Did my own ranking system... need tweaks

[Buckeyeskickbuttocks]

I don't think the NCAA website is good to use for SOS because I believe they just do it by win/loss record of opponents and I don't that that is the best way to measure SOS. What do you think of using a computer generated measure of SOS like Sagarin's rankings?

Agreed - that also will allow you to avoid the conflation of rank vs. SoS. Sagarin doesn't sugar-coat the SoS, he provides an explicit score - along with a simple ranking.

I would suggest using the Sagarin score - it provides a better measure of the true SoS than the ranknig, which implies that a schedule ranked 119 in the nation is well, worth 119th the value of the top-rated SoS.

For instance, the numerical difference between SoS for Florida and Ohio State is less than 2 units out of 74, rather than the NCAA SoS rank implying a 7 point difference on a 1-119 scale.

Putting it in layman's terms - don't conflate rank and score.

Good points.... when I get time, I'll put in Sagrins SOS ratings and see how that changes things.

 

[martinss01]

Also from the NCAA site, I obtained stat rankings for 27 different categories.

question, did you calculate these rankings per game and then factor in weight of just that game or did you simply take the averages for the season listed on the ncaa site and then factor in sos? the reason i bring this up is i don't think year averages provide an accurate estimate of a teams strength. if for example tOSU plays 3 games against Akron, YSU, and USC. the d gives up 98 yrds to Akron, 86 yrds to YSU and 460 yrds to USC. at the end of the 3 weeks they are giving up an average of 214 yrds per game. i suspect that the 214 yrds average will get far too much weight because USC is on the schedule. especially considering the yardage given up to USC.

i realize this is asking for ALOT more work and some of it may not even be realistic as i suspect this is at the heart of what is wrong with the bcs. but i get the feeling you are giving teams points they may not deserve.

excellent work though, seriously. i really like the approach you have taken.

edit: don't know how sagarin does their sos so you may already be addressing this point.

 

[ScarletInMyVeins]

Busy day at the office today BKB? Me too.

 

[Zurp]

Pretty cool. It's one of those things where you can never satisfy everyone. You may not be able to satisfy anyone who simply wants to poke holes in your results. You may not get much credibility when the only undefeated team in a BCS conference isn't in the top 10.

Excellent work. It looks like you've had a lot of fun doing this. If you aren't looking for suggestions, then feel free to quit reading now. But if you are looking for suggestions, I would say that you ought to try putting more emphasis on wins and losses, and who those wins and loses were against, rather than statistics. In my mind, an 8-4 team that averaged 550 yards on offense a game should be ranked lower than an 11-1 team that averaged 290 yards on offense. But, like I said, nice work. I'll check it out after this weekend's games and see if the rankings have been updated.

 

[Buckeyeskickbuttocks]

question, did you calculate these rankings per game and then factor in weight of just that game or did you simply take the averages for the season listed on the ncaa site and then factor in sos? the reason i bring this up is i don't think year averages provide an accurate estimate of a teams strength. if for example tOSU plays 3 games against Akron, YSU, and USC. the d gives up 98 yrds to Akron, 86 yrds to YSU and 460 yrds to USC. at the end of the 3 weeks they are giving up an average of 214 yrds per game. i suspect that the 214 yrds average will get far too much weight because USC is on the schedule. especially considering the yardage given up to USC.

i realize this is asking for ALOT more work and some of it may not even be realistic as i suspect this is at the heart of what is wrong with the bcs. but i get the feeling you are giving teams points they may not deserve.

excellent work though, seriously. i really like the approach you have taken.

edit: don't know how sagarin does their sos so you may already be addressing this point.

No, and I thought of the point you made recently as well... It's true that giving up 200 yards to Oklahoma is more impressive than giving up 100 to YSU, and my stats analysis does NOT factor that in.... Essentially I'm doing an "all things being equal" sort of approach.

Busy day at the office today BKB? Me too.

Did all the heavy lifting on this last night... Just had to format it today.

Pretty cool. It's one of those things where you can never satisfy everyone. You may not be able to satisfy anyone who simply wants to poke holes in your results. You may not get much credibility when the only undefeated team in a BCS conference isn't in the top 10.

Excellent work. It looks like you've had a lot of fun doing this. If you aren't looking for suggestions, then feel free to quit reading now. But if you are looking for suggestions, I would say that you ought to try putting more emphasis on wins and losses, and who those wins and loses were against, rather than statistics. In my mind, an 8-4 team that averaged 550 yards on offense a game should be ranked lower than an 11-1 team that averaged 290 yards on offense. But, like I said, nice work. I'll check it out after this weekend's games and see if the rankings have been updated.

For the moment, I've simply "normalized" wins and losses as they compare to statistical points. As I think more about it, I'll probably have to figure out what a win is worth, as compared to those stats... I'm thinking something like Wins and losses accouting for... I don't know... maybe 67 % of the total score, and the stats being 33%... but, that's also just an arbitrary selection....

Added Sagarins ratings... again, due to formatting I will not reproduce the whole 119 here.... but... It does look "better" I think...

A sample:
1....Texas
2....Oklahoma
3....Florida
4....Texas Tech
5....Southern California
6....Georgia
7....Penn St.
8....Utah
9....Alabama
10...Ohio St.
11...North Carolina
12...Oregon St.
13...Georgia Tech
14...Boston College
15...Boise St.
16...Ball St.
17...Virginia Tech
18...TCU
19...Oklahoma St.
20...Wake Forest
21...Michigan St.
22...Florida St.
23...Nebraska
24...Pittsburgh
25...Cincinnati

83. Michigan

100..Iowa St.
101..Florida Int'l
102..Indiana
103..Ohio
104..Utah St.
105..Army
106..Kent St.
107..UAB
108..Washington
109..Toledo
110..Eastern Mich.
111..La.-Monroe
112..Washington St.
113..San Diego St.
114..New Mexico St.
115..Tulane
116..Miami (Ohio)
117..SMU
118..Idaho
119..North Texas

Georgia at 6 seems too high for my liking. OSU at 10 seems better than 4, USC now ranked in the top 5

 

[BB73]

I'm not too keen on the idea of using statistical categories to rank teams. It would provide an incentive for teams to rack up yards, etc. in addition to just winning games, and could have a negative impact on the way games are played.

It's interesting stiff, but I'd really dislike any such system being used for BCS rankings.

 

[martinss01]

No, and I thought of the point you made recently as well... It's true that giving up 200 yards to Oklahoma is more impressive than giving up 100 to YSU, and my stats analysis does NOT factor that in.... Essentially I'm doing an "all things being equal" sort of approach.

well, when the bcs starts paying you billions to do their computer rankings remember to give me a job. and seeing as how you very likely are me you probably already have that covered

bama at #9 seems rather low expecially with georgia being so high. though it would be amusing to see the bama boards reaction should they see it. *snickers*

 

[LitlBuck]

As has already been mentioned that is quite a bit of work that you have done. Ron would be proud of you. You are either very good with Excel or an abacus:)

Please see my post in the BP Poll Week 14. I really don't think you can use statistical analysis to determine which teams are the best teams in the country. I really don't know how it can be done using any type of poll method either human or statistical. People complain about the BCS rankings but I really don't see any other way to do it. Perhaps you could get 10 college football experts then have them sit down every week and rank the teams as they see them. I don't know if you use writers, ex coaches, or the BP Recruiting Team but I just don't think it can be done any other way. I guess I was really helpful

 

[martinss01]

I would say that you ought to try putting more emphasis on wins and losses, and who those wins and loses were against, rather than statistics. In my mind, an 8-4 team that averaged 550 yards on offense a game should be ranked lower than an 11-1 team that averaged 290 yards on offense. But, like I said, nice work. I'll check it out after this weekend's games and see if the rankings have been updated.

hmmm... not sure i completely agree with that. W/L weight is obviously the most important statistic. but who the win or loss is to can be just as important. lets say team a plays texas, oklahoma, usc, florida and bama back to back going 3-2. they average 300yrds, give up 260, and looses by an average score of 4.5 points per game wining by an average of 11.5. now, lets say team b plays air force, army, navy, ntre ame, and washington going 5-0. they put up 500 yrs per game, gave up 120 and win by an average score of 30 points. could just be me, but i personally have a hard time thinking team b should get more points for its quality of work over those 5 games than team a.

this is my take. each team should be given a weight (a power rating as it were). while this rating would not be solely measured by its statistics they certainly need to be calculated into the formula. yrds allowed, points allowed, turnover margin, offense, special teams etc... these values obviously would fluctuate throughout the year and a team would gain or loose weight based not only upon their performance, but who their performance was against. in every game a team plays a certain number of points would be available based upon the weight of both teams. each "stat" in the formula would provide a team with either positive or negative points based upon performance. obviously W/L would be the highest. but performance comparatively to the other team would also be measured. the score available and each individual stat such as yrds allowed and score would provide either positive or negative points to the total value of your win.

sounds confusing i know but for example. tOSU vrs. USC. for USC tOSU is ranked #10 in the nation and for simplicity sake we will say it would be worth 109 points possible (max). 80% of those points are given if USC wins. however, they also get additional points for the various statistics in the game. score might be worth 5% of the total, turnovers 5%, yrds allowed 4%, yrds gained 4%, etc...

this would provide a measurable value to dominating a team but not overly reward a team for hanging 70 on a directional U.

for example if the full value of beating tOSU and getting full % points in every category is worth 110 points, doing the exact same thing against say duke would net a max of 15 points. so the potential value for points scored alone against tOSU might be 1-4 points in value where the value against duke would be .2 - 1.4. in this scenario beating tOSU by 10 might actually be worth more than hanging 80 on duke. in fact, scoring that many points on anyone would have little value as there is a point cap for each category. once you have scored enough points to get the max allowable points, no matter how many more you score, the # of points earned for that statistic is unchanged. so while beating duke by 3 might net florida .2 points for that stat. beating duke by 20 or 120 would give the same 1.4 points.

I'm not too keen on the idea of using statistical categories to rank teams. It would provide an incentive for teams to rack up yards, etc. in addition to just winning games, and could have a negative impact on the way games are played.

It's interesting stiff, but I'd really dislike any such system being used for BCS rankings.

but didn't that happen before the bcs though? in fact, aren't the voters far more swayed by huge point totals and offensive statistics than computers? could be me, but all i hear this time of year is how teams are trying to put up sexy wins for the "voters". i never hear anyone say they are trying to impress the computers. could be that they don't understand how the computers work. but i suspect you have a better chance of limiting the value placed on pounding a weak school into oblivion with a mathematical formula than with a person who only catches maybe 3% of the games and relies on sportscenter highlights to determine how they rank the top 25.

 

[Buckeyeskickbuttocks]

I'm not too keen on the idea of using statistical categories to rank teams. It would provide an incentive for teams to rack up yards, etc. in addition to just winning games, and could have a negative impact on the way games are played.

It's interesting stiff, but I'd really dislike any such system being used for BCS rankings.

I don't mind statistical output. Because it includes both sides of the ball as well as special teams (which includes punting, PR defense, KO Returns, KO D and such) I think things tend to normalize a little bit. Likewise, if you're very good at running, but suck at passing... you don't really gain anything. At some point, all things being equal... they remain equal... so Harrell throws for 600 instead of 500.... The TTU passing O doesn't really gain anything... they were 1 at 500 yards.. they're still 1 at 600.... The margin of their particular dominance doesn't count (At least as this ratings formula stands right now)

Eh.. I understand what you're saying, but I'm not sure any tendency to run it up would be encouraged (Nor would it be discouraged). However, superior/dominant play would be encouraged in all phases (though, that should be true even if no one kept any stats) (I should note, those phases also include things like 3rd Down rate (O and D) ./.. so it's not just like I look at a team that runs 2 plays for 40 yards each, call them the best, and end the discussion... a ball control team has equal opportunity to gain "stats points" by doing what they do best... converting short yardage)

 

[BUCKYLE]

tOSU still isn't #1. Keep tweaking.

 

[BB73]

but didn't that happen before the bcs though? in fact, aren't the voters far more swayed by huge point totals and offensive statistics than computers? could be me, but all i hear this time of year is how teams are trying to put up sexy wins for the "voters". i never hear anyone say they are trying to impress the computers. could be that they don't understand how the computers work. but i suspect you have a better chance of limiting the value placed on pounding a weak school into oblivion with a mathematical formula than with a person who only catches maybe 3% of the games and relies on sportscenter highlights to determine how they rank the top 25.

Per BCS rules, a few years ago the 6 BCS computer systems had to remove margin of victory from their formulas. That's why big scores don't impress the computers. It's all who a team played, and whether they won or lost, and the timing of a loss (early versus late) doesn't matter to the computers. Also, if a team beats a top-5 team early in the year, but that team finishes at 6-6, the computers don't care that the team was ranked when the game was played.

By the way, a Harris poll voter has been quoted telling a reporter at the Oklahoma-Okie State game last Saturday that he thought Penn State would play Alabama in the title game, since they were the only undefeated teams. That guy seriously needs to be removed from the Harris voting next year.

 

[BB73]

I don't mind statistical output. Because it includes both sides of the ball as well as special teams (which includes punting, PR defense, KO Returns, KO D and such) I think things tend to normalize a little bit. Likewise, if you're very good at running, but suck at passing... you don't really gain anything. At some point, all things being equal... they remain equal... so Harrell throws for 600 instead of 500.... The TTU passing O doesn't really gain anything... they were 1 at 500 yards.. they're still 1 at 600.... The margin of their particular dominance doesn't count (At least as this ratings formula stands right now)

Eh.. I understand what you're saying, but I'm not sure any tendency to run it up would be encouraged (Nor would it be discouraged). However, superior/dominant play would be encouraged in all phases (though, that should be true even if no one kept any stats) (I should note, those phases also include things like 3rd Down rate (O and D) ./.. so it's not just like I look at a team that runs 2 plays for 40 yards each, call them the best, and end the discussion... a ball control team has equal opportunity to gain "stats points" by doing what they do best... converting short yardage)

Here's an important argument against it.

Tressel-ball makes tOSU look worse than the team really is in many offensive categories, so using statistics to rank teams would negatively impact tOSU.

 

[WaitingforKickoff]

Great work Bbk, give yourself a GPA. I notice that you use the SOS as the key criteria in weighing the rankings. What happens if you use a combination of SOS, defensive rankings and offensive rankings? I'm just curious because I think a team's defensive ranking is probably the most accurate indicator of how high a team should be ranked, ie most of the time, the teams with the highest defensive rankings are at the top of the polls, which is not always the case with offensive output or SOS.

 

[Buckeyeskickbuttocks]

Here's an important argument against it.

Tressel-ball makes tOSU look worse than the team really is in many offensive categories, so using statistics to rank teams would negatively impact tOSU.

True, but I'm not sure that the results would be noticed too much. 27 Categories are quite a lot, and some of Tressel-ball's best are captured as well.... like... net punting It would probably be better if DBB was throwing this together with his differential stats and such, as I do think those stats are much more telling than the straight stats I've been trying to use...

At this point, I suppose I oughta list out the categories I used.... you can then determine for yourself if the array is vast enough to capture more than just throwing up big offensive numbers, or leaving your 1st Team D in against a team that can't move the ball....

No particular order, and some stats might actually capture the same information (Passing and Passing efficiency and turnover margin to varying degrees, for example)

Rush Offense
Passing Offense
Total Offense
Passing Efficiency Offense
Points
Rush Defense
Pass Defense
Passing Efficiency Defense
Total Defense
Points against
Turnover Margin
Fewest Penalties
Punt Return yards
Punt Return D
Kick Off Returns
Kick Off Return D
Turnovers Gained
Turnovers Lost
Sacks
Sacks Allowed
Tackles For Loss
Tackles For Loss Allowed
3d Down Conversion (Offense)
3rd Down Conversion (Defense)
Time of Possession
Red Zone Offense
Red Zone Defense

Essentially everything the NCAA site had (there were a few I ignored... like 4th Down O and D, for example)