovers.  Teams that have a good turnover margin (both fumbles and interceptions had an impact) tend to lose that edge the next season, and get worse because of it.  With all the Stanford alumni around here, this trend is cause for concern after the Cardinal posted the 5th best Turnover Margin Per Game last season.

Player Turnover

Player  movement clearly has a big impact on how teams play, just ask the Miami Heat.  While the first round of ratings projections don’t do a spectacular job of taking personnel changes into account, it does adjust some.  Losing both Cam Newton and Nick Fairley to NFL while returning only 7 starters means that the Auburn Tigers will likely need more than miracle to go undefeated again this year.

What to Expect

In college football the projected ratings successfully explain 65% of a team’s rating in the upcoming season. The other 35% is due either to factors we haven’t accounted for yet, or random variation due to unexpected circumstances (injuries, bad bounces, etc.).  We consider this  a good level of accuracy, helped by the consistency of college teams from year to year.

A comparable analogy comes from the world of biology.  Using similar regression analysis, it’s been determined that 30% of a child’s height is explained by the height of it’s mother.  The other 70% percent is explained by the father’s height (31%), environmental factors, and as always random variation.  So we’re pretty happy being able to explain 65% of college football ability before the season starts.

It’s easier to compensate for player changes in recruiting class numbers or returning starters than to compensate for the skill of free agents added or lost.  As a result, the NFL projected ratings suffer a little bit, explaining only 40% of a team’s rating in the upcoming year.  That doesn’t mean the numbers aren’t useful though.  Combined with our sophisticated season simulation tools it’s very useful in estimating team upcoming records.

While they’re not perfect yet we’re happy with the state of our ratings projections so far.  The statistics behind them reveal trends that can bypass even the most educated observers.  Maybe best of all though is the high ceiling that the ratings have.  More factors, individual player projections, and a larger sample size will all combine to make the pre season ratings useful TeamRankings.com tool for not only this year, but the future as well.