College Football Analytics: Predicting Exciting Games (Week 2)
Using numbers to plan our Saturday viewing schedule
A college football Saturday is beautifully chaotic. Not only do you have your own team to watch, but you have dozens of other games, all chock full of opportunities for excitement. With numerous games happening at the same time, it can be a little overwhelming figuring out what to watch in order to find those exciting games. Luckily for us, we have a way to predict which games have the best chance to become mega thrillers you don’t want to miss.
To accomplish this mission, we need a way to measure games for their “excitement”. Fortunately for us, collegefootballdata.com has a metric called the “excitement index”. Here is the definition behind the metric:
“Excitement Index is a measure of how exciting a game was to watch. It accomplishes this by measuring swings in win probability throughout the course of the game. The more extreme swings between both teams, the higher the excitement index will be.”
Now that we have our metric, we are going to train a model (a random forest model, which you should be familiar with from the explosive forest) to classify games based on what percentile the excitement index fell into. We are going to split up the percentiles into the following groups:
An Excitement Index in the 80th percentile or higher: “Game of the Season Potential”
Excitement Index from the 50th percentile to the 79th percentile: “Must Watch Potential”
Excitement Index from the 25th percentile to the 49th percentile: “Good Potential”
Excitement Index less than the 25th percentile: “I’ll Pass”
Feature Importance
The model had an error rate of about 10%, which is pretty good for what we have set out to accomplish (if its wrong you still got to watch college football, and that beats a lot of other activities!). Here are the variables used and how important they are to the models classifying process.
The spread of the game is by far the most important variable to exciting games, and you don’t really need a model to arrive at that conclusion. Closer games are going to see more swings in win probability, which means more exciting games and a better chance at an exciting finish. Following the spread are my own in house power ratings for both the home and away team. The TLDR version of my ratings are taking my preseason ELO model and beefing it up to include things like opponent adjustments and returning production.
The only variable that had virtually no importance was whether the home team was favored or not. Given that this excitement index is a measure of swings of win probability, this seems to check out. Of course, if your favorite team is favored at home and they win, you’ll probably think that was one exciting game!
Now that we know what is important and have a working model, let’s get to the games! Unfortunately this week there were no “Game of the Season Potential” games so we will start with the “Must Watch Potential” games:
Ok, so before you look at these teams and yell and scream at me, let’s look through this list. Every matchup in this list is within a touchdown in terms of the spread, and we have some QB’s coming off REALLY good performances in week 1. The one game I will definitely have my eye on is Kentucky-Missouri. Connor Bazelak had a slightly above average performance in terms of QBR (59.3), and he should only improve as the season goes on. Will Levis (82.9 QBR) is actually top 10 in the country in terms of QBR and EPA/Play. Again, it was only one week of play. But both of these QB’s came into these seasons with high expectations, and definitely ones to keep your eye on.
The Group of 5 matchup on this list i’ll be keeping tabs on is the Army-WKU matchup. The clash of polar opposite offensive styles. The triple option vs the Air Raid. Western Kentucky QB Bailey Zappe finished week 1 with a 92.7 QBR, and someone you need to keep your eye on as the season unfolds. If Army goes up early, forcing WKU to get more aggressive passing the ball, they have all the ability to do it. This game could go down as the “sneaky amazing game that people didn’t watch”.
These are the games the model classified as having the potential to be “good”. In my mind, this means you feel satisfied watching two teams you did not have any connection to leading into the game. There are a lot of games to pick from in this category, including the College Gameday stop Iowa State vs Iowa.
SMU vs North Texas isn’t expected to be a close game judging by the 22.5 point spread. I do want you to know that SMU QB Tanner Mordecai (formally of Oklahoma) is coming off a historic QB performance. As in the highest EPA/Play of the CFP Era:
I know, I know, its Abilene Christian. You would most likely filter this game out because of a FCS team. But look at the numbers! Absolutely clean slate on lost EPA, 74.3% success rate, and 65.7% of his plays were explosive. He will come back down to earth, but is still worth a look.
Iowa-Iowa State should be a college gameday game that lives up to its hype. Iowa dismantled an Indiana team that had high expectations on their own, and Iowa State came into this season believing they could reach new heights as a program. Even with a low point total, this game should see plenty of win probability swings
There you have it! A list of games to hold you over before or after your favorite team plays. A good strategy would be to pick a game from each list and give them a go. You never know you may end up finding some enjoyable football!
It should be noted that “exciting” games is a pretty subjective term. We all have things that we deem “exciting”, and some of those things could be things other people don’t find exciting. The excitement index measures swings in win probability, but doesn’t tell you whether a team passed their way back into a game, or took a lead running the ball. In the end, a good weekend of football with friends and family (be safe and healthy) is ALWAYS fun and exciting!
If you want to dive in to the data like I do, check out @CFB_Data and @cfbfastR on Twitter, where you can learn how to get started in the world of College Football data analysis!
If you want to see more charts and one off analysis, follow my twitter page, @CFBNumbers
Cover Photo Image: Jack Zinsky (https://studentunionsports.com/the-2013-iron-bowl-revisited/auburn-field/)