Thursday, January 27, 2011

Answering Mark Bradley

Mark Bradley is surprised that the Packers opened as a slight favorite over the Steelers:

The AFC champ has won nine of the past 13 Super Bowls. Pittsburgh was the No. 2 seed in the AFC, going 12-4. Pittsburgh has won two of the past five Super Bowls, and now, facing a wild-card qualifier that won two fewer games this season, Pittsburgh is …

A 2 1/2-point underdog.

Mark, you’re just looking at the wrong numbers.  Someone with serious skin in the game, i.e. the sharps who would kill the sports books if they based their lines on the wrong numbers, won’t look at who won the last 13 Super Bowls.  Maybe that issue was relevant in the 80s and 90s when the NFC won every year, but if the AFC had moved ahead of the NFC, that era has ended, as the NFC has won two of the last three.  The NFC was a whopping four games under .500 against the AFC this year, despite being dragged down by the NFC West. 

Sharps also won’t pay too much attention to the record of the two teams.  When we’re looking at a season of 16 games, there is the potential for a lot of noise in the records.  One or two bounces of the ball can cause one team to go 12-4 and another team to go 10-6.  Sure enough, the Packers went 4-6 in one-score games while the Steelers went 6-2.  Did that stat show that the Packers lacked a magical ability to win close games?  Probably not, since Green Bay has won a pair of one-score games en route to the Super Bowl.  In short, no one putting a $50,000 wager on a game is going to base his bet on the teams’ records.

So what it a sharp going to consider?  Yards per play.  Here is a reprint of a Chad Millman column making this point regarding the 2009 Michigan-Michigan State game:

Michigan State should have been the favorite when it opened. In fact, wise guys played some money on Michigan at first just to move the line a bit, so they could go back and play the other side for a better price. They knew most sharps would be on Michigan State. There were guys doing this with Cal and Oregon last week. Cal got bet up to 7.5 by wise guys looking for a better price because they were so sure of Oregon. Now, most college sharps build their math models around average yards per play. If you look at Michigan, it has gained 6.1 yards per play and allowed 5.5. Michigan State has gained 6.6 and allowed 5.1. Plus, I think MSU has played a tougher schedule, so that's why this game changed favorites.

Michigan came into the game at 4-0, having won a pair of nail-biters.  Michigan State was 1-3, having lost three tight games.  The yards per play numbers indicated that Michigan State was the better team.  Sure enough, the Spartans dominated the game and only won in overtime because Mark Dantonio did his best to make the game close.  Here is a podcast in which Millman explains in great detail why yards per play matters. 

So what does yards per play tell us about the Super Bowl?  Over the course of the season, it tells us that Pittsburgh was the best team in the AFC (+1.1 YPP) and that the Packers were one of the three best teams in the NFC (+.6).  However, YPP also shows that Green Bay has been dominant in the playoffs, as they were +1.0 against Chicago and +2.4 against the Falcons after a –.8 against the Eagles.  Pittsburgh was –.6 against the Jets after a +1.4 against the Ravens.  Green Bay has been great for the past two weeks, whereas the Steelers were a little lucky to survive against the Jets after being dominated in the second half.  So, yards per play reflects that the Steelers might be a slight favorite, but the Packers are the hotter team.

So what if we look at points-based models?  This is where we get the answer.  Sagarin would make the Packers a two-point favorite.  SRS would make the Packers a favorite of a half a point to a point.  When Green Bay has won, they have won comfortably.  The Steelers have won a number of close games.  We would expect these results from points-based rankings.

The overall point is that Bradley isn’t looking at the right numbers.  He is looking at the small sample – overall record – instead of the big sample – points and plays – that will give us a better sense of how good the two teams are.  I have to admit that I was a little surprised to read Bradley’s column.  He’s surely smart enough to realize that there are better ways to set a spread than “Pittsburgh was a higher seed and the AFC has been better over the past 13 years.”  His own sidebar reflects that he reads a lot of numbers-based sites (KenPom, Peachtree Hoops, Beyond the Boxscore, etc.), so he’s clearly aware of the blogosphere’s trend to being a reality-based community.  I’m inclined to reject the “can’t teach an old dog new tricks” conclusion, so are we left to believe that his editors don’t want him using meaningful numbers in his columns?   

1 comment:

Jeff said...

Good points. Bradley is also ignoring a key fact regarding the Packers--they were picked to finish way better than 10-6 in the preseason but were derailed the middle of the year by injuries. But most of the key injured guys are back (Clay Matthews) or have been adequately replaced (as with Starks at RB) so looking at their 10-6 record is irrelevant. By comparison, Pittsburgh may be worse than its record since their key injuries (most of their O-line) will be out for the Super Bowl.