2014 Big Ten Football Season in Review - Game Script Analysis

2014 Big Ten Football Season in Review - Game Script Analysis

Submitted by stbowie on August 18th, 2015 at 4:04 PM

Advanced Football Analytics puts together an interesting game summary, titled Game Scripts, for the NFL. Early last season I decided that I would put together these stats for the Big Ten, and here they are! (If this seems like an odd time to put together a summary of anything that happened in an actual football game last season, you’re not wrong. But it’s basically the last opportunity before there’s actual new football so… enjoy.)

What is a Game Script?
A Game Script (GS) is an average of the point differential for every second of a game. For example, if a team scores a field goal at the very end of the 1st quarter and that is the only score in the game (God save us from Big Ten football), the point differential was 0 for 900 seconds and +3 for the remaining 2700 seconds. The final point differential (Margin of Victory, or MoV) is +3 and the Game Script is +2.25.

Basically, gaining a lead early and maintaining it gives you a higher Game Script than keeping a game close and blowing it wide open at the end of the game. 

Why is it useful?
Useful is perhaps too high a bar to clear, as Game Scripts have little predictive value for future games, but it's certainly interesting and descriptive. A comparison of the final point differentials and Game Scripts can, at a glance, show which teams pulled off big comeback victories (a positive margin of victory, but a negative game script, indicating the other team held the lead for much of the game), which teams dominated the whole game (high game script and high margin of victory), and which teams played tight games but pulled away at the end (a low game script, but a high margin of victory). If you’ve watched a particular game from beginning to end, it doesn’t help much, but it does provide a good summary for games you weren’t able to watch in full.

I should be more clear when I say Game Scripts have little predictive value. Game Scripts correlate very strongly (~.94) with the final Margin of Victory for the same game - this should not be surprising, as both rely directly on points scored, i.e. you establish a lead, you maintain a lead, you tend to finish the game with a similar lead.


Average Game Scripts for the season correlate weakly with your results for the next game (~.28), and only slightly stronger than the correlation for average Margin of Victory for the season to your results for your next game (~.26). Sample size issues abound here, as we're looking at a single season and a stat that is only recorded once per game.

How’d Michigan do in this stat?
Rather dismally, I’m afraid, which is why instead of a mid-season project with weekly updates, this turned into an offseason summary project. As you are no doubt aware, last season Michigan lost regularly and emphatically. The good news is that Game Scripts offer plenty of schadenfreude to go around!

Game Scripts for all 2014 Big Ten Teams (In- and Out-of-Conference Games)

Team Table Chart
Illinois link link
Iowa link link
Northwestern link link
Wisconsin link link
Nebraska link link
Minnesota link link
Purdue link link
Indiana link link
Ohio State link link
Michigan State link link
Michigan link link
Rutgers link link
Maryland link link
Penn State link link

Which teams won in the most dominant fashion in the Big Ten this year (defined as biggest average Game Scripts in wins)?
MSU and OSU are the clear leaders in this category, with MSU averaging a 16.5 point lead in all of their wins (which includes a big late comeback against Baylor dragging their average down) and OSU averaging a 13.8 point lead in all of their wins. As we knew, they both had impressive seasons (certainly their last or close to last, right guys? Because Harbaugh?).

Some surprising teams are featured in the next echelon of this category, with Indiana, Minnesota, Michigan, Nebraska, Northwestern, Wisconsin, Purdue, Maryland, and Rutgers all averaging between one and two touchdowns in their victories. This is where strength of schedule (i.e. wins over cupcakes) and sample size (Indiana has a big game script in their wins… of which they had three) make a big difference. Michigan, for example, averaged a lead of 10.1 points in our wins, but that is anchored mainly by big wins over App State, Miami of Ohio, and Indiana. Similarly, Indiana won big over Indiana State and UNT, and squeaked by Mizzou with a 4-point victory. On the other hand, Wisconsin was just plain dominant in most of their victories this year, with the lone exceptions being their 2-point win over Iowa and their 3-point win over Auburn.

The last category has some of the teams you’d expect, with Rutgers, Illinois, Iowa, and Penn State all averaging a less-than-one-touchdown lead in their wins. Illinois deserves its own what’s-the-room-below-the-basement category, as its average lead in its five wins was a rather terrible 2.7 points.

This year’s biggest comeback wins (defined as largest negative Game Script in a victory)
MSU had the biggest comeback of the year against Baylor, with an average point differential of -7.8 for the game, and a 1-point final margin of victory. The next five closest were

  1. Rutgers over Maryland (GS: -6.7, MoV: 3)
  2. Minnesota over Nebraska (GS: -5.2, MoV: 4)
  3. The thoroughly embarrassing Iowa over Ball St (GS: -5, MoV: 4)
  4. PSU over Rutgers (GS: -4.6, MoV: 4)
  5. The absolutely glorious Northwestern over Notre Dame (G: -4.3, MoV: 3).

Michigan’s lack of any significant games in this category fits the narrative of the 2014 season – a lack of hope that this team could win a game once it fell behind. The lone comeback for Michigan on the season was against PSU, with a Game Script of -0.6 and a final margin of 5.

This year’s biggest comeback losses (defined as largest positive Game Script in a loss)
Wisconsin choking away its lead over LSU takes the cake in this category, with a Game Script of 5.9 and a final margin of -4. Not surprisingly, this category also features a few teams that were on the opposite end of the games mentioned in the section above – Nebraska, Rutgers, and Maryland were on the less glamorous end of those big comebacks. A miserable loss we wouldn’t want to overlook, however, is Iowa’s three point loss to ISU, with a Game Script of 4.1. Michigan’s only comeback loss was to Maryland, with a final margin of -7 and a Game Script of 1.2.

The depressing loss you knew you should have turned off in the 1st half but probably didn’t (defined as largest negative Game Script in a loss)
While Michigan’s embarrassing loss at Notre Dame was certainly its worst loss in this category, the good news is that that game doesn’t even make the top ten most depressing loss by a Big Ten team! Wisconsin leads off this category with its 59-0 loss to OSU in the BTCG, featuring a Game Script of -31.8. Other OSU wins (over Illinois and Rutgers) round out the top three. Wisconsin is the only team to both deliver and sustain a top-ten loss in this category (Maryland’s 45 point loss to Wiscy, with a game script of -25.3, comes in fourth on this list).

The dominant victory that made you feel all warm and fuzzy about the future of your team (defined as largest positive Game Script in a win)
Cupcakes fill out the majority of this category, with OSU’s win over Kent St (GS: 40.2, MoV: 66), MSU’s win over EMU (GS: 38.4, MoV: 59), and Wisconsin’s win over BGSU (GS: 29, MoV: 51) all placing in the top five. Michigan’s win over App State qualifies with a Game Script of 24.9, but falls outside the top ten.

Which games were closer than they seemed? (Game Script between -3 and 3, with a two-score win or loss)
Only seven games fell into this category:

  1. Illinois vs Purdue (GS: 2.5, MoV: 11)
  2. Northwestern vs Nebraska (GS: 2.1, MoV: 21)
  3. OSU @ Navy (GS: 2, MoV: 17)
  4. Illinois vs YSU (GS: 1.6, MoV: 11)
  5. Minnesota vs Mizzou (GS: -0.8, MoV: -16)
  6. Wisconsin vs Minnesota (GS: -2.7, MoV: 10)
  7. Purdue vs Iowa (GS: 0.6, MoV: 14)

I think OSU @ Navy and Minn vs Mizzou are the most interesting games from this category – the first because it shows OSU as a bit more fallible than they looked in the final score, and the second because the eventual SEC East Champion almost lost to two Big Ten teams that weren’t particularly close to the top of their divisions.

What does it all mean?
In perhaps the most obvious conclusion ever made in the history of sports analysis, scoring more points early (and throughout the game) is better and leads to more winning. Allowing your oppononent to score more points than you do in the later parts of games occasionally leads to unexpected losses. Quote me as frequently as you wish, and send all royalties to me by check via USPS.

Beyond that, I think you can argue that MSU had the more dominant season than OSU, given how close the GS was for their game, and the number of OSU games that featured low GS but high MoV (Navy, MSU, Indiana, Michigan, Alabama). Whether that's good news or bad news is up to you.

Happy to hear your comments and feedback on this analysis! In particular, I’d love to hear:

  1. How awesome you think this is;
  2. Suggestions on how to incorporate pass/run ratios into some of the above analysis;
  3. Suggestions on visual representations of the analysis;
  4. New categories to include in future versions;
  5. How to figure out those lightboxes for my image links...

I’m planning on continuing this in the 2015 season, and perhaps even providing weekly updates. If you’re interested, you can view the full Game Scripts data for the 2014 Big Ten season here. And of course, I urge you to check out the Advanced Football Analytics site (recently acquired by ESPN), which does all sorts of awesome analysis of NFL games.


Attempting to Evaluate the OL and Running Game Using Statistics

Attempting to Evaluate the OL and Running Game Using Statistics

Submitted by MichAero on October 7th, 2014 at 9:18 PM

Reading all of the debate about the offensive line and the running game, I decided to do some research about the matter. I looked at the 2013 and 2014 YPC and adjusted them against the opponents YPC allowed. I also looked at sacks allowed, and compared them to the opponents average.

Note: I used YPC as a way to control for tempo, and it helps to find a common link between each game. For reference, I added the opponents rushing YPC rank along the y-axis. They are chronological-- CMU is the top and KSU is the bottom of 2013, and App State is the top for 2014.

To start with, I looked at Michigan's total YPC against each team. I then took this number and subtracted each opponents YPC allowed. I outputted this information into a graph, below. Values above 0 are good, values below 0 are bad. 

The previous graph shows that out of our 13 games, we rushed better than the opponents average 6 times, and worse 7 times. However, only 2 times did we rush over 1 YPC more than the opponents average. On the flipside, 5 teams held us to 1 YPC less than their average or worse, with 2 teams obliterating us. It appears, IMO, that UCONN found our weakness and other teams after were able to capitalize.

Additionally, Minnesota (at 90th) and Indiana (at 117th) were poor run defense that shut us down. The final 2 games are a bit surprising. OSU can be chalked up to a rivalry game, or so I thought, but even with our backup QB we rushed decently against KSU (though only on 15 attempts).

The following graph shows the same data, but for this year. Some caveats apply: Only 6 games played thus far, with a large portion against poor teams, for instance.

From here, we can see that 2 teams have done better against us than their average, but not by nearly as much as 2013. Additionally, we have done a better job against the defenses we should, and even have an above average performance against what appears to be a good run defending team (Utah).

These numbers are subject to change throughout the season, but there appears to be a window for at least some hope.

Next, I looked at sacks allowed by our OL. Again, I subtracted the defenses average sacks from this number (adjusting it by taking out sacks against us). I did this here to get a view of how we stacked up against their other opponents.

Note: I also did these same graphs without adjusting (by taking out our sacks), and the charts are still roughly the same. The numbers skew a bit, but the trend is still there. Also, the numbers along the y-axis are the opponents rank for sacks per game.

The following graph is from 2013. Here, numbers below 0 are considered good, and numbers above 0 are considered bad.

Similar to the YPC chart, we started better and finished better, but struggled hard in the middle. We gave up an above average amount of sacks against teams ranked 100 and 103, and our best performance was against a team ranked 48. It is understandable to give up some sacks to Nebraska (20), but the amount is concerning. UCONN was the 100th team, by the way, again suggesing that they exposed a huge weakness.

The 2014 chart is next. This is subject to change much more, as the competition and small sample size make a more complete picture.

Thus far, the line appears to actually be doing a much better job of avoiding sacks, compared with how the opponents are playing against other teams. This is even against the 1st and 8th best teams as far as average sacks go. Utah, for instance, is averaging 5.6 sacks per game against everyone, and we "held" them to 4. Rutgers is averaging 4, and we "held" them to 3. Notre Dame is the lone exception this time, and I would contend that is more a product of having the lead that they did and didn't have to worry about us running nearly as much.

And lastly, I looked at a combination of the above. I took the sacks out of the rushing stats, and recalculated both our YPC and the opponents YPCA. The 2013 graph is shown below.

This actually looks worse to me. Now, we only have 4 performances above the average, and one just barely. 

The 2014 one is next:

Here instead, we are now below average only once. Our rutgers performance is a bit weaker now, as is Utah, but the other performances are better than in the previous graph.

You are free to draw your own conclusions from these. There is obviously a lot more football to be played, but the early numbers are looking decent. We are running better against better defenses, and actually performing better than average against a couple aggressive defenses. I think the sacks above average might start getting closer to 0 as we move into conference play, but that will be something to keep an eye on.

If you have any suggestions, comments, criticism, etc., please feel free to share. If there is interest, I will try and update this post as the season continues (assuming I have the time to do so).

UPDATE: I have added in a similar analysis using sack percentage. Thank you for the suggestion. I have also done an analysis on YPC, and sack % after the first 6 games from last year as a comparison.

The first graph is for the 2013 sack percentage above average. Negative numbers are good while positive numbers are bad.

As you can see, we still have 6 good performances and 7 poor performances. Unfortunately, all games against an opponent worse than 100 we did poor against. And again, it looks like we had some flaws exposed, but this time it suggests we might actually have done something at the end to fix them. Whether that is scheme, or players just producing and developing, I cannot say.

The numbers so far for 2014 are shown now.

Here, we see that our Rutgers performance was worse than the first analysis shows, and the Minnesota numbers become average. I'm not worried about the average Minnesota numbers because it was just one sack. The Rutgers number scares me a bit more, but if you look at the context I'm not sure it should. We were playing a night game on the road, like against ND. This time, though, we allowed just one sack in the second half, and that was on our opening drive of the 2nd half. Yes, we don't want to give up 3 sacks on those few passing attempts, but just throughout the game we saw some improvement IMO.

Next, I looked at the sack percentage from 2013, but looked at just how our first 6 opponents faired in their first 6 games.

We can see from this that the trends stay mostly in line, surprisingly. The CMU game and the Akron games look better here than they end up, and the UCONN game looks worse. The other games stay about where they are.

Finally, I did the same YPC analysis above, where I took out sacks, and looked at the first 6 games.

What we see is that the first 2 games look better here (CMU and ND), as do the last 2 (Minnesota and PSU). The middle two stayed roughly the same. The game against ND shifted by about 1.25 YPC.  I think that this shows that this isn't quite as good as it looked initially, but I don't want to make any sweeping conclusions here.

I wanted to add that I used data from cfbstats.com, and I got the rankings from teamrankings.com.

Calling Football Stats People - Download Box Score Data

Calling Football Stats People - Download Box Score Data

Submitted by stbowie on September 17th, 2014 at 9:52 PM

Does anyone know a good free source for downloading box score data? Primarily interested in this year's games, but somewhat interested in previous years as well. I'm looking to put together some statistical summaries of B1G games and would love to not have to enter data by hand if it's already out there (and free).

Thanks and Go Blue!

OT: Will this device for measuring head impacts revolutionize football?

OT: Will this device for measuring head impacts revolutionize football?

Submitted by Don on July 13th, 2013 at 1:12 AM

Reebok has developed Checklight, an impact sensor for the head which can be worn under a helmet and records the number of hits at different intensities. It has a flashing indicator that will indicate a level of impact sufficient to warrant removal from competition for assessment.

While football is the sport getting the most attention regarding concussions and other head injuries, hockey is another logical sport for this.