A nerd's look at predicting the SEC, Part II: Point differential

Remember ninth-grade geometry class?

All those jokes and references you made about the “Pythagorean Theorem” that really weren’t all that funny or clever, just so you could associate yourself with the $100 term you learned in math?

If you’re nodding your head, you’re worthy of the in-vogue nerd label. In case you’re oblivious to cultural trends like the dad bod and the popularity of nerds (not the candy), being intellectually obsessed is well-received in general culture as well as in the current sports climate.

We’ll focus on the latter for now, as we’re spending this week looking at a few of the game-by-game, outcome-related numbers that remain useful, even in college football.

(For an introduction on the series, an explanation of the place of analytics in college football, and Part I — predicting the SEC based on every team’s record in close games — go here.)

Speaking of that Pythagorean Theorem from geometry class, there’s a related equation that’s an important predictor of team performance. Created by sabermetrics demigod Bill James for baseball, current Houston Rockets general manager Daryl Morey modified the Pythagorean theorem for football in the 1990s.

The formula translates a team’s points scored and points allowed (collectively, in one season) into an expected winning percentage. A team’s point differential in its most recent season is a much better indicator of its future winning percentage than a team’s actual record.

The problem with predicting future performance based on a team’s record is that individual games are weighted equally.

For example, the week after Georgia was forced to suspend then-Heisman Trophy candidate Todd Gurley, the Bulldogs traveled to eventual SEC East champion Missouri and blasted the Tigers, 34-0.

That win counts just the same as a 35-32 squeaker at home against unranked Tennessee, although the former is much more impressive than the latter. No one objective would claim that Georgia played as well or better against the Vols.

For you nerds, the Pythagorean theorem formula is like assigning a specific number grade rather than pass/fail. (After all, wouldn’t you be flustered if you scored in the 97th percentile, yet you received the same label as Johnny Snotnose, who skated by in the 71st percentile?)

The formula looks like this:

Points For^2.37 / (Points For^2.37 + Points Against^2.37)

Let’s use the 2013 Auburn Tigers as an example. In coach Gus Malzahn’s first season as head coach, the team skated past Georgia and Alabama late in the season, won a surprise SEC championship and lost in the BCS title game to finish 12-2.

The ’13 Tigers scored 553 points and gave up 346 points. So our formula is 553^2.37 / (553^2.37 +346^2.37) = .752. That was Auburn’s expected winning percentage from its point differential. The team played 14 games that season, and 14 x .752 = 10.5. In other words, Auburn outperformed its true level of performance by about 1.5 wins.

Of course, the Tigers finished 8-5 last season, representing a four-win decline. Although that seems extreme, it’s actually right in line with recent history.

Since 2012, seven teams have outperformed their true level of performance by at least 0.5 wins. Those seven teams saw their total wins decrease by 3.7 on average the following season. So, given Auburn’s luck in ’13, the team’s four-win decline is pretty normal.

TEAMS THAT OUTPERFORMED TRUE LEVEL BY >0.5 WINS IN 2012 AND 2013

Notice that five of the seven teams that fell into this category in 2012 and 2013 dropped off by at least four wins the next year.

Not a single one of these teams won more games than it did the year before, and the only team that achieved a +/- of zero — the 2013 South Carolina Gamecocks — went from 11 to 7 wins in 2014. Yikes.

PYTHAGOREAN THEORM, 2012 AND 2013

The predictive nature of this formula isn’t limited to outliers who outperformed their true level.

Here’s a look at every SEC team that either won fewer games than expected (underperformed) or more games than expected (underperformed) in ’12 and ’13.

That’s a relatively small sample size. But the formula has been proven by people far smarter than I am and with far more time on their hands. Feel free to extend it back a decade or more using SEC data and the results will be more or less the same.

As I mentioned earlier, SEC teams who outperformed their Pythagorean expectations by a wide margin in the last few years have predictably lost more games the following season.

BIGGEST YEAR-TO-YEAR GAINS, 2012-14

The formula also has successfully predicted every big jump in wins since the SEC expanded to 14 teams in ’12, many of them surprises.

HOW DOES THIS STUFF APPLY TO 2015?

The following chart depicts teams that either underperformed or outperformed their true levels in the 2014 season.

So what does this mean?

• Bad news, Mizzou fans. You too, Vandy fans, if you’re still out there. To borrow a mountaineering term used to describe extreme altitude, you’re in the Death Zone this season. The previous scenario, with the seven teams that lost nearly four wins on average the following year? That’s you in 2015. That’s not a good sign for the defending SEC East champions, an indicator that the team will fail to reach 10 wins this fall. (At least if you’re Vanderbilt, you only won three games in ’14, so a winless ’15 still would only equate to a three-win drop-off.)
• Arkansas, this formula is a bonanza for your 2015 hopes. Essentially, the Razorbacks performed like a 10-win team last season despite the 7-6 final record. So that’s a very strong indication that the Hogs are in line for more than 7 wins this fall.
• By the way, Arkansas hosts Missouri on Nov. 27. The Razorbacks, which lost 21-14 last year, opened as a seven-point favorite in Vegas, so the sportsbooks already agree with the projection.
• Other teams who underperformed last year include Ole Miss (-1.3 wins), Georgia (-0.9 wins), and Tennessee (-0.8 wins).

EVERY SEC TEAM RELATIVE TO TRUE PERFORMANCE IN 2014