Due to an unforseen dump taken on me by some superiors, this is just a little late. I'll have this week's BCS Breakdown up before the end of the week, and we'll be on schedule from there. -HB
It's been a long wait for some of you, but I've finally returned with my spreadsheet to make shutting off your brain at work just that much more difficult. One day a week you'll find yourself minimizing your Excel documents, pulling up Burnt Orange Nation, and looking at mine, which is exactly what you want. OK, you say, I get it, cut to the numbers.
That would be easy, but all in all, too easy. First, I would like to point out that there potential advantages of writing this breakdown after the week's games have been played. I can look at the data for a team and compare it to their game this week. Maybe we'll glean a little knowledge as to whether the polls can point out over/under-rated teams to us beforehand. It's a good idea, but we'll have to check it out first, and so, on to the madness...
First we'll take a look at how well the voters agree on each team's ranking as compared to my predictive model that I created in my original BCS analysis way back when.
Sorry about the crappy picture quality, the teams are in order of BCS ranking, increasing from left to right, in case you can't see. Since this is the first post of the season, I'll go ahead and explaborate a little. This graph shows the standard deviation (which I'm saying is a measure of uncertainty by the voters) of a team's ranking minus the expected value from my model, which is just a curve fit for all these values averaged over several years. Last year, I said that teams with significantly higher than expected standard deviation were good candidates for upsets, as usually the computers are down on them while the humans are high. Sometimes it can be the other way, but usually it works out like that, as there are rarely any really talented teams outside of the usual suspects that the humans like to vote up anyway. In theory, all of the bars should be small and pretty randomly distributed above and below zero (expected), but that usually doesn't happen until more towards the middle of the ranking period, as right now the humans and computers haven't seen quite enough to equilibrate.
That said, the first thing you should notice is that there are a lot of really tall bars, which means uncertainty is very high. A little of that is to be expected in the first week of the poll, as it's unlikely that many of the human voters have seen the computer rankings before now and haven't had their ballots subject to a second opinion, but I think this year a lot of it is due to some of the wacky upsets and inconsistent play of the teams. Take USC, and tOSU for instance. The computers and humans basically have their rankings flip-flopped. USC is 4th/5th with the humans and 10th in the computers, and tOSU is 5th in the computers and 10th with the humans. Did the computers not see that game? No, they didn't. As far as they know, USC beat tOSU 1-0 in a game exactly like every other game ever played in the history of football. I'm not sure where Ohio State is getting credit for their schedule, but it doesn't look like the Oregon State loss shines very favorably on USC. (Note: This is last week's poll, tOSU and USC are T-10th and 6th in the computers this week. No MoV to the rescue after that slop-fest against Arizona.)
So what can we tell from the performances we just saw Saturday compared to this graph? Penn State had an enormously high standard deviation(~3.2), but came through and beat tOSU(~0.7). Looking at the numbers, the computers hated PSU at #7, likely due to their weak schedule. This we can chalk up to the computers not having enough info to get the ranking right. USC(~2.1) nearly got stuffed by Arizona(NR), which seems to indicate that the computer hatred directed at them previously had merit. Tech(~2.2) obliterated Kansas(~1.4), which gave them some computer love where they needed it. Florida(~1.8) ran through Kentucky(NR) like a ski-boat through a manatee, showing that they are likely underrated by the computers. LSU(~2.9) got smoked by Georgia(~-0.8) in Death Valley, validating the computer hatred (T-19th!). TCU(~-1.9) went all Florida on Wyoming's(NR) Kentucky, which may mean that we know how good TCU is, if everyone is already in agreement. I think that came from playing OU pretty well early on. The rest is just garbage.
Notable games without wacky standard deviations involved? Hmmm... well, Okie State hung close with Texas, validating their computer love, and that's about it. I'm liking the results here so far for my theory. Games with wacked standard deviations either validated computer love in cases where the computers had enough to go on, or exposed the computers' lack of knowledge in games where teams hadn't been tested. I'm thinking this week's numbers could help you with you picks this week, if you're willing to give it a shot. I'll have them out ASAP.
Billingsley Report Card:
Ahh, yes. Our old friend, RB. Sometimes I'll have trouble piecing together a post, but this section always just flows right out onto the screen. It helps when you've got such a slam dunk caseload of evidence to back up your claim. Sorry for the lawyer speak, I'm actually pressed for time, so I'll have to pass up this opportunity to fire off a few shots. Luckily this week's numbers are already here, so my next opportunity isn't far off. Madness:
Billingsley looks crazy. Surprise. He got thrown out 14 times out of 25, and had the highest standard deviation from the average by almost 2.5 points. I'm watching an old episode of Dexter right now and Billingsley makes this look like an episode of Full House. Feel free to use the comments to point out your favorite ranking redonks.