Differentiating a differential

Matt Melton has been exploring a new metric.

So what is first half point differential (1HPD)? Unlike advanced stats, it is easy to define and calculate. It is simply the scoring margin (positive or negative) in the first half of a game. In the 2019 College Football National Championship Game, LSU led Clemson 28-17 at the half. The 1HPD for LSU was +11 and -11 for Clemson. Simple stuff. With more than ample time on my hands, I calculated the 1HPD for all FBS teams in conference play back to the first season of the College Football Playoff (2014). That is six seasons worth of data. What did that data show? I thought you would never ask.

For starters, 1HPD in conference play is positively correlated to a team’s conference record. In fact, with an R squared value of .72, the correlation is stronger than YPP in conference play (which surprised me). So if 1HPD has a strong correlation with conference success, what happened to those teams that saw their expected record based on 1HPD differ significantly from their actual record? Wow, its like your reading my mind.

What happened is pretty interesting.

For YPP, I consider a difference of .200 between actual and expected record significant, so I used the same logic here. Between 2014 and 2018 101 FBS teams saw their actual record differ significantly from their expected record. 52 teams under-performed relative to their expected record and 49 exceeded their expected record. Those teams that under-performed tended to see their conference record improve the next season.

The average team improved by 1.41 wins in conference play and nearly 70% of the teams improved by at least one win.

On the other hand, the teams that exceeded their expected record more often than not declined the following season.

The average team won about 1.80 fewer games the next season and more than three quarters of the teams declined by at least one win.

His latest post runs that through the P5 conferences.  Here’s what he comes up with for the SEC:

Finally, here are the SEC standings.

And the SEC 1HPD.

I know its fun to hate on Alabama, but don’t go throwing dirt on the Tide just yet. Alabama was nipping at the heels of LSU in 1HPD and if we look at the other seven conference games both teams played, Alabama actually had a better differential (+137 to +113). Of course, that classic in Tuscaloosa does count and lets also give LSU credit for never trailing at the half in any conference game (they were tied against Florida and Auburn). In the East, Georgia finished with a healthy 1HPD margin over Florida and Kentucky and despite the relatively disappointing season, the Bulldogs led by double-digits six times in SEC play (tied with Alabama and LSU for the most double-digit leads).

(By the way, is anyone surprised that Georgia was third in the conference, behind the two offensive powerhouses?)

The question is whether you see a team there that’s destined for a little regression to the mean.  Matt sure does.

… Since all the Power Five conferences are grouped together here, I am only going to list those that significantly over or under-performed (a difference of at least .200). We’ll start with the overachievers.

You’ll enjoy his explanation.

… Tennessee has been one of the hardest teams for me to get a read on this offseason. Were the Volunteers good in 2019? They closed the year on a six-game winning streak, beating three bowl teams in the process (Indiana, Kentucky, and UAB). Their YPP numbers were solid (fifth overall and third in the East). On the other hand, their APR numbers were much less glowing (ninth overall and fourth in the East). They also lost to Georgia State and were pounded by the three best teams on their schedule (Alabama, Florida, and Georgia beat them by a combined 82 points). And since when are Tennessee fans delighted by victories against Indiana, Kentucky, and UAB? Their 1HPD also shows them to be pretty weak. Despite winning five of their eight league games, they were outscored in the first half in conference play and actually trailed at the half five times, including four times by double-digits. The Vols do have the benefit of drawing Arkansas in their rotating cross-division game in 2020, but Alabama, Florida, and Georgia are still on the schedule as well as a non-conference trip to Oklahoma. 8-4 seems like the ceiling for this team with a real possibility of 6-6 or worse.

Anybody who wants to answer his “since when are Tennessee fans delighted by victories against Indiana, Kentucky, and UAB?” question can step right up in the comments to do so.

Aside from that, what do you think of his correlation?

11 Comments

Filed under Stats Geek!

11 responses to “Differentiating a differential

  1. Corch Irvin Meyers, New USC Corch (2021)

    What I see there is how Floriduh isn’t anywhere near close to us, and that Kentucky, having to cobble together an option offense about a third of the way through the year, is nipping on the heels of media-made-genius Sideshow Dan the Clown.

    That gives my prediction that Kentucky wins 9-10 regular season games this year and finishes second in the East far more credence.

    Like

  2. Muttley

    Vince Dooley’s teams would have driven this guy crazy.

    Like

  3. junkyardawg41

    “since when are Tennessee fans delighted by victories against Indiana, Kentucky, and UAB?”
    2020 – I had it right after Australian Wildfires but before Murder Hornets.

    Liked by 3 people

  4. NotMyCrossToBear

    Since it’s not the 90’s and now they suck?

    Liked by 1 person

  5. David H.

    Intuitively, I think the reason first-half point differential is a good predictor of success is that both teams are guaranteed to be trying all-out in the first half. In the second half, if the game is a blowout, then one or both teams are playing out the string, and the second-half scoring may not be representative of the true team strength.

    The best advanced stats (like Bill Connelly’s) filter out garbage-time data, so they also reflect well how teams play when the game is still in doubt. I’d bet that a version of yards-per-play that’s adjusted to remove garbage-time plays would do as well or better than first-half point differential for predicting team strength. (First-half point differential does have the advantage of being far easier to calculate.)

    Like

  6. Snoop Dawgy Dawg

    I don’t intend this as a snarky response, but it may read that way. He did a lot of work to say that teams leading going into half time tend to be better teams with better records. And that, the bigger the lead going into the half, across a season, the better the season. Am I wrong to read this in the same context as an announcer stating before kickoff, “it’s really going to come down to who can score more points tonight”?

    The indices used to predict success the following season also seem to be self-evident. Outside of a handful of teams that sustain success, most teams who are good tend to drop a game the following year or who are bad, win a bad extra the following year.

    The biggest thing for me here is that in spite of our offensive woes, our defense really locked down most teams. Most of that defense returns, and we got basically everything we could hope for this year for a big offensive improvement. Except there is also the Rona floating around, which means this season probably doesn’t happen, and what could have been a special year is taken from us.

    Like

    • FlyingPeakDawg

      Elsewhere in statistical analysis, the team that scores more points has a R factor at .8945 deviated from a squared mean of .0934 multiplied by the KPI factor usually wins.

      Did the Snarky for you. You’re welcome. 😉

      Like

  7. DawgPhan

    since Derek Dooley?

    Like

  8. 69Dawg

    A simple answer is if you have a great defense, and you can jump on your opponent early, your mostly home free. Ala, LSU and us were not going to let many teams that were down get back in it.

    Like