051: What I think I know, so far, about the length of college football games

Assumptions after analyzing six weeks of college football game duration data.

As I was waiting for the North Carolina and Syracuse game to kickoff just after 3:30 pm EST on Saturday, ESPN flashes a graphic that the game would be starting on the ESPN App. The LSU and Missouri game was three hours past its noon start time.

So, I tried following these steps¹:

1. Close YouTubeTV

2. Open ESPN App

3. Search for UNC game

4. Wait I need to login, so open - espn.com/activate - on my phone

5. Enter a six-character code

6. Authenticate through YouTubeTV

7. Select the UNC-Syracuse game again

8. See a screen that it hasn’t started yet despite it being 3:34 PM EST

LSU beat Missouri 49-39 and the game lasted three hours and 45 minutes. After the game ended, ESPN airs an interview with LSU head coach Brian Kelly before switching over to the North Carolina-Syracuse contest.

The average duration of a college football games this season is three hours and 24 minutes. This average is derived from 431 games this season that feature at least one FBS team.

The games are too long.

---

Trying to predict game length

Based on a variety of factors, I’ve tried predicting the game length. The model takes the total number of drives, plays, points, punts, point differential, and turnovers. It uses those inputs and tries to project the duration.

Here is how the model compares to the actual or observed minutes so far:

Total possessions is the strongest indicator in game time

The more possessions or drives in a game, the longer it generally lasts. Total drives is calculated by adding together the each team’s offensive drives.

NC State’s 48-41 win over Marshall in week six featured 37 combined drives. The game lasted three hours and 59 minutes.

Michigan’s 31-7 week four win against Rutgers had 15 combined drives. The game went two hours and 51 minutes.

In general, the higher combined point total translates to longer games. This can be noisy because a combined point total of 80 points could be a blow out or a back-and-forth type of game.

More punts might mean a game lasts longer too. This makes sense because a punt signals a change in possession or more drives. Lots of three-and-outs can take a long time.

BYU’s 14-0 win against Sam Houston State in week one included a combined 19 punts. The game went three hours and twenty minutes.

By and large, more plays means more minutes added to the game duration. The total plays adds together the rushing attempts and passing attempts from both offenses. This excludes special teams plays right now.

Fewer turnovers means longer games in general. The assumption here is a game with lots of turnovers means that it’s likely a blowout. Utah State beat Colorado State 44-24 in week six. Nine total turnovers in a game that lasted three hours and 30 minutes.

Closer games, with a smaller point differential, tend to last longer. This makes sense that games with a wide point margin often speed up towards the end. Toledo’s 71-3 win over Texas Southern last three hours and eight minutes.

Weather, replay reviews, and networks offer outliers in the data

You can plan a pretty picnic, but you can’t predict the weather.² The weather is responsible for most outliers on the graph.

A weather delay is at the discretion of the scorekeeper when submitting game duration. This makes it unpredictable, especially when there are long delays.

Eight games are listed as lasting longer than 300 minutes or four hours and twenty minutes. All eight of those games involved a weather delay and all occurred no later than week three:

• Miami (OH) 41, UMass 28: five-and-half hours

• Arizona State 24, Southern Utah 21: two hours and 50 minutes

• Boston College 31, Holy Cross 28: two hours and 15 minutes

• Notre Dame 45, NC State 24: one hour and 45 minutes

• Wake Forest 36, Vanderbilt 20: two hours

• Louisiana 41, UAB 21: one hour and 51 minutes

• West Virginia 56, Duquesne 17: two hours

• Marshall 31, East Carolina 13: two hours

There are other oddities in game duration too. It’s possible there are data entry errors³. For example, Miami (OH) beat Cincinnati 31-24 in week three and the game duration is 95 minutes. The game went to overtime. Confident it’s wrong, but see the box score and the duration entered:

Replay reviews are hard to track in the data
There no is event or detail added to the public play-by-play when a call is reviewed. These reviews add significant time and commercial breaks to the game.

I’ve listened to this Alberto Riveron interview with Ovies and Giglio a couple times, and after seven weeks, it doesn’t feel like the ACC has made the review process⁴ more efficient or transparent.

It’s anecdotal, however, we’ve seen simultaneous catches and fumbles that are reviewed, and these take more than a few minutes to determine the final call.

For example, see this blocked punt turned first down call from yesterday in North Carolina’s 40-7 win over Syracuse:

This play took about 20 seconds of game time and over eight minutes of real time to determine the final call. Two commercial breaks.

It’s hard to track these replay reviews because they’re hidden from public data that is available.

Does a game on a specific network take longer than other networks?
This is another part of game data that isn’t simple to track. Others have hinted towards tracking networks, but I’m skeptical there is a significant difference between networks. I do think television and the broadcast matters though.

For example, do Notre Dame games on NBC last longer than say ESPN? The difference is hard to track without doing it manually.

All models are wrong, some models are useful

This model is not a great fit for the data right now.

This model could improve by tracking more data, including more factors, or controlling for weather. I think tracking it over the course of the entire season is worthwhile to see any other trends or differences, especially with the increase in competitive games against common opponents.

Right now, the analysis also lacks tracking of special teams plays and any garbage time too. The overall point total and point differential would factor in any special teams or defensive scores. I’m not confident yet in how to track a special teams drive or the best way to include/discard any garbage time drives⁵ though.

I’m not sold that the new rule changes are adding more commercials and giving us less football either. I’m confident though that the rule changes took into account that television dictates the sport.

If we want game duration to go down, it’s likely easiest to adjust what you can control. For example, shortening the 20 minute half time or improving how long it takes to complete a review.

---

🤟 Thanks for reading this far. You can find code for this chart and more information about games this season at fbs-logs.blessyourchart.com. Please consider subscribing below 🤟

1. This is way too many steps. The switching between games no longer being a flip of a “channel”, but an open and login of a new app stinks out loud. Sorry for complaining about it.

2. Shoutout Outkast.

3. All data is human-made in my mind. A human trains a model or decides how to train it. Humans make mistakes.

4. Riveron did tell the truth about game length. The rule changes intent was player safety and not to make the games shorter.

5. Big fan of Brian Fremeau’s data project. I might explore using some of those definitions for garbage time tracking.