MNCs, Rankings, Probability Fields and RPS

This time of year, there are many discussions about ranking various teams and who "deserves" to be the MNC. The discussion, and the way humans and computers rank teams, assumes that each team, or element of a team, has an absolute value (FEI, RPI, ranking, SOS x outcome, etc.). The job of the evaluator is to determine what that value is for each team. I think this is completely wrong. First off, when two teams play, there is a probability greater than 0 that each team will win. Whether that's two teams where the outcome seems 50/50, or where one team has almost no chance to win (but sometimes does). This is much like in nature, where sometimes the cheetah gets the gazelle, and sometime the cheetah starves. One can give odds in advance, but until the game is actually played, there's no actual "better team," there's only a probability that one team will beat another. In sports, there's no one better team, there’s only a probability field that fluctuates until the clock ticks to 0, the probability field collapses, and there’s a winner.

So let's look at the cheetah/gazelle thing again. Over time, one will win out more often than not, and either all the gazelles get eaten or all the cheetahs starve. Yet neither gazelles nor cheetahs are extinct. Why? That's because nature is not a 2-player game. It's more like rock-paper-scissors. Here's an interesting article discussing: Link.  The bottom line is that there are always at least three species competing, and it's almost always an odd-number, just like RPS. Cheetah beats gazelle. Gazelle beats hyena. Hyena beats cheetah (by stealing his food). (Yup, look it up: Here).

So what does all this have to do with football and ranking teams? Well, first off, trying to pick which team is better based on results on the field, while the best method, is far from perfect. The sampling size is just too small. Secondly, even if results were absolute and replicable, RPS makes a hash of rankings. I don't believe the transitive property would apply, even if sampling size were large enough, because different teams, like different species, adopt differing strategies. UM regularly beat supposedly "superior" ND teams, which would beat MSU, which would beat us. While luck is involved, I think it was also that our teams were particularly well-suited to beating ND (Denard), but not necessarily well-suited to beating MSU (anti-Denard). A couple obvious notes should go along: 1) Teams "evolve" like species, so a team may be poorly-suited to beat a rival one particular year, or even in one week, but well-suited the next (See, Ohio 2010, 11). 2) The basis for what will make one team well-suited to beat another is not always obvious. Some manball teams do very poorly against spread teams. Others seem to do quite well. Coarse analysis will not work.

So what is the difference? I think that the items most often discussed (run offense v. run defense, etc) are all but useless. If they worked, Vegas would be broke. My guess is that the differences are often due largely to luck (oblong pork bladder, players’ fragility, and the law of averages discussed above). Some significant fraction of the difference, however, is based on metrics that are difficult, but not impossible, to determine in football. Here are some elements that I think might be relevant but are almost never discussed in game previews, though computer analysis would likely be required to prove/disprove:

1)            Blocking style vs. defensive style: does the offensive team use reach-blocking? Does it pull linemen? Does it emphasize speed or strength? Does the defense emphasize speed or strength? How does it fill gaps—with LBs or DBs or DL? How are players pad levels? How well does it emphasize tackling in space? Notice that none of this necessarily has anything to do with 3-4 vs. 4-4. It has to do with how one team's philosophy/scheme matches another team's. It's why a team like Iowa may do well against UM but not MN or ISU.

2)            Running style vs. containment style: Do runners tend to run North-South or bounce-bounce-bounce? How often do they cut back in open seams vs. following blockers? How fast is the D to the corner? How aggressively does it attack gaps? Does the defense sell-out to the LOS? How well do CBs come off blocks?

3)            Aggressiveness: Does the team tend to gamble? In what situations? Is it predictable? How good is the other team at predicting? A good example on this one was Borges calling conservative plays against Illinois. Seeing how well the Defense was playing, a conservative approach was appropriate. Against Ohio, not so much.

Note that each of these metrics, which are themselves neither exhaustive nor all-encompassing, impact each other. The running style of the offense under #2 is affected not only by the scheme of the defense, but also by factors under #1, e.g., the blocking style of the offense and the DL style.

What does this mean going forward and in reviewing the season past? I have a couple thoughts based on my memory, but I would be interested in discussing others’ views: 1) Our offense tended to do well when it either could take advantage of having extra blockers or else could manhandle the DL. It did poorly against MSU and VaTech because our emphasis on line play and speed in space (exacerbated by injuries vs. VT) was a poor strategy against their personnel and schemes but a good strategy against other teams. 2) Improvements on our defense this year are too great to be explained solely by greater experience, increased talent and improved coaching. A big part of the difference is that our survival strategy changed. My theory is that 3-3-5 works against teams spreading the field. In that sense, I think the 3-3-5 is not dissimilar to VaTech's defense, which is good at stopping lateral spreads and offenses based on speed. It did very poorly in the B1G. 3) Borges’ potluck approach this year is good in an environment where one plays a number of different types of teams, but would be less effective against very good teams that require a very high level of competency in a specific strategy (See, Bo’s teams, success vs. Cooper). 4) Given the complexities involved, and how teams develop over a season, it's no surprise that pre-season predictions are so horrible. Going into the season, I thought we would be best served with a Coker/Hopkins-type-substance that emphasized power. Over the season, we were best served by Fitz’ slashing-type running, based on an ability to see the holes developing. He didn’t even have the vision necessary going into the season. Finally, naming a MNC, or even coming up with a ranking, is an exercise in futility. How do you rank rock vs. paper vs. scissors?

I know this was long, but I felt a need to get this down, as it seems like much discussion assumes that one team will be better than another based on some absolute value. LSU is a 10. Bama is a 8. LSU therefore will beat Bama. I think reality is much more nuanced, and that one must break out particular values for numerous variables to have any real guess as to what team will beat another most often, and even at the end of that process, there is no one "best" team. I’d be interested in what others think are relevant metrics and what strategies would work best in the B1G generally and against Ohio and MSU in particular.