How Much Should The D Improve? Not Enough

How Much Should The D Improve? Not Enough

Submitted by Brian on July 12th, 2011 at 12:17 PM

Let's take some small sample sizes and extrapolate wildly. It will be fun. Here's Bill Connolly breaking down expected improvement from teams that return varying numbers of defensive starters:

So Cincinnati returns 11 defensive starters. That's probably a good thing, right? But how good? And how much can a bad defense improve in one offseason just because of experience? Let's take a look, shall we?

Average Change In Def. F/+, Last Three Years
N Avg Chg in
Def. F/+

1 1 -12.4%
2 4 -10.9%
3 10 -8.4%
4 32 -2.1%
5 53 -1.1%
6 69 -0.5%
7 85 1.1%
8 56 1.5%
9 37 4.2%
10 9 6.0%
11 3 5.4%

So basically, if you return between five and eight starters, you are likely not going to change much, but three or fewer is a problem, and nine or more is a good thing.

F/+ is Connolly's advanced metric; it's play-based instead of drive-based like FEI. Don't be fooled by the % symbol—the metric is percentage based and from context it's clear the difference is meant to be added to the score, not multiplied. Since the best defenses are around +17% and the worst around –13%, 6% is about a fifth of the entire scale.

Michigan is, unsurprisingly, right at the bottom of that scale at 115th. They were 12% worse than an average defense down-to-down. The good news is they return 9-ish starters, losing Greg Banks, James Rogers and Jonas Mouton while reacquiring Troy Woolfolk. (They also lose Ray Vinopal and Obi Ezeh, but Ezeh had been replaced and Michigan should get JT Floyd back so let's call it a wash.)

The numbers are thin at both ends of the spectrum but, hey, extrapolating wildly from small sample sizes. Doing so says Michigan's defense will storm forward from 115th nationally to…



I have no source for this, unfortunately.

But wait! Our sample sizes are not small enough and our extrapolation is not making out with other nubile young extrapolations in front of a television camera. Bill added a second factor, the previous year's defense, and finds that a defense with an F/+ under –10% that returns nine starters should expect (for a given confidence level that is not high at all) to improve by 8.6%, which would see them get to…



You might be able to argue that Mike Martin wasn't right and the team was even younger than the average team that returns nine starters and GERG is rubbing stuffed animals on the faces of other stuffed animals at a tearful tea party and for the first time in a long time they'll just run one damn defense per year and that they should expect to improve even more. You're probably setting yourself up for disappointment. Like installing the spread 'n' shred, digging out of a hole this big is a multi-year project.