070: College football games are still too long

Recap of what I learned trying to track game duration over the 2023 college football season.

Did the recent rule changes in college football games result in shorter game times?

This past college football season, I put together a project that tracks the duration of games to surface the data. I tracked 910 total games and 792 games between FBS teams, and you can check out the data here:

This is a longer post, so if you’re not interested, feel free to move on with your day. I plan to share more college basketball data in the near future, so please consider subscribing to the newsletter if you haven’t done so already.

Here is a recap of what I learned trying to track this data.

First, what were the rule changes?

There were a series of rule changes put in place prior to the 2023 season.

The primary change was the clock no longer stops when the offense gains a first down except with less than two minutes remaining in the second and fourth quarters.

This is the change that brought curiosity after week zero games. The number of offensive plays went down, however, the duration of games only decreased by about four minutes after week zero¹.

So, did the rule changes result in shorter games?

No. The rule changes didn’t result in significantly shorter games.

Specifically, the average game duration this past season was 203 minutes. This is four minutes shorter than last season (207 minutes).

A two percent decrease in game length doesn’t feel all that impactful.

But, was the intent of the rule changes to reduce the duration of games?

No, it was not.

The best explanation of the rule changes that I’ve heard came from Alberto Riveron, the ACC Supervisor of Officials, in an interview with Ovies & Giglio last summer.

The rule changes were designed to reduce the number of plays and improve player safety. Not expedite the game.

So, did the number of plays go down?

Yes, the number of plays and drives went down overall.

This is where we get into the weeds on how to track this information.

Data definitions for FBS logs:

• Plays: combined rushing and passing attempts on offense. Does not include other play types (kickoff, punts, field goals, etc. )

• Duration: listed in minutes. If a game includes a weather delay, it’s at the discretion of the scorekeeper if the delay is included in the duration.

• Drives: combined possessions that includes all raw data and any “garbage” time drives for games between two FBS teams.

These definitions result in the number of plays going down from 68.7 in 2022 to 66.8 in 2023. The average number of combined drives decreased from 24.5 in 2022 to 24.1 in 2022.

Better data definitions

Brian Fremeau provides an excellent college football data resource at bcftoys.com. Brian was kind enough to reply to an email of mine this fall, and shared more precise definitions for tracking drives or possessions.

These definitions cover blind spots in my analysis². For example, defensive possessions and special team possessions.

This analysis is more precise and surfaces that the average number of drives was 24.6 per game, which is the fewest number of drives since 2007. It’s a slight decrease from the 25.1 average in 2022.

Overall, this spells out a two percent decrease in the average number of drives and a two percent decrease in game duration.

How is duration or game length tracked?

In my analysis, I purchased data from SportSource Analytics. This provided lots of game stats and reliable drive data³. SportSource is behind the useful resource - cfbstats.com - and it aggregates a lot of this data too.

The game length or duration is provided by the scorekeeper⁴. This means it’s at the discretion of that scorekeeper and it can be subjective.

There are outliers in the data. This includes significant weather delays and blatant data entry errors too. For example, the week two game - Miami (OH) 41, UMass 28 - clocked in at 410 minutes due to weather and the week three game - Miami (OH) 31, Cincinnati 24 - was entered at 95 minutes likely due to a data entry error.

I don’t think either of those two game times is all that accurate, but it’s what was provided.

Should we control for weather when tracking game duration?

There are more weather delays in earlier weeks of the season. The second week of the season, around the second week of September, produced a lot of outliers in game duration.

At first, my goal was to adjust from any weather delays. This became a lot of work and I was also nervous that any historical data didn’t follow this same logic, so I chose to leave it as is.

The median duration still hovers around 200 minutes or 3 hours and 20 minutes most of the season.

Does game duration even matter?

Timothy Burke started to track broadcast length and surface it via the social platform BlueSky. I asked about game duration, and got this reply⁵:

I disagree that game duration has no value at all. I think there is some value or at least a degree of value. This did provide some confidence that discarding some of those outliers was safe.

It also hammered home the point that college football is about television and broadcast length probably matters more.

So, do we know why games are so long?

To be clear, any sort of statistical modeling is out of my circle of competence. This was a learning exercise to try and predict game duration based on a limited number of stats from those games. It’s also a follow up from this previous post.

The simple linear regression model uses several variables from game stats to try and determine the predicted game duration in minutes.

When excluding outliers of games over 300 minutes and fewer than 150 minutes, its predictions explain about 30 percent of the variation in game duration and its typically off by about 18 minutes.

Games with more combined penalties, drives, incompletions, and combined points tend to last longer. Games with more punts, more turnovers, and a large point differential often are shorter.

Here is a concrete example with a game where it was accurate:

• Michigan beat Nebraska 45-7 in week five

• The actual game duration is 179 minutes and the predicted game duration is 178.8 minutes

• Start with 135.53201 minutes - this is the estimated game duration when all other variables are zero

• Plug in the following stats and multiple by each factor . . .

• More of these factors, result in longer game durations:

  • 4 combined penalties (4 * 1.16711)

  • 19 combined drives (19 * .09647)

  • 18 combined incompletions (18 * 0.72039)

  • 52 combined points (52 * 0.28030)

  • 72 combined rush attempts (72 * 0.08389)

• More of these factors result in shorter game durations:

  • 4 combined punts (4 * -0.01291)

  • 1 total turnover (1 * -0.02131)

  • 120 combined plays (120 * -0.03983)

  • 38 point differential (38 * -0.27911)

  • 0 special teams touchdowns (0 * -0.78208)

What is the model missing?

The biggest miss in the model is any tracking of video reviews. This information is not tracked in any of the play-by-play data that I can find.

The strongest factor in longer games is the volume of penalties. For every penalty, you can add over a full minute of game duration. You can infer that more penalties likely mean some video or official reviews, so that might capture some of that data.

More granularity would improve the model. For example, point differential by quarter, type of penalty committed, or game duration by broadcast provider (Fox, NBC, ESPN).

Overall, the two most obvious way to reduce game duration in my mind are:

• Shorten halftime⁶: it’s 20 minutes now, can you reduce it by five minutes or two minutes without major impacts?

• Expedite reviews: the review process is not uniform across different leagues, and it can take a long time to get calls finalized. In games this season, we’ve seen 20 second plays take eight minutes of real time to conclude a call. That’s not good.

There is a lot more important problems with the sport than the game duration. The calendar is bonkers, NIL, coaching salaries, and more.

College football games are too long though. And it might be because that is how television wants it to be.

Technical bits about the project

You can find the code for the project here.

This was a challenging project. Once the season is in the full swing, there are games Tuesday through Saturday, and it was tricky to automate updates.

I expect there is a lot of mistakes in my code or analysis. This was a way for me to try and learn out in the open, so if you have suggestions or questions, please let me know.

🤟 And thanks again for reading this far 🤟

1. This tweet of mine somehow found its way to the Washington Post, thanks to Steven Godfrey. I don’t consider myself a college football statistician, but this was a good signal people were interested in these rule changes.

2. Here are the nitty gritty definitions of drives:
Offensive Drives (96.8 percent of all possessions) consist of at least one scrimmage play run by an offense in possession against the opposing team defense. Drives conclude in one of eight ways: touchdown, field goal attempt, punt, turnover on downs, fumble, interception, safety, or time expiring at the end of the half.

Field Goal Attempt Possessions (0.02 percent of all possessions) are situations in which, due to limited time remaining at the end of the half, no offensive scrimmage plays are run and only a field goal is attempted.

Defensive Possessions (1.1 percent of all possessions) are initiated as a result of an opponent interception or fumble and, with the defense in control of the ball, conclude on the same play in one of three ways: touchdown, fumble recovered by the offense, or safety.

Special Teams Possessions (1.6 percent of all possessions) are kickoff return, punt return, or failed field goal return events that conclude in one of four ways: touchdown, fumble recovered by the kicking team, onside kick recovered by the kickoff team, or safety.

Overtime Possessions (0.5 percent of all possessions) include overtime offensive drives (teams alternate drives starting at the opponent's 25-yard line), overtime field goal attempt possessions, overtime defensive possessions, and overtime conversion possessions in which teams alternate two-point conversion scoring attempts from the opponent's 3-yard line.

3. cfbfastR is also an excellent resource, however, I’ve found the play-by-play to be rough from ESPN (the main provider).

4. As a former Sports Information Director, the data input when keeping score is subject to human errors or mistakes. It can also be amusing when information isn’t provided too. I remember one SID entering in the name of his dogs as the umpire names in a college softball tournament in February.

5. The “no value at all” response was quite discouraging. I do understand where Burke is coming from though.

6. I know the 20 minutes is for the “pageantry” or bands to perform. I agree, those traditions are important. I’m not saying make it 12 minutes like the NFL, but 20 minutes is way too long.