Consistency of Performance

Football Outsiders just had a post on covariance, which I'll describe in a moment.  You can read the full article at but I want to highlight what it says about VT thus far this season here rather than rehash their entire article.

We're all familiar with teams that seems to always play to the level of their competition, and others than seems to do the opposite.  Covariance is a measure of that.  It starts with the idea of "Adjusted Scoring Margin" which takes both teams in a game and converts their performance into performance against a hypothetical average team.  So if you play really well against, say, Alabama but lose 20-17, you probably had a blowout performance against an average team and the adjusted scoring margin might be something like 34-9.  Makes sense, right?

The heart of the article is about how the adjusted scoring margin for each team varies along with the strength of their opponent.  So if your adjusted scoring margins are great against good teams and poor against bad teams ("playing to the level of the competition") then you get a high negative score; if the opposite is true, you get a high positive score.  Wazzou in particular is highlighted, as they kept it close against some of the best in the country but then laid eggs against some of the worst and have the second most negative score.

In this regard, since I'm sure you're curious, Virginia Tech is slightly positive but not enough to matter this season - basically how well we play seems unrelated to how good the team we are playing is.  Also, for what it's worth, there is little if any relationship between this metric and how good you are, so it's more just a measure of the "personality" of the team.  There are great, and terrible, teams on both ends of the scale.

But here is what is interesting...along with the covariance metric, the table also reports the standard deviation of performances (again using adjusted scoring margin) as a measure of consistency.  A very consistent team will have a low standard deviation, while a very inconsistent team will have a high standard deviation.  Again this is not really correlated with how good your team is, with good and bad teams at both ends.  Here are the top 10 teams based on standard deviation:

Bowling Green
Virginia Tech
North Carolina
Georgia Tech
Std Dev

I don't think it's huge news to anyone reading this, but we have been really, really inconsistent this year - 5th in the country in fact.  But I thought it was interesting to see some numbers around it and see just how we have ranked relative to the other teams in the country.



Negative Covariance Also Means Equaly Dominance on All Teams

Pretty cool statistic, but it undermines the truly dominant teams.
Oregon (4th most negative) and KSU (8th most negative) are dominating all teams, good and bad. But, because they're beating the good teams by 20+ and the bad teams by slightly more, they're achieving a high negative covariance. So, while an indicator of playing to your level's competition, negative covariance is also indicative of being completely dominant, regardless of competition.

VT is very close to zero, which is probably more to do with the lack of ability to play well on the road, regardless if the team is good or not.

It's definitely not an indicator of whether a team is "good" or "bad"...more just a personality thing.

The standard deviation thing is interesting, because the two (along with the variation in your opponents abilities which is not reported) are mathematically tied. Assuming you play a variety of teams (as most have easy opponents, average, and difficult ones), then in order to have a highly positive or negative covariance you would have to also have a pretty high standard deviation, because you're varying wither with or against your opponents abilities which are inherently varying (I have a graph demonstrating this).

Unfortunately they don't show the adjusted game scores mid-season, so we can't pick apart different performance but instead just have these summary statistics. But Kansas State has not been that consistent at all - their standard deviation is 14.

as a stats nerd....

thank you.

Loria, Beamer, Wiley, Russell, Drakeford, Gray, Banks, Prioleau, Carpenter, Charlton, Midget, Sorensen, Whitaker, Bird, McCadam, Pile, Hall, Wilds, Green, Fuller, Williams, Hamilton, Rouse, Flowers, Harris, Chancellor, Carmichael, Whitley, Hosley, Fuller