Computers vs human judgment in picking college football games

Since the beginning of the 2014 season, The Power Rank’s college football prediction service is 56.4% in picking spreads and totals (137-106-6). Mike Craig and I use a computer model based on my algorithm that makes accurate adjustments for strength of schedule.

However, we never only rely on computers and numbers to make predictions. We always consider subjective factors like injuries and situations.

This reliance on both computers and human judgment is reminiscent of the world of chess and a New Yorker article on Magnus Carlsen, the 23 year old World Champion from Norway.

Computers and Chess

For most of the 20th century, the Soviets dominated chess. They relied on “focus, logic, and, above all, preparation” to produce world champion after world champion.

However, they met a new foe in the late 1990’s: the computer. These inanimate objects are all about focus and logic. When computing power hit a certain level, computer programs could use brute force search to pick the best move. This led to the Deep Blue computer’s famous win over World Champion Gary Kasparov in 1997.

Nowadays, computers always beat the top human players. When asked what strategy he would bring against a computer, a grandmaster said “I would bring a hammer.” Computers are also a training tool for chess players.

But Magnus Carlsen is different. As he told The New Yorker, the Norwegian champion never trains with a computer. He admits to preparing less for tournaments and relying more on his judgment.


Predicting football games

Predicting football games is more complicated than winning a chess match. Players don’t occupy a square on a board. Football also adds the element of randomness, as events like fumbles defy any quantitative attempts to predict them.

This makes human judgment even more important in predicting college football games than chess. For the college football prediction service, we always consider subjective factors like injuries and situations. For instance, we needed to make an adjustment for Baylor as they will start third string QB Chris Johnson against North Carolina.

Both computers and human judgment play a role in giving you the best possible predictions.

Last night to take advantage of bowl season discount

The prediction service, which has gone 55.9% (66-52-2) in 2015, is available for bowl season. The price goes up at midnight, Eastern time on Tuesday, December 22nd, 2015.

To take advantage of this early bird price, click here.

Study shows interim coaches have little impact on bowl games

Screen Shot 2015-12-18 at 11.00.21 AMIn an excellent article over on Sports on Earth, Ross Benes used my college football team rankings to study the impact of interim coaches on bowl games.

He looked at all bowl games since the 2005 season and split the 42 games with an interim coach from the remaining 295 games. He found the statistics of both groups to be roughly the same.

The Power Rank actually performed better in predicting the results of games with an interim coach (71% compared with 59% in games without an interim coach). However, this is mostly likely a fluke due to small sample size.

The study does suggests you can ignore coaching changes when predicting the winner of bowl games.

To read the article, click here.

The Power Rank 2015 Bowl Season Cheat Sheet


If you’re entering a bowl pool, you might be interested in the 2015 bowl cheat sheet.

The study by Benes uses my calculations based on points, or margin of victory adjusted for strength of schedule. In the cheat sheet, I combine these predictions with others based on yards per play. The larger ensemble of predictors makes for stronger predictions.

These prediction picked the winner in 76.5% of college football games this season. In addition, I expect the results of the study to apply to these predictions: ignore the impact of interim coaches.

For example, suppose you’re trying to pick the winner of Georgia versus Penn State. Georgia fired Mark Richt and installed wide receiver coach Bryan McClendon as interim coach. Georgia also won’t have the services of either coordinator for the bowl game.

It most likely doesn’t matter. My bowl cheat sheet predicts Georgia by 8.6 points over Penn State, which corresponds to a 74% win probability. Go with the Bulldogs.

To get the 2015 Bowl Season Cheat Sheet, click here.

Bowl predictions by win probability for 2015

You don’t have access to view this content

Will the Carolina Panthers go undefeated and win the Super Bowl?

Carolina Panthers seems like the best team in the NFL. They have an unblemished 12-0 with wins against Green Bay and Seattle.

The numbers love Carolina’s defense. They rank 2nd in pass defense by my yards per attempt adjusted for strength of schedule. Their secondary deserves most of the credit, as their pass rush ranks 15th by sack rate adjusted for schedule.

The pass defense has made a huge leap with the emergence of cornerback Josh Norman. The fourth year player from Coastal Carolina took over a starting role late last season and has blossomed into a shutdown corner this season.

However, other factors suggest Carolina is overrated.

Cam Newton and Carolina’s offense

The offense limits the ceiling of this team.

Cam Newton seems like an elite quarterback at the helm of an undefeated team. However, Newton has completed a below NFL average 58.4% of his passes, and Carolina ranks 18th in pass offense by yards per attempt adjusted for schedule.

The advanced statistics also don’t like Newton. ESPN’s QBR metric, which accounts for rushing plays, ranks Newton 14th in the NFL behind studs like Brian Hoyer and Ryan Fitzpatrick. A closer look at Newton’s numbers shows that the defense has made big plays after his turnovers.

A weak schedule

The schedule has also greatly helped Carolina’s pursuit of perfection. They play in a weak NFC South division, and drew a putrid NFC East out of division.

In addition, Carolina finishes the season with 4 opponents in these divisions: Atlanta twice, Tampa Bay and the New York Giants. They have a healthy 23% chance to end the regular season undefeated.

However, Carolina’s path to the Super Bowl gets more difficult in the playoffs. By my ensemble calculations that combine metrics based on points and yards, the Panthers would be a slight home underdog to Seattle and Arizona. In a neutral site Super Bowl, New England and Cincinnati would be favored to win.

Let’s not crown the Panthers Super Bowl champions just yet.

Advanced NFL statistics

The numbers cited in this article, such as sack rate adjusted for strength of schedule and predictions based on ensemble numbers, are available to members of The Power Rank. To learn more, click here.

Bleacher Report college football playoff predictions after week 13

Screen Shot 2015-12-02 at 3.38.31 PMOver on Bleacher Report, I both predicted the committee rankings on Sunday morning and calculated playoff probabilities based on the actual rankings on Tuesday night.

This week’s most interesting team is Ohio State. The idle Buckeyes need help from others to make the top 4. I put some numbers behind the scenarios.

I also preview North Carolina vs Clemson in the ACC Championship and Michigan State vs Iowa in the Big Ten Championship game.

To read this analysis, click here.