FPI, LSU and how probability in betting can work for you

Four years ago, in April 2019, I sat on the Daily Wager set and made the case that former Ohio State transfer Joe Burrow, coming off a season at LSU in which he ranked 14th in QBR, was a little better than his reputation indicated. Understatement of the decade, it turned out.

It was part of the justification I was making for our models — ESPN’s Football Power Index and the Allstate Playoff Predictor — believing that LSU was a value to win the national championship at 25-1 odds.

And in retrospect, that segment was an example of perhaps the most valuable lesson analytics can teach us about sports betting: think probabilistically.

The case was three-fold.

1. Joe Burrow played better the year before than he was given credit for.

2. LSU could reach the playoff without beating Alabama.

3. This was the most important one; that LSU had a 25% chance to actually beat Alabama and that when thinking about that game, you shouldn’t round that 25% chance down to zero. Alabama was probably going to beat LSU that year, our models thought. But not definitely.

In other words: assigning a win or loss to every game on a team’s schedule is a perfectly fine game to play for fun, but it’s the exact opposite of what we should be doing in betting.

Ultimately, the models’ conclusion was that LSU had a 7% chance to win the national championship, while the odds implied a 3-4% range. That’s value.

You probably know what happened next (well, like eight months later), but just in case: LSU did win over Alabama — 46-41 in Tuscaloosa — and then went on to win the national championship. The bet hit.

The lesson here is not that our models are clairvoyant. The LSU argument is incredibly easy to make … in retrospect.

In fact, you could completely flip the numbers around, if you want. ESPN Analytics said there was a 93% chance that LSU would not win the national championship, and yet they did. How bad is that! FPI also gave Alabama a 75% chance to beat the Tigers and yet they still lost. And in that same segment, I also mentioned both Clemson and Michigan as other teams the model liked; it hadn’t singled out just LSU.

One last piece of evidence that this is all much easier to say with the benefit of hindsight: I was so confident in what I had said about LSU that, a couple months after that segment when I happened to be in a legal betting state (weren’t as many back then!), I placed a single dollar on LSU to win it all. At 50-1, mind you, but still: one single dollar. When my friend I asked why I’d bet only the one dollar I said, “If I’m going to light money on fire I’d prefer it were not very much.” Ouch.

The point is that sports are uncertain. And as sports bettors, we should embrace that uncertainty and be price conscious. ESPN Analytics found its way to LSU via not denying the possibility that LSU could upset the Crimson Tide and even win the whole thing, but also because of the price: 25-1 (it had even been 60-1 at one point in the offseason, per SportsOddsHistory.com). Had the price been 10-1, we never would have mentioned LSU at all. It’s about probability and value.

Right now you might be thinking: OK, that was all the past. What does the Allstate Playoff Predictor say about 2023? It has two words for you: Ohio. State.

While we typically find that FPI and the Allstate Playoff Predictor are fairly in line with the betting markets, the models have an unusually large disagreement with the Buckeyes’ odds. Ohio State currently is +750 to win the national championship at Caesars Sportsbook, but ESPN Analytics makes the Buckeyes’ chances at a whopping 36.7%.

Some quick math: at +750, the breakeven percentage on the Buckeyes is 11.8%. In other words: if you truly could predict the future probabilistically, you would need to know that Ohio State has greater than an 11.8% chance to win the national title to justify betting on them.

However, the odds also imply that Ohio State’s chances are not quite that high. Caesars has a vig on the market, which means that if you add up the breakeven percentages of every team it sums to more than 100%. Thus, the implication is that Ohio State’s chances are slightly less than 11.8%.

Why so high on Ohio State? You can read more about that here but the abridged version is that FPI believes the Buckeyes have the best non-QB offensive returning production in the nation — led by WR Marvin Harrison Jr. — and have an impeccable track record of efficiency and had excellent recruiting in recent years. Crucially, Ohio State is by far the best team in the Big Ten, while teams 2-4 in FPI’s rankings are all in the SEC.

FPI gives the Buckeyes a rating of +31.5, meaning it would anticipate them being an average FBS team (someone like Washington State or UTSA) by 31.5 points on a neutral field. Or Alabama by about a field goal.

But remember that uncertainty we’re emphasizing? It’s here, too. When we simulate the season thousands of times to form our projections, the Buckeyes (and every other team) won’t always have a rating of exactly 31.5. Sometimes it will be a little better, sometimes a little worse. Occasionally it will be much better or much worse, and it’s important to capture those iterations of the season that could unfold, too.

I’m not here today to convince you to follow along with our models this season or bet on Ohio State. That’s up to you. I’m just making the case to think probabilistically.

Ohio State probably won’t win the title anyway. We’re 64% sure of it.