Expert Commentary

Touchdown! Analytics in College Football

Hook 'em Horns! Roll Tide! O-H-I-O! T-R-O … J-A-N! USMA Rah Rah!, 1-2-3-4, 1-2-3-4, C-L-E-M-S-O-N! T-I-G-E-RRRR-S! Boomer Sooner! Go Irish Go!

It’s that exciting time of year again—college football season.


Big Data and Predictive Analytics
September 2016

By the time you're reading this, many games will have already been played with outcomes no one expected. Underdogs will win. Heavy favorites will fumble. Players will have been injured. Controversies will have developed. Pundits will be embarrassed. Lee Corso may have put on the wrong team helmet.

College football is fun and a big money machine. Everyone and their (bull)dog (e.g., Uga, Rhett, Butler Blue II, Jack, and Handsome Dan) endeavor to leverage the sport for their business ends. Clothing, food, tchotchkes, parking spaces, advertising, TV shows, betting, and travel directed at National Collegiate Athletic Association football followers ride the coattails of the mania.

"Comin' to Your City"

How do these businesses know where and when to invest their money? Is it no wonder ESPN College GameDay isn't announced Sunday or Monday for the following week? If a team gets on a winning or losing streak, it can make or break businesses connected with that team. The major football ranking polls shift weekly, building to a crescendo culminating after the last league tournament championship in early December. People are continually trying to predict where the next broadcast will be hosted. (See Predicting College GameDay Visits for 2016 College Football Season.)

It's hard to miss the story about the beginnings of analytics in sports as depicted in the book/movie Moneyball. The Oakland Athletics first deployed analytics in 2002 led by then General Manager Billy Beane. Mr. Beane sensed there should be statistical evidence to be able to predict the future productivity of players so they'd know who to draft. Since then, nearly every sports franchise in every sport has adopted some analytic strategy.

Premium Decisions

Our industry has gone gangbusters over analytics since the mid-1990s. Well, let's just say many companies have. Either way you look at it, the companies getting ahead have mostly been those who've embraced analytics and related technologies as core competencies and leveraged their investment in various processes throughout the enterprise.

Some companies have analytic teams numbering over 100, while others struggle to find, and then retain, the one magic bullet "data scientist."

Others have merely dipped their toes in the water, perhaps confirming pricing strategies with generalized linear models but not yet embracing a data-driven culture. Underwriters, marketers, claims adjustors, and middle management of all the above are dragged kicking and screaming into deploying these technologies because of years' old traditions and a "we've always done it this way" attitude.

I remember, however, studying business cases in college where the case conclusion was insanely "obvious." Most of us followed the most obvious trail we were led down and came up with the same erroneous answer because we didn't explore the "not so obvious" alternatives. We learned from trial by fire that it's smart to look for the sometimes opposite answer than the obvious as there's usually at least some truth that's getting overshadowed by the obvious.

My point is to be supportive of the "analytical naysayers" by considering there may be some truth in their approach versus taking a 100 percent data-driven approach because the "numbers don't lie."

In support of this hypothesis, I will reference a study conducted by Valen Analytics and documented in "The Perfect Pair: Human Judgment and Analytics" by Bret Shroyer, which was published in 2016. The paper illustrates the power of combining analytics with human judgment in the case of applying debits and credits or other premium adjustments to workers compensation risks. The study showed a lesser precision of both the pure data-driven prediction and the judgmental (underwriting) prediction as compared to the combination of both inputs. The article makes the statistical case for the underwriters using their judgment in combination with information from predictive models to yield the best results. This begs the question of how to do so successfully. (My next article will be about this subject.)

From Gator to Brutus

Rob Oller in the Columbus Dispatch captures another case in point in the August 24, 2016, sports commentary, "Stats Guru Says These Buckeyes Don't Add Up." Mr. Oller makes note that college football, Columbus's revered Ohio State Buckeyes included, is also deep into using analytics to select their best players. However, he states that Coach Urban Meyer is not in favor of using analytics in place of using his judgment and senses as well as that of his coaching staff.

Last year, based on coaches, press polls, and statistical analytics, the Buckeyes were ranked first going into the season. Well, Alabama did it again with Ohio State finishing fifth in the overall final rankings. In 2016, Mr. Oller cites rankings published by Bill Connelly, an analytics insider for SB Nation. He's got the Buckeyes ranked 14th (coaches and the Associated Press have them ranked fifth and sixth, respectively). Of course, the Buckeye Nation is furious at Mr. Connelly's assertion, placing their "judgment" ahead of his statistical approach.

I have not yet seen a model predicting the best coach. However, the AP does have a poll ranking the coaches that employs a "model-like calculation." The rankings involve more than just a sparkling resume, but success matters most. Body of work is worth 15 points; recruiting, 10 points; an ebullient personality, 8 points; newsmaker, 6 points; and catchall (for example, friendliness toward media), 2 points.

In this poll, Coach Meyer ranks second, bested by Nick Saban of the University of Alabama. The difference was 1 point of the tally, but it's obvious the success of winning 4 national titles in the past 7 years is hard to beat, despite the other factors.

for·tu·itous

\fȯr-ˈtü-ə-təs

Adjective

Happening by accident or chance rather than design.

"The similarity between the paintings may not be simply fortuitous."

Synonyms: chance, adventitious, unexpected, unanticipated, unpredictable, unforeseen, unlooked-for, serendipitous, casual, incidental, coincidental, random, accidental, inadvertent, unintentional, unintended, unplanned, unpremeditated, "a fortuitous resemblance"

The Best of Both Worlds

The world of sports is no different than the insurance industry. Losses are certainly influenced by risk characteristics well proven over time. We sometimes look at risks statistically and follow these analyses like it's the only truth. In football, there are still humans who might wake up feeling anxious or stressed. They compete with other humans that might tweak an ankle or notices his family in the stands and errors by over extending themselves beyond their capability.

There is heart, there is willingness to go the extra mile, and there is upbringing. These performance factors cannot be coded into a data model to be bucketed. They must be observed and "measured" by a human to assess the combined effect of the nonquantifiable factors plus the quantifiable ones for a complete perspective. These nonquantifiable factors are what can separate the statistically similar from each other. Losses are also influenced by uncertainty.

The same holds true for insurance risk performance. We try to assess these behavioral factors for personal lines from prior loss reports, driving records, or credit reports. But how do you distinguish between people with clean driving records and high credit scores from each other? So many losses are the result of something immeasurable from any statistic. Agents used to serve as front line underwriters, but that process is also becoming extinct.

For business risks, the immeasurable becomes even more complicated. Pride of ownership, customer service, business partnerships, business motivation, retirement plans, succession planning, and use of environmentally friendly processes are all risk factors that may not be able to be measured. But they can be observed and considered in the underwriting process. Underwriters and agents can balance these factors with the quantifiable model outputs for even better results than either process on their own.

Put Me in Coach

"I'm ready to play." But how do I convince the coach to put me in the game? Surely stats will speak for themselves: 40-yard sprint speed, ball control, blocking ability, physical stats, lateral quickness, reach, vertical leap, bench press, ball drops, sacks, etc., are all there for the record. What about the impact of extra time in the gym, knowledge of the playbook, ability to play through pain, dedication to team, acceptance of responsibility, and leadership skills? These are the things Coach Meyer looks for in his go-to players. Yes, you don't get on the field without good stats. Football, like insurance, is competitive. Competitive people will learn how to supplement the numbers with intuition and game knowledge.

Thank goodness for the mystery of it all, or Saturdays wouldn't be so exciting.

9/20/2016 - Cases in point (e.g., Oklahoma, Louisville, Houston)? And it's only just 3 weeks into the season!


Opinions expressed in Expert Commentary articles are those of the author and are not necessarily held by the author's employer or IRMI. Expert Commentary articles and other IRMI Online content do not purport to provide legal, accounting, or other professional advice or opinion. If such advice is needed, consult with your attorney, accountant, or other qualified adviser.

Like This Article?

IRMI Update

Dive into thought-provoking industry commentary every other week, including links to free articles from industry experts. Discover practical risk management tips, insight on important case law and be the first to receive important news regarding IRMI products and events.

Learn More



Navigation

Social Media

User ID: Subscriber Status:Free