Statistics on Predictability, Part 2

As a follow-up to last week's post on predictability of our offense, a couple of commenters mentioned that a better comparison might be to the best teams in the country rather than all teams (since we aspire to be one of the best and most years are). In this post we'll compare the behavior seen by VT with that of the top twenty offenses in the country as ranked at http://www.footballoutsiders.com/stats/ncaaoff. We could debate all day about whether these are the exact 20 best, but I don't think anyone will argue that those are pretty good offenses that move the ball well, and the rankings on the site are adjusted for opponent's defensive ability.

In the first part of Part 1 we saw how VT quickly abandons the run when down and turns to heavy run-use when up at any point in the game, and while that behavior is somewhat expected, we are much more extreme than others. Here is the same analysis, but only using the Top 20 offenses instead of all other teams:

Unfortunately the same pattern exists. It is true however, that the Top 20 offenses are a little more extreme than average, but not nearly as much as we are. The data would be painful to collect, but it would be interesting to see if teams with extreme behavior like ours are also less likely to come back from a deficit or hold a lead.

Next we saw that on third down, VT's behavior becomes much more predictable than others depending on the yardage and field position. We saw our odds of rushing vary from <20% to >80%, while the remainder of teams only exceeded those bounds in very extreme situations. When looking at the Top 20 offenses, we unfortunately see that they fit the mold of all teams and do not quickly become predictable like the Hokies:

For a minute though, let's forget about the situational stuff and just look at predictability more broadly. Suppose we looked at every play all year and used the same basic criteria to predict odds of rushing, and looked at the distribution of those odds. Obviously you want few plays in the extremes, and most plays towards the middle. Here are histograms of that analysis, with VT's plays on top and the Top 20 offenses on bottom:

VT runs way too many plays where the likelihood of running (or passing, which is saying the same thing) is to easy to guess. The most concerning are the extreme ends, where there is close to a 0% or 100% chance of the play being a run.

Finally, I diagramed all WR Screens to determine how many of them were a good call and how many were not and charted them: