clock menu more-arrow no yes

Filed under:

Arkansas Bye Week Advanced Stats: Why Is This Happening?

New, 10 comments

Defensive problems, quick preview of the next four games, and a new statistic are topics of this bye week special.

NCAA Football: Mississippi at Arkansas Nelson Chenault-USA TODAY Sports

“The bye week comes at the perfect time” is one of the biggest clichés in college football. Is it ever a bad time for a bye week? (We’ll test that theory next year when the Hogs have one after just two weeks.)

But it’s more than just a cliché this year. The Hogs get a reset after a 53-point loss to Auburn.

Last Saturday was, without a doubt, the worst Saturday for Hog football since (at least) the Toledo loss in September 2015. It’s not just because Auburn lost: TCU and Ole Miss lost too and are now a combined 7-7. So those wins aren’t looking as good as we thought.

What happened to the defense?

It’s been a roller-coaster under Robb Smith. After Willy Robinson’s final defense declined in 2011 from a solid 2010 (something there was no reason for, given that Arkansas returned almost everybody that year), he was fired. In the throwaway season of 2012, Paul Haynes improved the Hogs against the run, but the bottom fell out against the pass. Chris Ash’s only season (2013, Bielema’s first) was plagued by bad communications and blown assignments, and was particularly bad against the run.

Smith entered the picture in 2014, inherited a veteran defense, and instantly dominated. Arkansas improved by every metric in 2014. The bottom has fallen out since, though. Arkansas looked good against the run in 2015 as well if you only look at raw totals (10th in actual rushing yards allowed per game), but as advanced stats fans, we know better than that. Arkansas actually fell to 51st in Rush Defense S&P+ and simply benefited from teams throwing much more than running to allow so few rush yards.

If you look at the changes from 2015 to 2016, it’s obvious what Smith was trying to do: he figured the run defense was fine with linebackers Brooks Ellis and Dre Greenlaw returning, so he decided to make some changes to improve the pass rush, which would help boost a pass defense that finished 115th (in S&P+).

In addition to unknown decisions regarding the playbook and how the team practices, Smith made two obvious moves, and I would argue that both have backfired:

  • Moved Jeremiah Ledbetter (and Sosa Agim) inside. With so many defensive ends on the roster, Smith decided to turn the bigger Ledbetter into a disruptive 3-tech defensive tackle, like what Darius Philon was in 2014. On paper, it made sense as it would doubtlessly improve the pass rush. Why it failed: Ledbetter has been solid in that spot, but he’s not as big as Philon was and thus doesn’t command a double-team, so he can’t be as good against the run. Further, his move has reduced the number of solid defense ends, creating a new void at his old position.
  • Made Deatrich Wise an every-down end. Despite Wise’s prowess as a pass rusher, he was not a starter in 2015 when he lead the team in sacks (and was often the only one getting consistent pressure). He played most of his downs in obvious passing situations, while Ledbetter or JaMichael Winston handled early downs. With Ledbetter’s move inside and Smith’s desire to get more of a rush from his front four, Wise became an every-down player. While he hasn’t been bad at rushing the passer - given the hype, he’s probably disappointed a little bit - it’s easy to see why he didn’t play early downs until this season. His inability to set the edge makes him a major liability in the run game.

The trade-off (up to 57th in pass defense but down to 124th in run defense) was clearly not worth it, and now the general fan dislike of Robb Smith is growing after a historically-bad showing against Auburn. If Bret Bielema is considering replacing Smith, he’ll have to ask himself if this season was due to bad luck from a well-thought-out decision, so shaking up the staff isn’t worth it... or if the decision showed bad judgement and warrants replacing the coach who made it.

Looking ahead

Arkansas S&P+ Ranking: 62nd. Yikes, a 45-spot fall from preseason. Not good.

  • Florida: 9th
  • LSU: 8th
  • Mississippi State: 64th
  • Missouri: 40th

S&P+ says the Hogs will get blasted by Florida and LSU then split the final two to reach a bowl at 6-6. Arkansas’ chances of losing out and missing a bowl are 26% and the chances of reaching 7-5 or better are 29%.

I’m not sure about any of these games, and I think we could still be in for some madness this season. Florida doesn’t feel bad for a variety of reasons; among them, the Hogs are coming off a bye week while the Gators are coming off a rivalry game against Georgia, and Florida is not a good running team, ranking just 91st in Rushing S&P+. In case you haven’t noticed, run defense is kind of a problem for Arkansas.

LSU could also go several ways. Obviously, the most likely result is Leonard Fournette rushes for as many yards as he wants to and LSU wins comfortably. But remember, LSU will play Arkansas the week after Alabama. If the Tigers are drained physically and emotionally, that could go Arkansas’ way as it has the previous two seasons.

Mississippi State and Missouri continue to look like possible victories. Mississippi State is surprisingly good as running the football (10th Rushing S&P+) – almost all of that from their quarterback, Nick Fitzgerald. However, that’s their only strength. They are a very poor passing team (83rd), have a bad pass defense (106th), and don’t get much pressure on the quarterback (101st in adjusted sack rate).

Missouri is a wildcard. The Tigers are 2-5, but have a sneaky-good offense lead by improving quarterback Drew Lock. However, any offensive improvement has been wiped out by a total defensive collapse. Last week’s 51-45 loss to Middle Tennessee State (!) saw the Tigers surrender 629 yards of offense and marked three straight games of allowing 40+ points. Missouri is average at running the ball (51st), decent at passing it (25th), really bad at run defense (97th), and pretty bad at pressuring the quarterback (81st). Those numbers are tailor-made for an Arkansas win.

Why 7-5 could be good enough to call this season a success

If you can’t win 10 games, you have to find the little things. One of those little things is playing the long game; that is, passing other programs in the SEC West pecking order.

Here’s that pecking order entering the season, based on my best guess:

  1. Alabama
  2. LSU
  3. Texas A&M
  4. Ole Miss
  5. Auburn
  6. Arkansas
  7. Mississippi State

Arkansas can solidify its spot ahead of Mississippi State if MSU goes 4-8 (their most likely result, and 3-9 is definitely on the table) and the Hogs beat them. But what about moving up? That’s where Arkansas’ 34-30 win over Ole Miss comes into play. The Hogs have beaten Ole Miss three straight years and now, for the first time during those three, appear poised to finish with a better overall record. Ole Miss is 3-4 and still has games against Auburn and Texas A&M on the schedule. Hugh Freeze seems to be (finally) treading water in Oxford. He still can’t establish the run, and his defense has been subpar for two years now. Once Chad Kelly leaves, does Ole Miss remain a competitor?

If Ole Miss goes 6-6 and Arkansas goes 7-5, the Hogs may have done enough to move up to #5. That definitely could be an overreaction, given that Ole Miss was a 4th-and-25 away from winning the West last year, but it sure feels like the Rebels have stagnated as a program. Passing Ole Miss doesn’t seem like much, but in the very long term, it could matter.

Meet a new statistic

I promised a new statistic in the Auburn recap, so here it is. It’s called expected value (EV). EV assigns an “expected point value” for every down and distance at every spot on the field. Each play in the game changes the expected value, and the changes are documented in a chart.

Here’s how it works. Let’s take the following sequence from the Ole Miss game:

  • Arkansas has 1st-and-10 at its own 25. Based on my data, the average drive from that spot ends with 2.34 points. That’s the starting value.
  • On first down, Rawleigh Williams runs for 24 yards out to the 49. That gives Arkansas a 1st-and-10 from its own 49, which is worth 3.29 points. So that run by Williams was worth 0.95 points (3.29 – 2.34).
  • On the next play, Williams gets only two yards. My formula really doesn’t like short gains on first down, and 2nd-and-8 from the opponent 49 is worth just 3.05 points, so Williams loses 0.24 for that run (3.05 – 3.29).
  • His two-run average is now about 0.35 (0.71 points over two carries).

The process of getting those values was the hardest part, and now that it’s done, tweaking them is much easier. Here’s how I did it, for those wondering:

  • I downloaded an Excel spreadsheet of every (FBS) play of the 2015 college football season and the first six weeks of the 2016 season. That’s about 200,000 plays, but unbelievably, it’s actually not enough (I’ll explain below).
  • I ordered every play by drive, and ran a formula to assign how many the points the drive ended with to every play (for example, every play of a six-play touchdown got a “7” while every play of a five-play punt drive got a “0”).
  • I pulled every play’s down, to-go, yardline, and drive-end data to a separate spreadsheet, where I “averaged” the number of points ultimately scored from every down and distance at every spot on the field.
  • I grouped all the data by down and ran a regression function on each set, generating four equations, one for each down.
  • The function checks the play-by-play to determine the down, and then pulls the yards to go and the yardline and plugs it into that down’s formula to generate the starting EV for the play. Changes in EV are documented from play to play.

Here’s an example of how we can chart EV and walk through the different values. Here’s the Ole Miss game:

Let’s start with Field Position EV. This is simply the cumulative expected value based on starting field position for every drive. As I mentioned above, having 1st-and-10 at your own 25 is worth 2.34 points, and that is counted as Field Position EV. It’s worth noting that you cannot have a negative FPEV, since it simply adds up starting field position, so this value will always be significantly larger than Offense EV or Special Teams EV (which are often hurt by negative values). It’s also worth noting that this figure has some overlap with Special Teams EV (since a good punt by one team creates poor field position for another), so the team with higher Special Teams EV often wins FPEV as well.

Offense EV is, of course, the most interesting one. In a defensive struggle, Offense EV could be negative, since failing to score points squanders almost all the EV you started with from field position. For example, to open the Ole Miss game, Arkansas had 2.34 EV from its starting field position but went three-and-out. The Offense thus had -2.34 EV at that point, since it left those points on the table. Failure to convert on third down leading to a punt causes a loss of all almost EV on the drive (for example, Austin Allen being sacked on 3rd-and-10 on the first series was valued at -1.26, which was all the EV Arkansas had at that point).

What’s amazing is that my calculation of EV is almost identical to success rate. For a 1st-and-10 situation, success rate formula says you need 4 yards (40%) for a play to be called “successful.” In my EV formula, a 4-yard gain on first down is recorded at roughly 0 EV: a break-even point. Your chances of scoring don’t go up, but they don’t go down either.

For those wondering, here were Arkansas’ biggest offensive plays from scrimmage, according to EV:

  • Austin Allen’s 51-yard touchdown pass to Dominique Reed (3.74 points)
  • Austin Allen’s 13-yard touchdown pass to Drew Morgan (2.73 points)
  • Rawleigh Williams’ 53-yard run (2.50 points)

All came in the first quarter. And here were the worst plays:

  • The fumbled snap that was kicked backwards for an 18-yard loss in the third quarter (-3.50 points)
  • Austin Allen’s interception in the third quarter (-2.50 points)
  • Rawleigh Williams’ 2-yard run on 3rd-and-3 deep in Ole Miss territory, leading to a field goal (-1.99 points)

A touchdown is worth 6.97 points (the reason why it’s not 7.0 is below), so scoring a touchdown is worth the difference between the starting EV of the play and 6.97. This is obviously high-value; for example, Allen’s two-yard pass/toss to Williams in the third quarter was worth 2.16 points.

Penalty EV can play a role, too. A second-quarter pass interference call on Henre’ Toliver was worth 1.28 points to Ole Miss.

Special Teams EV is a little more complicated. It has three components: Punting, Kickoffs, and Placekicking. Let’s start with punting. A “neutral” punt (worth 0 EV) has a net of 40 yards. If a punt nets more or less than that, then the difference in field position counts to the punter’s total. For example, after Arkansas’ opening three-and-out, Toby Baker boomed a 53-yard punt, forcing Ole Miss to start at its own 35 (or, 2.74 Field Position EV). If Baker’s punt would have been “neutral”, that is, if it would have netted just 40 yards, Ole Miss would start at its own 48 and have 3.25 EV. So Baker is credited with the difference, or 0.51 EV. If a team is punting from opponent territory (so a 40-yard punt would travel inside the 10 or be a touchback), it only has to get to the 10-yard line. Any better without a touchback and the punter gets EV.

As you can see, Baker clearly out-punted Ole Miss’ duo of Will Gleeson and Gary Wunderlich. In one significant sequence in the third quarter, Baker dropped a punt at the Ole Miss 5 (for 0.20 EV). The Rebels went three-and-out, and Gleeson’s punt went just 36 yards, and Jared Cornelius tacked on four more on a return – the value of the punt was -0.32, or a 0.52-point swing in Arkansas’ favor based on the two kicks. The Razorbacks only had to drive 38 yards for the go-ahead touchdown. Right now, Cornelius’ return can only be counted against Ole Miss’ Punting EV; I’m considering adding Return EV as a new category.

Kickoff EV is similar. A “neutral” kickoff is one that nets 40 yards, forcing the opponent to start at his own 25. Any difference and the field position is counted for or against the kickoff unit. This is not a strength for Arkansas.

Placekicking EV is also based on an expected value. I assume that field goals <29 yards are good 90% of the time, field goals 30-39 yards are good 75% of the time, field goals 40-49 yards are good 60% of the time, and field goals from 50+ are good just 45% of the time. So Adam McFain’s 43-yard field goal in the second quarter had a “neutral” value of 1.8 (60% x the three points its worth). He made it, so he gets the other 1.2 points of its value. Extra points are similar. They’re good 97% of the time, so every made extra point is worth 0.03 points to the kicker if he makes it and -0.97 if he misses it. That’s why Ole Miss is in the negative.

What can we learn from EV?

EV is better than success rate in showing you where you are strategically weak. It also highlights special teams much more efficiently (my game previews and recaps over the last couple years have been very light on special teams). As I mentioned above, EV nearly matches success rate in what the standard is for a play to be successful. But EV is more versatile because it tells more than just “yes/no” on whether the play is successful: it tells you how successful, and how value each play really is.

We can see from the Ole Miss game that the cracks in the run defense that Auburn exposed were already there. It didn’t look like Arkansas performed that bad against the run, but the Rebels’ 6.20 Rushing EV suggests otherwise. Keep in mind, I deemed a couple of Chad Kelly’s runs – including a 20+ yard gainer – as pure scrambles, not called runs or run-pass options, so they counted towards Passing EV. We can also break EV down on a per-play basis, which opens up a whole new range of analysis, including comparing two running backs’ EV per play.

I got the idea for most of this data from two sources: Football Outsiders, which uses a number of similar metrics, and MGoBlog, an excellent Michigan fan site that has used EV (and a related stat called PAN, that I may try to adapt into my previews/recaps once I get enough data) for several years now. The stats guy over there has more than 1 million plays in his database, so his figures are probably a bit more accurate than mine.

I love feedback. If you have any suggestions on how I can present EV in a way that helps analyze the game, please let me know.

In the meantime, I’ll leave you with one other one. Leave a comment with any analysis you have for it: