offensive statistics

SOME INTERESTING FACTS ABOUT BIG TEN SCORING OFFENSES: 2000-PRESENT

I find the things I am uncovering to be interesting (hopefully, you do as well), I am extending this series another week, and this time, we are going to poke around at some of the historic stats on scoring offense in the Big Ten. I even included Nebraska this time.

Since 2000, the conference’s football teams have scored 49,281 points over a stretch of 692,579 yards of total offense, or roughly the distance from Detroit to Springfield, Illinois. The conference has also amassed 5657 PATs to go with 6,168 touchdowns, as well as 2,076 field goals and 4,776 total yards per year.  Actually, here’s a small chart with the four most common scoring types and their relative occurrence:

In the 1,792 games that all this encompasses, the Big Ten has managed to maintain an average rate of 27.5 points per game and 386.5 yards per game, which is not the West Virginia-Baylor game of recent note but is also not bad. It still means an average ranking nationally in the mid-50s, which essentially means there have been about as many terrible offenses in this stretch as there have been good ones, but slightly fewer bad ones. Yes, very technical statement there.

It doesn’t look like it on the field sometimes, but take thirteen years of data and do a table of “percent of total” for a few things and you get this:

Nebraska is, of course, the anomaly here. Illinois and Indiana show their protracted stretches of relative ineptitude even here, as the difference between Wisconsin and Illinois, for example, means a veritable sh*t ton on scoring over 13 years even if the percentage is small. For the most part, things are more even than I would have thought, but again, fractions of a percent here hide off seasons.

Here are the totals / averages by team from 2000 to the present:

It may or may not be the variation you would expect. I sorted the table by average points per game and was not entirely shocked by the order of the teams myself. All things considered, maintaining an average ranking of 42, in our case, which would be the upper reaches of the second quartile of teams, is not that bad at all when compared to the grand mean of 56.

So, similar to the other two diaries that I did recently, I asked myself the question – which of these nearly 150 teams in this spreadsheet were very good at scoring, in relative terms? Using a similar method, I decided to create from the excessively large table a small table of teams which were above average in at least four of the following: Total yards, TDs, FGs, PATs, and Points.

You get 63 teams that compare as follows:

Here, from a historic average of 27.5 points per game, you jump to 31.8 points per game for the teams that fit the criteria for this table. I then did the same thing with the remaining teams, and you see the following from the remaining 23 teams:

These teams were scoring at an average rate of 35.2 points per game, or slightly more than 1 TD per game more than the Big Ten grand mean in this time period.

Not shockingly, being able to actually get the ball across the plane or through the uprights on a consistent basis makes a considerable difference. The Big Ten’s cumulative winning percentage since 2000 has been 0.564, but when I did the first elimination, that jumped to 0.686, and then on the second one, it leapt to 0.753. Essentially, it is the difference, in scoring terms, between 7 and 9 wins in a season based on historic numbers.

TL;DR CONCLUSION:

Once again, this was an exercise conducted under an admittedly arbitrary set of assumptions, but it is interesting to see  the improvements that mere points will bring in numerical terms and give an added dimension – hopefully – to what occurs on the field and how much it means to, well, score.

“SOME CHARACTERISTICS OF HIGHLY RATED PASSING OFFENSES IN THE BIG TEN: 2000-PRESENT”

In a companion diary to my last entry, I took a similar dive into the passing statistics of the Big Ten since 2000 to see what some of the characteristics of the highly rated passing offenses were. In an attempt to be a little more thoughtful as well, I also looked at the passing efficiency data, particularly since TD and INT percentages are part of the calculation of efficiency ratings. I like to believe that these two percentages really matter more than the average total yards in that they provide insight into what a team does with the yards they managed to accumulate. This is also the reason that the selection process for this exercise varies a little from the method employed when looking at rushing offenses.

Some high-level trivia:

• Since 2000, a Big Ten quarterback has thrown the ball to someone else 53,905 times, and on 31,128 of those occasions, someone caught it. That’s good for a 57.7% completion percentage and 381,792 total yards.
• When the ball was caught, teams averaged 12.27 yards per completion. When it was not, it was thrown an average of 7.08 yards.
• For all that passing, 2,640 touchdowns were produced, or an average of 1.5 passing TDs per game. There were also 1,646 interceptions thrown, or 0.93 INTs per game.
• The touchdown percentage of the Big Ten in that space was 4.06%. The interception percentage was 3.09%. Michigan fell slightly above the average in both cases, incidentally.
• The cumulative passer efficiency rating of the Big Ten in this timeframe is 127.59
• The average yards per game passing in this time turns out to be 216.69 yards
• The cumulative winning percentage of the conference? 0.562

So, once again, I laid all this out in an egregiously large spreadsheet and then put it aside to do some actual work at work. I came back to this later and decided to pay particular attention to four factors which are considered in the efficiency statistics. In this case, I thought it would be interesting to use the following – average yards per game, touchdown percentage, interception percentage and completion percentage.

As it turns out, there were only 27 passing attacks in the group which were above average in all four areas, but here is what those teams were typically capable of doing:

• Average completion percentage: 61.02%
• Average interception percentage: 2.22%
• Average touchdown percentage: 6.30%
• Average yards per game: 248.16
• Average passer rating: 142.65
• Average yards per attempt: 7.68
• Average yards per completion: 12.59

These are noticably better than the grand means in each category. Another interesting improvement is in total years for the season. For the entire sample, it was 2,669 yards, but for this statistically elite group, it was 3,102 yards. Further, the cumulative winning percentage of this group is 0.653, so having an efficient passing game gets perhaps one more win each year in the Big Ten.

I eased the restrictions a little for the next sort just to see if I could squeeze out a list of the best of the best, if you will. For the next step, I took teams from the smaller sample that were above average in at least two of the four statistics and managed to get a group of 13 teams. Their means are:

• Average completion percentage: 62.40%
• Average interception percentage: 1.97%
• Average touchdown percentage: 6.65%
• Average yards per game: 262.17
• Average passer rating: 148.89
• Average yards per attempt: 8.00
• Average yards per completion: 12.82

Those teams that made the final cut under these assumptions are:

TL;DR CONCLUSION:

Like the rushing version of this from earlier in the week, the point of this was to simply run through a short exercise on finding a potential way to discover from a large set of data which teams stood out among their peers in the conference in a specific set of statistics. I chose to go with statistics that I thought pointed towards an efficient passing attack, not necessarily the most prolific, although the two do in fact overlap somewhat. There are probably better ways to think through this, but I was working with easily available data.

It is also rather intriguing that, at least under my own assumptions in doing these two diaries, having an efficient passing attack and an effective rushing game produce the same typical bump in winning percentage, at least when looked at separately like this.

RANDOM ENTERTAINMENT:

Because I was missing "Animalympics" earlier...

“CHARTING THE JOURNEY THROUGH CONFERENCE PLAY”

Now that there is some respite from the meat grinder that seems to be the Big Ten basketball schedule (at least until this weekend), I felt it might be an appropriate moment to step back and look at some of the basic numbers and breakdowns for our Wolverines. Much has been said in the postgame threads over the last stretch of games, and indeed, some of it bears itself out in the trends that you will see here. The caveat here is that the conference schedule is not yet complete and these numbers are not final.

I had wanted to do something like this since the Indiana game, but I held off because there simply was not enough data regarding conference play to make much of a determination about where the areas of focus should be at that point. Now, I think you can see some definite trends. I also compiled our statistics in a Michigan win versus Michigan loss format so you can easily see just how stark some of the differences are in some cases.

TABLE 1 – “Michigan Win Vs. Michigan Loss”

The one thing that leapt out immediately, at least to me, is that in conference play, we are shooting about 11% when we win as opposed to when we lose, which is significant considering that our four losses have been at the hands of some of the most defensively efficient teams in Division I basketball, not just the conference. The difference is small for our opponents, who shoot only about 6% better in wins as opposed to losses. It’s a fairly similar story for three-pointers – we are down about 13% in losses compared to wins, whereas our opponents again show a difference of only 6% between the two scenarios.

Here is the shooting data broken out into individual games:

Many of the findings aren’t entirely unexpected – we have fewer assists when we lose, we rebound less, and so on, but there are actually sustained trends that are worth noting at this point. Below are trends for point totals and the running average of points:

In both of these, you can see an overall decline in our own production and a gradual increase in the production of those we have played. Since Indiana, in fact, we are giving up three more points per game on average, which may not seem like much, but when you consider that the fewest we have given up since then is 65, it is noteworthy. Our average in the same period has declined about two points, but our average is bolstered some by some of our early performances in conference play.

Tied somewhat to that would be offensive and defensive efficiency, shown below. This is the running number, cumulative as of each game:

The trends are obviously not favorable, but overall, the efficiency numbers have not slid too much, as you will note. In both case, it is less than a 10% slide. It is enough, however, to say that there are items to address soon on both sides of the ball.

Rebounds and assists have also tailed off somewhat, but turnovers show one notable aberration:

TL;DR CONCLUSION:

This is here for your perusal. The discussion which hopefully follows will become the conclusion of the board, or at least that is my intention. If there are other statistics that you would like to see charted, let me know and I will insert the data as time permits. I thought I might just get the discussion going with what I did here.

OBLIGATORY:

“SEEKING RELATIONSHIPS IN THE RUSHING GAME”

In a similar fashion to a diary about correlations in the passing game that I presented a little over a week ago, I decided to do a quick analysis of all 144 regular season Big Ten games and their rushing statistics. Once again, this is to see if the relationships that we believe we see do in fact present themselves mathematically in some fashion.

One of the driving forces behind this particular analysis, as with the passing correlations, is to probe certain (perhaps insignificant admittedly) aspects of the so-called “eye test”, which to me has always been a somewhat nebulous term that people used to encompass broad perceptions of games, regardless of the actual relationships which may exist within the game.

I managed to collect the total carries, total yards and yards per carry for each of the 144 regular season games for the 2012 season, so this should be a sufficient sample size to find some evidence of the three tested relationships that I chose to look at. These are “Carries / Yards”, “Carries / YPC” and “Yards / YPC”.

RESULTS:

DISCUSSION:

On two of the above relationships, there is a rather strong correlation. In the case of carries being positively correlated to total rushing yards, we know that this is generally true unless your team either has a substandard rushing attack which they wield anyway, or alternatively when a team runs often against a superior rushing defense. Both scenarios are evident in the individual game statistics of the Big Ten this season. Yards and Yards Per Carry have the strongest relationship, and in my own opinion, one of the major contributors to that are games where good rushing teams have gone against substandard rushing defenses, accumulation prolific yardage and, by extension, longer runs.

So, as was true with the passing diary a week or so ago, you aren’t necessarily being told something you didn’t already know implicitly. The idea here was to explore the possibility that the relationship you believe you see does in fact exist in the numbers, and in the case of the rushing game, at least for this season, it would seem to be the case.

SUMMARY STATS:

ONE OF MY FAVORITE MOMENTS FROM "THE CRITIC", JUST BECAUSE:

For a number of years now, it has been a popular position that the time off between the end of the regular season and the beginning of bowl season is excessively long, and leads to poor play early in games. Certainly I have found this to be an attractive explaination for the prodigous quantities of DERP we saw this bowl season.

But DERP is subjective, and the plural of anecdote is not data. Is there any substantial quantitative support to the notion that bowl games start off unusually sloppy? There are a great many factors at play; I chose to look at relevant statistics on a quarter-by-quarter basis. I don't have the massive database that seems popular around these here parts, and the typical box score doesn't give a per-quarter breakdown of anything but score. Bummer. Well, maybe if we go ahead and look anyway at a

Chart?

chart of percent of total points scored per quarter, we can find something elucidating. All bowl games are included. Each team's per-quarter scoring is normalized by their total score in the game. Averages and standard deviations are then computed, based on the entire bowl team population. It seems plausible that excessive pre-bowl layoff will result in a substantially higher standard deviation in the early part of the game, when either offense or defense might be DERPerrific.

Well, hmm. We could weight the per-quarter score fractions by the total number of points scored per team (0-3-3-3 is less telling than 0-21-21-21), but we find that this makes little difference. In the unweighted case, we find that the first half standard deviation is 29% higher than the second half standard deviation. The first quarter standard deviation is about 16% higher than the average quarter.

This seems to lend mild support to the idea that bowl games start off unusually sloppy. How does this compare to regular-season results? I compared to games from weeks 12-15, but only if the game involved two teams that ended up bowl-eligible (I counted Arizona State, because I graduated from there and it was Wisconsin's/SJSU's fault anyway and if you don't like it then tough). I toyed with the idea of removing rivalry games, because rivalry games are weird, but I did not.

Stupid Fresno. Anyway, we continue to see elevated stdev for the first quarter of regular season games between bowl eligible teams, but by a lesser degree. For these regular season games, the first half standard deviation is 10.8% higher than that for the second half. The first quarter standard deviation is about 11% higher than the average quarter.

There are certainly some serious issues with this methodology. Does a sloppy defense give up more scores to a sloppy offense than when both are playing carefully? I don't know the best answer, but I'm sure it varies on a case-by-case basis. There are also many late-game effects for which I have not accounted - a prevent version of a dominant defense might give up the only score of the game in the 4th quarter when the game is out of reach (hi there Sparty!). This would give a large standard deviation value to the 4th quarter, erroneously implying sloppiness. I do not know how to account for these sorts of errors with the data set I have available. Further, I do not suggest that this is an all-inclusive list of methodological problems.

Still, 29% vs 10.8% seems vaguely compelling, give 70 bowl and 98 regular season scores. My statistical background has faded badly since undergrad, so I'm going to refrain from a hilariously misguided attempt at error bars. The sample size is large, but boy those data are noisy. Any time my standard deviation is as big as the average, I start to feel a little woozy...

So, I had some spare time at work and decided to look into something that I have been wondering for a while: does it seem like Rich Rod sticks to the run too much, even in games where we are losing?  At this point I’m sure you all know RR likes to run the ball—duh.  But how often, exactly, does he run the ball?

This analysis is just a basic overview of my dataset.  I will follow up with more in-depth looks at point margins and down and distance, but I thought you all might be interested to see basic percentages for our offense through Purdue.  If anyone has anything specific regarding playcalling vs. score margin vs. down and distance, let me know and I will see what I can do.

A few notes and stipulations on the dataset: 

1.) All data is taken from Brian’s UFRs for games this season. 

2.) Analysis stopped at end of UFR, so if Brian didn’t include it in his UFR (i.e. blowouts), it is not in my analysis.  If Brian doesn’t think its worth looking at, well, neither do I.

3.) Plays in which either team got a penalty are included ONLY IF the ball was snapped, since if it didn’t, we can’t know what play was called.

4.) Every snap weighted the same regardless of time left in half/game, because my thought process, we are almost as likely to run the ball in a 2 minute drill as the rest of the game (almost, though not quite).

5.)2 pt. conversions left out.

Disclaimer: 4th down numbers are not very accurate due to low sample size.

Now, for a chart:

A few things here are obvious.  First, RR likes to run on 1st down (69.1% of the time).  His affinity for running decreases every down, with the exception of 4th down.  Most of the runs on 4th down are 4th and short, and the passes 4th and long.

Next, we will take a look at the distribution when the game is tied:

What we see here is that RR is more likely to run the ball on every down when the game is tied than his average, except fourth down.  

Now, we look at when we are winning the game:

So, RR is more likely to run when we are winning, though not to a statistically significant level except on 1st down.

And when we are losing:

So, RR is much less likely to run on all downs (except 4th) when we are losing.  

This data all flushes out pretty much as expected, but I thought I'd share anyway.  In the next edition, I will analyze run-calling affinity per down based on score margin.  Stay tuned, folks for all upcoming editions of 2010 Playcalling so far: An Analysis.

P.S. If anyone is interested in seeing my whole dataset for validation or to do your own analysis, please leave a comment/message me (can you message people on here?) and I will be pleased to share!

EDIT: Title and Tags edited to be more informative