In-Game Win Probabilities
And how we calculate them
Good morning and happy Thursday. There were some crazy games this past weekend.
Louisville came within five yards and a better pair of cleats away from beating Clemson
LSU took Alabama to the brink but was a hail-mary short
Purdue upset yet another Top 5 team in Michigan State
Wake Forest’s 8-0 streak came to an end in a no-defense-necessary 58-55 loss to UNC
And South Carolina took down the Gators
Needless to say, this is why we love college football. And even though it was played at a reasonable hour, I regret to inform you I still didn’t watch the Oregon-Washington PAC-12 matchup. It’s tough to compete with ‘Bama-LSU, despite how far LSU has fallen from just two seasons ago.
But just how close were some of these matchups? Let’s take a look at the in-game win probability charts, and talk about win probability in general while we’re at it.
The first one is my personal favorite. A game that was close down to the very end. Clemson went in with a slight edge, but things swung in Louisville’s direction, largely thanks to the incredible play of Louisville’s Malik Cunningham—who if you haven’t seen play yet you really need to check out. Cunningham was compared to Lamar Jackson, but with better passing ability. Now that’s high praise.
If you’re wondering how this chart is generated, here’s a primer. For each play, we use a model that takes into account a bunch of factors in the game and weighs them all compared to games it’s seen in the past, and then spits out how often teams in that situation win or lose the game. This is the in-game win probability. Here are the factors:
Points leading or trailing
Time left in the game
Down and Distance
Yards from your opponent’s endzone
Home and Away timeouts remaining
Your pregame win probability according to your Elo rating
The pregame spread
Whether the game is over or not (this is to help make sure the win probability goes to roughly 100% by the end)
Let’s take a look at another game.
This one is painful for Gators fans. Florida was heavily favored heading into the game, by about 19 points! Elo also had them as huge favorites, with a 79% win probability despite being away from home. That’s largely why the start of the in-game win probability graph is down around 15-20%. However, by halftime they were down 30-10 and things were not looking great. Even a 19-point favorite is unlikely to come back from a 20-point halftime deficit, which the model knew.
While the final score showed a close game, this win probability chart says otherwise. Probably due to the fact that Alabama was a 91% favorite by Elo and a 28.5 point favorite. ‘Bama also held a 20-7 lead until deep into the 3rd quarter, where you can see LSU’s chances creep back up a little bit. But in the end, although LSU had a shot, it was still a hail-mary, and the model likely took that difficult field position into account when declaring the game all but over on the final drive.
Another number to note on these graphs is the post-game win probability. This stat is calculated by collegefootballdata.com and tells you how likely to win the winning team was, simply based on how they played. So in this case, Alabama did get a bit lucky to come up with the win with only a 60% post-game win probability. This means that if they play that game that way 10 times, they end up losing 4 of them. Lucky for them, LSU wasn’t playing great either. Clemson’s 23% post-game win probability was even worse. They got really lucky that Malik Cunningham slipped on the last play, and that he was injured and out of the game, stalling out one drive resulting in a missed FG, and another leading to a punt. Those two possessions may have made all the difference.
Here are the last two games, just for fun. You can tell that the Wake-UNC game was a back-and-forth, exciting game just by looking at the chart, and you might be able to say that Michigan State was a bit unlucky to lose to Purdue. The postgame win probability indicates that the result was more of a toss-up.
These charts can be both cool to look at, and informative. However, even a few weeks ago, our charts didn’t always look like this. Sometimes, for one reason or another, they looked like this:
If you’re interested in the data behind the charts and how we’ve improved them throughout the season, then be sure to checkout an upcoming article on Staturdays.com that will go into the details of how they’re built and how we’ve improved them to make them more accurate. For now, let’s look at this week’s slate of games.
It pains me that we’re already in the double-digits of the season. Thankfully, with bye weeks and conference championships, we’re still looking at a solid 5 more weeks of football before bowl season.
Slight note on the expected values this week: I’ve excluded any games that have under $1 of positive expected value, because I found that this is actually the most profitable betting strategy. Some of these games where the value is in the pennies are just not worth the risk, so from now on I’ll be excluding them. This will help avoid those pesky negative weeks. And it’ll be less work on your part. It’s a win-win!
North Carolina at Pitt should be an interesting one. They showed us this week that they can score with the best of them when they’re hot, but are unwilling to play defense. So take your pick.
Let’s look at the individual win probabilities. For non-Power 5 conferences, click here.
We have one mismatch in the ACC in Boston College, who are 2-point dogs to Georgia Tech.
No mismatches here, but Michigan at Penn State in the “helmet-stripe-out” is looking to be the most competitive of the weekend in the Big Ten.
Arizona State is a 6-point favorite, but Washington is at home which shifts things slightly in their favor for Elo.
After a major win last weekend, Elo has South Carolina as the slight favorite over Missouri, who didn’t fair nearly as well against the Georgia Bulldogs.
And after going toe-to-toe (kind of) with Alabama last week, LSU open as underdogs to Arkansas this weekend. Elo disagrees. Personally, I think it’s a toss-up, as the spread implies.
Thanks so much for reading. Hopefully I didn’t bore you with too much win probability talk. Savor the college football this weekend. We’re getting up there in the weeks!
Talk next Thursday, and make sure to follow @staturdays throughout the weekend to see those in-game win probability charts come out in real time after each matchup.