Dullard's Computer's Top 25 College Football Rankings - Week 6

[gistech1978]

Originally posted by: dullard

Originally posted by: Saulbadguy
Kansas #22...woot! Rock Chalk Jayhawks! (Sorry Mikey)

I still think it is a fluke. Kansas will likely start a long losing streak. I'm a student at Nebraska and so I pay attention to the Big 12 a lot. Kansas has impressed me so far, but I think this fairy tale start will not last long. Who nows, maybe they will keep it up. You've got two wins ahead of you: Baylor for sure and either Colorado or Texas A&M. But looking at the rest of the schedule I see many losses (ie your last 5 games).

you dont think the fat man can pull it off?
i think the jayhawks beat the cyclones for their 2 conference wins.
okay 3 i neglected to see baylor.
kansas beating AM or CU? no way.

 

[dullard]

Originally posted by: royaldank
Biggest problems I see are Georgia and Minnesota. Minnesota always opens with their customary tour of local high schools. Didn't Georgia loose and then have a by week? You're saying they are better than LSU that beat them? I know this logic doesn't always hold true, but I think in this case it does. Georgia shouldn't be that high, Minnesota shouldn't be that high either.

I don't think Arkansas should be ranked that high, but I can understand how it's possible given their record and the use of a algorithm. Thus, no compaints there.

I agree with your statement. There are always going to be teams ranked ahead of teams that they lost to. The classic example is when teams play each other twice in the same season - and usually they each win one game. How can any rating scheme ever have both teams ahead of the other team in that case? Or a much more common example: X beats Y, Y beats Z, Z beats X. Clearly at least one team must be ranked ahead of someone they lost to.

Really the main difference betwen Georgia and LSU is strength of schedule. According to my software LSU ranks #89 in strength of teams played so far (Minnesota ranks #85 for comparison). Georgia ranks #6 - only five teams have had tougher schedules so far. The pansy schedule that LSU has played so far drags them down. Contrary to what Dead Parrot Sketch has been saying in this thread, the pansy schedule so far has kept LSU lower in the rankings in my computer program.

As for Minnesota, they have also played a pansy schedule. Increasing the penalty for a weak schedule would pull LSU down as well - and make the score predictions worse. Minnesota has some tougher games coming along. If they can't win them, Minnesota will drop like a rock. But if Minnesota continues scoring 40+ points per game, keeping the opponent to an average of 13 points, and maintains a perfect record - then why not keep them in the top 25?

 

[Thraxen]

Or a much more common example: X beats Y, Y beats Z, Z beats X.

We've seen a big one like that already this year:

Notre Dame beats Washinton State beats Oregon beats Michigan beats Notre Dame

 

[DT4K]

Originally posted by: ShotgunEd
were are the Oregon Ducks? for sure they are better than the stupid beavers

Apparently you missed saturdays game.
The way Oregon was playing, they would have lost to the local high school JV team.

I'm surprised they only dropped to 19 in the AP.

 

[Tom]

You didn't answer my question about what the penalty for losing is..

 

[pecel]

BOOMER SOONER !!

Place Team Rank
1 Oklahoma 147

😀

 

[dullard]

Originally posted by: Dead Parrot Sketch
You didn't answer my question about what the penalty for losing is..

It varies from week to week in order to obtain the best estimate of next weeks scores - so I cannot give a definate answer. Typically you will get about a 2-3 rank deduction if you lose by 0 points to a team with equal ranking as you. Of course, you cannot lose by zero points - the minimum loss is one point since football scores are integers. So in reality the minimum deduction you will get is

(2 to 3) + 1 * (score weighting factor) / 2 = ~ 2.5 to 3.5

The score weighting factor is normally close to 1.0, so we could ignore it for simplicity for now. The score is divided by two since the winning team moves up by half the score and the losing team moves down by half the score. Now if you lose by more than one point you will be hurt even more. Suppose you lose by 10 points. You get a deduction of

(2 to 3) + 10 * (score weighting factor) / 2 = ~ 7 to 8

So a loss of 10 points to a team equal to you will hurt you by 7 to 8 ranking points.

But you also must include the strength of schedule. If you lose to Buffalo by 1 point your ranking will plummet due to the 2.5 to 3.5 rank deduction AND the strength of schedule hit you took for playing a pansy team. However if you lose to Oklahoma by 1 point you may actually rise in the rankings (since the strength of schedule may boost your score more than the 2.5 to 3.5 rank deduction you got). Of course if Oklahoma blows you away you will go down in the rankings.

Do you need a detailed calculation? Lets assume it is the first game of the season for two teams (X and Y). Lets start them both at a ranking of 100. Lets assume team X wins by 10 points played on a neutral field. Lets assume for this week the computer optimization gets a score weighting factor of 1.0 and a loss factor of 2.0. The score is considered first. So team X gets a ranking of 105 and team Y gets a ranking of 95 (note the 10 point differential - one team is moved up by half and the other team is moved down by half). Now the loss is applied. Team X is unaffected since it didn't have a loss. Team Y gets a 2.0 rank deduction. Final rankings are 105 for team X and 93 for team Y. Now do this same procedure for all the games you play, average them together, and iterate until the rankings stay constant and you'll get the season ranking.

 

[royaldank]

Fairly interesting thing you got going here. Sure, there are things to disagree about right now, but the season is young. I'm interested in seeing how it starts looking deep into November. How does your program normally finish up? You normally have the top teams pretty close to how the final AP and Coaches poll end up?

One more question...do you rank the teams at the beginning of the season for use in calculations? If so, where do these rankings come from? Not questioning your methods here or looking for something to harp on, just curious.

 

[Tom]

Where are Wisconsin and Penn State ranked, out of curiosity ?

 

[dullard]

Originally posted by: royaldank
Fairly interesting thing you got going here. Sure, there are things to disagree about right now, but the season is young. I'm interested in seeing how it starts looking deep into November. How does your program normally finish up? You normally have the top teams pretty close to how the final AP and Coaches poll end up?

One more question...do you rank the teams at the beginning of the season for use in calculations? If so, where do these rankings come from? Not questioning your methods here or looking for something to harp on, just curious.

The final rankings are usually within 3-4 places of the polls. For example the polls may have someone ranked #7 and my computer may typically be from #4 to #10. So it isn't exactly the same, but close. The polls are much harsher for losses than my program is (meaning the polls favor a team that has a slightly easier schedule with 1 loss and the polls harm a team with a slightly harder schedule with 2 losses - but those two teams may be equal in my rankings).

By the end of the season it usually predicts the correct winner in 80%-85% of conference games (a bit less with non-conference games). For conference games it typically gets the score within +-7 points. So its average error margin is one missed touchdown pass or one fumble. It may predict team X to beat Y by 20 points but the final score may be 13 to 27. Often it does much better but sometimes it bombs (there are always upsets in football) and the bombs really hurt the average.

If it predicts A to beat B by 10, C to beat D by 10, and E to beat F by 10 but in reality A beats B by 10, C beats D by 10, but F beats E by 11 then the average error is: (0 + 0 + (10+11))/3 = 7 points. Sure it got two games perfectly but the third game which was an upset really means that averge error is high. For all games (including non-conference games) the average error raises to about 10. Its not perfect - and doesn't claim to be. I don't think anyone could get closer than 7 points on average since just one play can affect the score by 7 points.

I answered your last paragraph in a post above. I could easilly use the last years rankings as a starting for this year. But I choose not to since there may be many factors year to year (good players go to NFL or graduate, new coaches, etc). So I set all teams of each division to an equal starting point (but the divisions vary). So before the teams get 3-4 games under their belt the computer prediction is not too accurate.

 

[JohnCU]

YEAH BUDDY! Clemson in the top 25, say goodbye to Maryland this weekend. 😉

 

[dullard]

Originally posted by: Dead Parrot Sketch
Where are Wisconsin and Penn State ranked, out of curiosity ?

Wisconsin (where my dad graduated) is ranked #50.
Penn State (where many of my cousins graduated and a recent rival of my school Nebraska) is ranked #42. Yes Penn St. has more losses, but their strength of schedule is far harder than Wisconsin's so far this year. Wisconsin is near the bottom of the long list of 1 loss teams - but they are just above some other of the better known 1 loss teams (Oregon, Texas Tech, and Missouri).

 

Dude, there's no way Michigan deserves to be 20, while Miami (OH) is in the top 10.

 

[dullard]

Originally posted by: JohnCU
YEAH BUDDY! Clemson in the top 25, say goodbye to Maryland this weekend. 😉

It is very early for the computer to make really good predictions, but so far it says Clemson by 7 over Maryland on a neutral field. I find the home team on average does about 2.5 to 3.5 points better than my computer predicts. So lets just assume Maryland gets a 3 point home field advantage. That makes its prediction:

Clemson wins by 4 over Maryland.

I'm sure it'll be off by quite a bit. But if both teams play average, that is the result. If Maryland gets hot or if some Clemson players get injured Maryland could easilly win this close game.

 

[dullard]

Originally posted by: jumpr
Dude, there's no way Michigan deserves to be 20, while Miami (OH) is in the top 10.

<- Dullard looks up at the top 10 he listed. Dullard gets confused.

 

[Tom]

I should have asked for Wisconsins and Penn States numbers too..😱

So what is the prognostication for mighty mighty Minnesota versus lowly Penn State this week ? (just interesting in testing how well it works) :light:

 

Originally posted by: dullard

Originally posted by: jumpr
Dude, there's no way Michigan deserves to be 20, while Miami (OH) is in the top 10.

<- Dullard looks up at the top 10 he listed. Dullard gets confused.

Oops, I meant top 15. 😛

 

[dullard]

Originally posted by: Dead Parrot Sketch
I should have asked for Wisconsins and Penn States numbers too..😱

So what is the prognostication for mighty mighty Minnesota versus lowly Penn State this week ? (just interesting in testing how well it works) :light:

Well I'm getting a lot more response for this ranking this year than I did the previous two years at ATOT. I might as well just list everyone instead of just the top 25.

It says Penn State over Wisconsin by 3.5 (I use fractions in the program but didn't bother to paste the fractions into ATOT). Add in a few points if you wish for the home field advantage. But I think Wisconsin has been playing better on the road the last few years (and Penn State's home crowd booing has hurt it at home). So I personally will keep it at 3.5 points for Penn State. Or possibly even subtract a couple for a home field disadvantage.

Here are the next 25 for anyone who feels the need to ask:

Louisville 127
Oklahoma St. 126
South Carolina 126
Washington 125
Virginia 125
Fresno St. 125
Florida 124
Boston College 124
Utah 124
Wake Forest 124
Boise St. 124
Michigan St. 123
Pittsburgh 123
North Carolina St. 123
Syracuse 123
Cincinnati 123
Penn St. 121
California 121
Iowa St. 121
UNLV 120
Maryland 119
Bowling Green 119
Stanford 119
Louisiana Tech 119
Wisconsin 118