SEC Power Poll: Week 3 Ballot

by David Hooper on Sep 23, 2009 6:03 PM EDT

The SEC Power Poll will be released shortly this evening; here is the RTT ballot. It's based on the same computer engine that is currently tripping out over the BlogPoll, but with one caveat: the actual power poll ballot submitted is slightly different than the calculated rankings. This is because the calculated rankings were last tweaked after the ballot was submitted. This makes no significant difference, even despite flipping a couple of teams, because some teams are too numerically close to care about their differences at this stage anyhow.

Powerpoll2009logo_medium

Explanations after the jump.

Here are how the teams currently ranks in the RTT computers:

SEC	Team	WL	SOS	PED	RD	3DO	TD	PEO	OPPG	TO	3DD	W SUM
1	Florida	1.338	-1.318	1.627	0.723	2.586	1.449	1.756	1.705	2.054	0.626	107.308
2	Alabama	1.338	-0.123	0.675	1.715	0.622	1.927	0.990	0.866	1.735	0.739	106.809
3	LSU	1.338	0.995	1.023	0.080	1.348	0.261	0.496	1.230	-0.710	-0.562	77.126
4	Auburn	1.338	0.447	0.772	-0.432	1.195	-0.115	0.918	0.063	1.770	-1.061	75.541
5	Kentucky	1.338	-1.206	0.993	0.676	1.809	0.719	0.403	1.030	0.487	0.280	64.633
6	Mississippi	1.338	-1.948	1.362	0.292	0.346	0.595	0.607	1.413	0.585	2.607	45.873
7	Tennessee	-0.781	0.687	0.957	0.461	0.323	1.779	-0.539	0.720	-0.282	0.590	22.978
8	South Carolina	0.278	-0.214	0.157	0.867	-0.635	1.135	-0.232	0.318	0.260	-0.865	21.976
9	Mississippi St.	0.278	0.207	1.263	-0.526	-0.865	0.343	-0.338	0.354	-0.400	0.396	12.275
10	Georgia	0.278	0.667	-0.785	0.236	0.123	-0.808	1.368	-1.214	-0.195	0.476	7.338
11	Vanderbilt	-0.781	0.686	1.128	-0.488	-0.570	0.995	-1.825	1.121	-0.666	1.379	-7.425
12	Arkansas	-0.252	-0.479	-1.936	0.283	-1.440	-0.327	2.105	-0.886	2.072	0.254	-19.006

The only real difference between this and the ballot was that Tennessee edged out South Carolina in the revised computer tally. However, their difference is so slight (1 point) that it's really the same as saying they're tied in the computers. Despite Tennessee's extra loss, their better numbers in most categories make them indistinguishable in the computers. Some interesting in-conference notes:

Arkansas and their lack of defense: Look at those defensive numbers. For a rough guide, anything that's -1 is pretty bad. Numbers around -1.5 (3DO - 3rd Down Offense) to -2 (PED - Passing Efficiency Defense) are outright horrific. Since they basically only pass the ball and neglect all other phases of the game, anybody looking for supplemental scouting tape can dredge up old 1980s WAC games. They'll do in a pinch.
But... having a bottom score of only -19 is phenomenal for a conference. No other conference has a better bottom end.
Georgia suffers a similar malaise. Their opponents have overall been better, but UGA still has no defense. Improve Arkansas's 3rd downs and decrease their 1sts and 2nds, and you have Willie Martinez.
Balanced is best. The top five teams all have very well-balanced metrics, which seems to be the running theme in the computer in general. More on that in coming weeks as the numbers stabilize.
Tennessee's defense is really that good. If you compare Florida's offense numbers (anything ending in an 'O') to Tennessee's defensive numbers, you see that both units are among the best in the league. Factors in that they account for a third of each others' scores, and the effect is remarkable.

Looking around the league, here are how the conference averages stack up:

Rank	Conference	WL	SOS	PED	RD	3DO	TD	PEO	OPPG	TO	3DD	W SUM
1	SEC	0.587	-0.133	0.603	0.324	0.403	0.663	0.476	0.560	0.559	0.405	42.952
2	BIG EAST	0.477	-0.156	0.254	0.770	0.257	0.615	0.418	0.471	0.349	0.160	35.639
3	PAC-10	0.490	0.195	0.316	0.587	-0.199	0.358	-0.173	0.506	0.022	0.470	28.802
4	BIG 12	0.411	-0.286	0.316	0.435	0.236	0.190	0.230	0.593	0.672	0.280	26.856
5	BIG 10	0.519	-0.133	0.001	0.294	0.525	0.074	0.244	0.308	0.305	-0.065	24.473
6	ACC	0.102	0.576	0.276	0.027	-0.100	0.276	0.126	0.140	-0.240	0.476	17.088
7	IA-IND	-0.075	-0.468	-0.176	0.486	0.443	0.193	0.105	0.002	-0.103	0.199	-1.809
8	MWC	0.043	-0.653	-0.017	0.025	-0.180	0.036	0.215	-0.053	0.003	0.009	-9.483
9	MAC	-0.781	0.482	0.041	-0.654	-0.275	-0.263	-0.317	-0.490	-0.660	-0.504	-33.590
10	C-USA	-0.252	-0.468	-0.593	-0.152	-0.108	-0.624	-0.298	-0.650	-0.274	-0.383	-36.124
11	WAC	-0.664	0.293	-0.650	-0.875	-0.158	-0.713	-0.214	-0.592	0.045	-0.599	-39.978
12	SUN BELT	-0.840	0.405	-0.693	-0.832	-0.588	-0.711	-0.591	-0.926	-0.637	-0.410	-55.256

The SEC is still on top. Something else to note is that, despite the -efenses of Georgia and Arkansas, the SEC has the best defensive metrics of any conference, leading in Pass Efficiency Defense and Total Defense, and third in 3rd Down Defense. Among the major converences, only the ACC and PAC-10 have higher strengths of schedule at the moment, so it's not entirely a cupcake problem.

If you're surprised to see the Big East at number two, don't be. So far, none of the Big East teams have had truly sucktastic starts, and minimizing the bottom end is the biggest difference between the top conferences and the middle of the pack so far.

The MWC is leading the mid-major battle by a considerable margin, but last weekend really hurt them. Before then, they were actually 5th overall, and wins from BYU and Utah would have had them most likely at 4th or 3rd. But hey, that's why the games are played. I'm not sure the MWC can recover enough to get back into the top six, but it'll have to come from the bottom of the conference, where San Diego State, Wyoming, and New Mexico are dragging the anchor.

Tweet Comment 10 comments | 0 recs |

Do you like this story?

Comments

Just curious...

can you work moral victories and some sort of flu index into the formula?

by CornFromAJar on Sep 23, 2009 9:13 PM EDT reply actions 1 recs

Can, yes.

LOL

by David Hooper on Sep 23, 2009 9:45 PM EDT up reply actions

lol

Tennessee WILL beat Georgia on the way to 9+ wins in '09!!!
Eric Berry For Heisman!!!

by VolBrian on Sep 23, 2009 9:54 PM EDT up reply actions

Perhaps I am merely stupid

But how does the WSUM figure correspond to the other columns?

The numbers don’t seem remotely close to adding up, unless you use various weights that I’m not aware of. In fact, as I type, WSUM could quite easily mean Weighted Sum. So, I guess, where can I find the weights?

by Nashville on Sep 23, 2009 9:55 PM EDT reply actions

Oh, sorry about that.

I explained that in better detail here, but forgot to link it in. At that other link, the weights are listed below the big table. Use the Normalized Weights, as copy/pasted here:

Weights, Normalized: W/L = 27.47, SOS = 16.48, PED = 9.89, RD = 8.79, 3DO = 7.69, TD = 10.99, PEO = 5.49, OPPG = 3.30, TO = 8.79, 3DD = 1.10

Just to be sure I’m clear, you multiply each stat by its corresponding weight, then add up all those terms to get the weighted sum. (I’m sure you understood, but I’m just trying to be sure I’m thorough.) And if you think there’s a better way to weight the factors, please speak up. This is still a work in progress and there’s a lot to nail down.

With the exception of W/L and SOS, the order in which the different factors are listed are the order in which they correlated to team success over the last few years. So the earlier factors tend to get heavier weights (though I bumped up TO and TD because it’s so early in the season).

by David Hooper on Sep 23, 2009 10:09 PM EDT up reply actions

TYVM

I appreciate the link. Neat to see the total data segment and it makes more sense now.

I am frankly not good enough with statistics to critique your data presentation or weighting decisions, but I do have several questions which may help me understand your decisions.

First, where are you getting the SOS from? I didn’t realize that the NCAA itself had developed a way to track it and I couldn’t find it among the list of statistics it tracks.

Second, why do the non-normalized weights only add up to 91 and not 100? (I note that the normalized weights add up to 99.99, so this may just be innumeracy on my part.)

Finally, as an offensive metric, you may wish to consider adding non-penalty first downs offense. It’s a pretty good measure of how proficient each team is at consistently generating offense, which is a key to sustaining drives. If it’s too much difficulty to separate out those FD not earned on penalties it’s no big deal, as the remainder will still be a pretty good proxy.

Thanks for listening Hooper, very interested to see how these rankings shake out over the course of the season.

by Nashville on Sep 24, 2009 12:51 AM EDT up reply actions

Oh!

One final request. It may well not be possible, but if you could make each column sortable, that would be pretty interesting to see as well.

by Nashville on Sep 24, 2009 12:52 AM EDT up reply actions

Thanks. I really appreciate the feedback.

I’ll take a look at the non-penalty first-downs offense. That’s not a bad idea, as first downs are much like hits in baseball: so long as you keep getting them, your half of the ‘inning’ resumes.

Re: the SOS

I use Colley’s SOS numbers (the decimal values, not the rankings), primarily because they can be linked to (unlike the NCAA pdf sheets).

Re: the weights.

I adjust the non-normalized ones to play with the balance of the inputs. But I don’t worry about making them add up to 100. (What I do is I divide each weight by the sum, then multiply by 100 to get the normalized weights. The normalized weights are the ones that are actually used, and since they always add to 100, the magnitudes of the sums are consistent from week to week.) The normalized will also add up to 100 if you had all the digits; I truncated the numbers before presenting to make them easier to read on the web page.

By using that non-normalized / normalized system, I can tweak the weights without having to think much. Say I want to increase the effect of SOS. I can simply give SOS a bigger number and the spreadsheet automatically calculates the right adjustments to everything else for me.

Re: the weighting decisions.

Those are arbitrary with guidance. It’d take me a while to find the data again, but these were the strongest metrics that Doc Saturday (who was, at the time, SMQ) had found correlated to team success. If I had the time, I’d set up an actual regression model for the last several years’ worth of data to find the right balance of weights, but I can’t take that long on it so I play the weights by ear. As a general rule, the weights will descend from left to right, as that’s the order of correlation. But I’m open to thoughts on this.

Re: sorting the data

I’m working on a way to provide the data in sortable fashion and hope to have that out in the next week or two. In the meantime, you could always copy/paste into Excel. (It may be easier to copy/paste into Notepad first, then into Excel from there.) Then, either define the table as a “table” in Excel (if that makes sense), or simply highlight everything and use the Sort command. (I recommend that you don’t include the “rank” column when you sort.)

by David Hooper on Sep 24, 2009 8:30 AM EDT up reply actions

Thank you. Like I said, some of it’s above my head, but very interesting stuff.

by Nashville on Sep 24, 2009 5:48 PM EDT up reply actions

Thanks for catching me on that, by the way.

I will have a better means of reporting the weights in the future. Joel used to make the spreadsheet available for download every week, but I’m not quite there yet. I’ve really done a lot of tweaking to it, but I do hope to release it for review in the future.

by David Hooper on Sep 23, 2009 10:13 PM EDT up reply actions