Serious computer index question

[Lehigh Football Nation]

My beef with Ralph's GPI isn't so much that it is an index but that by its nature it's an index that never really gets the human element right completely. I don't feel comfortable with an index that is by its nature more tilted towards the computers.

Do you think there's value in coming up with an index that is constructed in the following way:

Total = (ARC/6 + ARP/3)/2

ARC (Adjusted Rankings of the Computers) = Sum of 6 computer ranking numbers (removing the max and min number from the 8 used)

ARP (Average Rankings of Polls) = Sum of 3 polls: TSN, Coaches', AGS

Ralph's GPI model uses MAS = Massey (http://www.masseyratings.com/rate.php?lg=cf), ASH = Ashburn (http://www.atomicfootball.com/), SAG = Sagarin (http://usatoday30.usatoday.com/sports/sagarin/fbt12.htm), LAZ = Laz Index (http://www.lazindex.com/), KEE = Keeper (http://home.comcast.net/~keepersfootball/site/), SEL = Self (https://github.com/jeffself/collegefootballrankings), BRN = Born (http://www.bornpowerindex.com/). I would definitely add the SRS to this mix and consider using a different computer models.

I'd also consider other polls (College Sports Madness?).

Thoughts? Is this a good idea, making human and computer a true 50/50 in an index and adding SRS to the computer models? Or would this be a colossal waste of time?

[Pard4Life]

Aw LFN, no need to be upset that the computers treat Lehigh much more poorly than the human polls. Computers are blind to bias. :D

[Lehigh Football Nation]

Computers are blind

+1

[dystopiamembrane]

The human element is the issue here, in my opinion.

[Lehigh Football Nation]

The human element is the issue here, in my opinion.

Go on... Not enough?

[MTfan4life]

If you're going to be using the Coaches poll, I think you are also going to have to include the Illinois State fraternity poll, and the poll of the employees of the Alaska Department of Transportation. All three pay about the same amount of attention, so if you're going to include one, you're going to need to include them all.

[dbackjon]

If you're going to be using the Coaches poll, I think you are also going to have to include the Illinois State fraternity poll, and the poll of the employees of the Alaska Department of Transportation. All three pay about the same amount of attention, so if you're going to include one, you're going to need to include them all.


And should include the CS.com poll, as that is better than at least two of the polls, with pollsters more knowledgable about the game than the coaches. Without the CS poll, GPI remains flawed.

[Catbooster]

If you're going to be using the Coaches poll, I think you are also going to have to include the Illinois State fraternity poll, and the poll of the employees of the Alaska Department of Transportation. All three pay about the same amount of attention, so if you're going to include one, you're going to need to include them all.
xlolx xbowx xnodx

[lionsrking2]

Humans are biased and computers lack common sense... Other than that it's a perfect system.

[danefan]

Playoffs. **** yeah



no one can complain. Every league has an AQ now that wants one. That makes polls irrelevant. Hence my lack of f Ralph posts in the GPI thread.

[Pard4Life]

Taking the average of two averages is not very sound. Why not just include the human polls side by side with the computer polls and divide by 9? It would be weighted correctly in that manner...

[thebootfitter]

Why is the average of two averages not sound? Depends on what it is you are trying to communicate with the result, I suppose. If you want to equally weight the collective "human" ranking component against a collective ofsimulation/computer based models, the average of averages should be fine.

[Lehigh Football Nation]

Why is the average of two averages not sound? Depends on what it is you are trying to communicate with the result, I suppose. If you want to equally weight the collective "human" ranking component against a collective ofsimulation/computer based models, the average of averages should be fine.

This was my thought process.

[thebootfitter]

Why not include enough human polls to justify throwing out the min and max from those as well? CS? AGS? What others are there to pick from?

The more I think about this, the more I like it. It also, I think, offers an easy way to compare the collective human polls against the computers. It would be fun to track over time how computers perform vs humans after the playoffs happen each year.

[Wallace]

My beef with Ralph's GPI isn't so much that it is an index but that by its nature it's an index that never really gets the human element right completely. I don't feel comfortable with an index that is by its nature more tilted towards the computers....

You personalizing this at me reflects your character. Mike Jenkins created the GPI over 13 years ago, I had nothing to do with it. The GPI advisor is Kenneth Massey who has a lot more experience and knowledge than anyone here on this subject and he states that the GPI is fine. His only beef is that the polls are out of whack and not enough computer ratings are included.

Go ahead and copy the GPI for your business and hack at CSN, then maybe you can make even more money off the FCS. xnonono2x


The human element is the issue here, in my opinion.

THIS!

[MR. CHICKEN]

The human element is the issue here, in my opinion.

18403......IFIN' DATA....IS ANYTHIN'....LIKE SOME AGS.....POLL PICKERAHS...POLLS.......AH'D FACTOR IN.....ZOO GORILLAS/FARM CHICKENS......O'....JES' SPIN DUH BOTTLE..............xintx........BRAWK!

[UAalum72]

Why is the average of two averages not sound? Depends on what it is you are trying to communicate with the result, I suppose. If you want to equally weight the collective "human" ranking component against a collective ofsimulation/computer based models, the average of averages should be fine.Because the unspoken assumption of an index is that the strength of one type reduces the weaknesses of the other, but ignores the reverse. Whichever you think is better (human or computer), adding the other to it regresses the result away from the better.

[dystopiamembrane]

18403......IFIN' DATA....IS ANYTHIN'....LIKE SOME AGS.....POLL PICKERAHS...POLLS.......AH'D FACTOR IN.....ZOO GORILLAS/FARM CHICKENS......O'....JES' SPIN DUH BOTTLE..............xintx........BRAWK!
Would someone please translate?

[RichH2]

Because the unspoken assumption of an index is that the strength of one type reduces the weaknesses of the other, but ignores the reverse. Whichever you think is better (human or computer), adding the other to it regresses the result away from the better.

Well said ,both aspects of this issue are based on value judgments by humans, Does more input mandate a more accurate system?

[RichH2]

Would someone please translate?

xlolx Been enjoying his posts for yrs. I would guess that he is opining that all of these ratings are about as accurate as spinning the bottle.
Perhaps his view s/b added to any new system to reflect the inherent skepticism of the validity of ratings;)

[Lehigh Football Nation]

You personalizing this at me reflects your character. Mike Jenkins created the GPI over 13 years ago, I had nothing to do with it. The GPI advisor is Kenneth Massey who has a lot more experience and knowledge than anyone here on this subject and he states that the GPI is fine. His only beef is that the polls are out of whack and not enough computer ratings are included.

Go ahead and copy the GPI for your business and hack at CSN, then maybe you can make even more money off the FCS. xnonono2x



THIS!

There are two GPIs.

[MR. CHICKEN]

Because the unspoken assumption of an index is that the strength of one type reduces the weaknesses of the other, but ignores the reverse. Whichever you think is better (human or computer), adding the other to it regresses the result away from the better.

18406......OWN-LAH....SO MUCH WAWA...YOU CAN ADD...TA GRAVY.....B/4...IT NEEDS...MO' FLOUR.........BRAWK!

[darell1976]

If you're going to be using the Coaches poll, I think you are also going to have to include the Illinois State fraternity poll, and the poll of the employees of the Alaska Department of Transportation. All three pay about the same amount of attention, so if you're going to include one, you're going to need to include them all.

http://i5.photobucket.com/albums/y178/darell1976/SP.png (http://s5.photobucket.com/user/darell1976/media/SP.png.html)

[Lehigh Football Nation]

Why not include enough human polls to justify throwing out the min and max from those as well? CS? AGS? What others are there to pick from?

The more I think about this, the more I like it. It also, I think, offers an easy way to compare the collective human polls against the computers. It would be fun to track over time how computers perform vs humans after the playoffs happen each year.

The trouble is that there are not enough human polls to justify taking out the min and max. You'd need at least 5, in my estimation. You'd have TSN, the Coaches', AGS, and say CS.com. You'd need at least 1 more, and the more, the better.

There's also the issue that the same people vote in different polls.

But overall I just think that pure, human ballots and computer models should be weighted the same in an index, 50/50. And I believe that going through this exercise might be interesting.

Maybe you do the computations and CSN's GPI and this index end up the same... I have no idea. But I think some of the worse things I've seen from the GPI in recent years, like teams with sub-.500 records in the Top 25, might be alleviated somewhat.

[Wallace]

The trouble is that there are not enough human polls to justify taking out the min and max. You'd need at least 5, in my estimation... But I think some of the worse things I've seen from the GPI in recent years, like teams with sub-.500 records in the Top 25, might be alleviated somewhat.

There's the issue, you think you know better than experts and you have no basis for it. BTW, the "teams with sub-.500 records in the Top 25" are in the polls, not the computers in the GPI.


There are two GPIs .

No, there isn't. Stick to Lehigh adulation. xcoffeex

[Lehigh Football Nation]

There's the issue, you think you know better than experts and you have no basis for it. BTW, the "teams with sub-.500 records in the Top 25" are in the polls, not the computers in the GPI.

Last season UNI was in the Top 25 of the GPI with a sub-.500 record.

[Wallace]

omg, cherry picking much? Look at the current rankings. I think you have a problem with the GPI (besides you wanting to make more money in your business by hacking at CSN) because your everlovin' Lehigh has such a bad schedule every year that hinders their rating. Can't help that you know? Just play good teams and win.

[Lehigh Football Nation]

omg, cherry picking much? Look at the current rankings. I think you have a problem with the GPI because your everlovin' Lehigh has such a bad schedule every year that hinders their rating. Can't help that you know? Just play good teams and win.

Nice, I prove your statement wrong, then you accuse me of cherrypicking.

Computer models assume an FBS team is a good team even if they're terrible.

I'll assume you also have computer models that predict in March that teams on your schedule like perennial playoff participant New Hampshire are going to fall precipitously to No. 50. Please, by all means, let Lehigh's AD know so they can schedule accordingly.

[walliver]

The trouble is that there are not enough human polls to justify taking out the min and max. You'd need at least 5, in my estimation. You'd have TSN, the Coaches', AGS, and say CS.com. You'd need at least 1 more, and the more, the better.

There's also the issue that the same people vote in different polls.

But overall I just think that pure, human ballots and computer models should be weighted the same in an index, 50/50. And I believe that going through this exercise might be interesting.

Maybe you do the computations and CSN's GPI and this index end up the same... I have no idea. But I think some of the worse things I've seen from the GPI in recent years, like teams with sub-.500 records in the Top 25, might be alleviated somewhat.

I think the real issue is the lack of useful human polls at the FCS level.

For the big boys, reporters and AD's can watch a plethora of games via DVR, highlights via ESPN and endless discussions on Sports Center. For the Big Boys, their top teams are on TV almost every week. In the Coaches poll, the coaches may sign off on their lists, but these guys are too busy coaching to make up an accurate list and usually let someone else (like the AD) fill out the list.

At the FCS level, there are limited telecasts and frequently the games are picked for political reasons (everybody gets a TV game) not because they are good matchups. Many of the telecasts are regional, not national. There is no FCS highlights show. Most beat writers have little knowledge of other teams in their own conferences, much less the national "big picture". I'll admit that most of my FCS knowledge is limited to the SoCon and the Carolinas' Big South teams. Yes, I have heard the south canadians have a good team, but have not seen a game, or significant highlights, involving NDSU, SHSU, UNI, Towson, etc., and most of what I know of them comes from AGS, and we know that all AGS sources are free of bias. I suspect a lot of voters in many of the polls have less national knowledge than I.

Most national FCS polls are garbage. And when you put garbage into a calculation, you get a garbage result.

I don't particularly worry about these polls, since, at the end of the year, there is usually a consensus about 90+% of the playoff field, and the last 1 or 2 spots can often be guessed at by looking at the committee's historical biases (with a WTF thrown in every year or two).

[ysubigred]

http://i5.photobucket.com/albums/y178/darell1976/SP.png (http://s5.photobucket.com/user/darell1976/media/SP.png.html)

I'd hit it ^^ xlovex

[Hammerhead]

You can't trust The Sports Network poll if a 2-3 North Dakota team gets 20 votes. xspankx

http://www.sportsnetwork.com/merge/tsnform.aspx?c=sportsnetwork&page=cfoot2/misc/cfoot225poll.aspx

[dystopiamembrane]

Computer models assume an FBS team is a good team even if they're terrible.
If a computer model is set up in this fashion, the model itself is terrible.

[darell1976]

You can't trust The Sports Network poll if a 2-3 North Dakota team gets 20 votes. xspankx

http://www.sportsnetwork.com/merge/tsnform.aspx?c=sportsnetwork&page=cfoot2/misc/cfoot225poll.aspx

Of course you can ISU was a tough win. Ok seriously I think it's our SOS since we have already played 3 top 10 teams in a row and soon to be a 4th on Saturday.

[Lehigh Football Nation]

If a computer model is set up in this fashion, the model itself is terrible.

My assertion is that since there is not enough of a body of work in a given year, it is impossible to determine how "strong" FCS vs. FBS games are in any given year. In an RPI situation in basketball, there are lots of OOC data points to compare, so this isn't really an issue, but an FCS team has 0, 1 (most common) 2, (sometimes) or 3 (rarely) competitions against FBS teams.

IMO, many computer models try to get around this weakness in a variety of ways, but ultimately computer models (notably Sagarin) tend to overweight FBS wins in general. Ironically, the only way I can think to counteract this lack of data is to use some amount of human polls - human polls at least have the potential of having the "eye test" to know that a win over Oregon State is a fabulous win, and a win over Georgia State, not so much.

A picture-perfect example of this can be demonstrated in Sagarin (http://www.usatoday.com/sports/ncaaf/sagarin/) this week.

77 McNeese State AA = 67.71 5 1 55.43( 161) 0 0 | 0 0 | 68.73 72 | 66.52 84
95 Sam Houston State AA = 63.86 4 1 32.57( 251) 0 1 | 0 1 | 67.79 76 | 60.00 109

Before we get started, I like both of these teams - they've obviously both earned Top 25 rankings.

But most humans look at SHSU's loss to Texas A&M, and McNeese State's 41-6 loss to Northern Iowa and make a different choice as to which team is better.

More importantly, look at "schedule strength". McNeese State's SOS is twenty points higher!

Compare the schedules:

McNeese: USF (FBS, W), West Alabama (D-II), UAPB, Weber State, UNI (L)
SHSU: TAMU (FBS, L), Houston Baptist, Incarnate Word, Texas Southern, Eastern Washington (W)

With a human eye, essentially, they have the same type of schedule: a SWAC, one sub-D-I program (Houston Baptist, West Alabama), one bad FCS program (Weber State, Incarnate Word), one FCS Top 25 (EWU, UNI) and one FBS.

This is why I feel like the models are screwed up.

On the schedule side, Sag overweights McNeese's win over USF, discount their loss to UNI, overweight SHSU's loss to TAMU, and underweight significantly their win over Eastern Washington.

[Wallace]

Just Ralph being Ralph,very terratorial

Yeah, I like good and factual FCS information. RichH2, inspect your spellchecker :D

[Wallace]

My assertion is that since there is not enough of a body of work in a given year, it is impossible to determine... A picture-perfect example of this can be demonstrated in Sagarin (http://www.usatoday.com/sports/ncaaf/sagarin/)...

Chuck, how often will you continue to display your lack of understanding in this matter? All of the experts state there is enough points to determine a conclusion but YOU say there is not! What are your credentials to dispute what experts state? Sagarin is well known as an inaccurate measure of the FCS due to his formula mis-measure of non-D-I opponents. Chat about Lehigh and light-heartedly about the FCS in general but do not bite off more than you can chew. Just friendly advise.

[RichH2]

Yeah, I like good and factual FCS information. RichH2, inspect your spellchecker :D


Senior moment,typing on phone

[Wallace]

Senior moment,typing on phone

lol, I recently got one of those really big phones so I can type better on it. ASUS fonepad. :)

18408

[RichH2]

lol, I recently got one of those really big phones so I can type better on it. ASUS fonepad. :)

LOl ,we've been around too long old friend. My kids got me the phone as payback I think.

[caribbeanhen]

Humans are biased and computers lack common sense... Other than that it's a perfect system.
people lack common sense but some Humans can program computers to be biased...... playoffs baby

[caribbeanhen]

Last season UNI was in the Top 25 of the GPI with a sub-.500 record.

makes perfect sense of only 24 teams were better than them

[Wallace]

people lack common sense but some Humans can program computers to be biased...... playoffs baby

my experience is biased computer systems do not have legs.

[Lehigh Football Nation]

Chuck, how often will you continue to display your lack of understanding in this matter? All of the experts state there is enough points to determine a conclusion but YOU say there is not! What are your credentials to dispute what experts state? Sagarin is well known as an inaccurate measure of the FCS due to his formula mis-measure of non-D-I opponents. Chat about Lehigh and light-heartedly about the FCS in general but do not bite off more than you can chew. Just friendly advise.

1. If Sag is such an inaccurate measure of the FCS, how come it's included in the GPI?

2. If it's "mis-measuring non-D-I opponents", isn't that for a reason? What is it?

3. By saying that he "mis-measures non-D-I opponents", how far of a leap of faith is it to think that maybe it's overweighting FBS opponents?

4. Are experts always right?

[GoAgs72]

What is going on here?

They have some personal issues - best to stay out of it.

[Mr. C]

Would someone please translate?
What, you don't understand Chickenspeak? You just be a newby to AGS Land.

[HensRock]

If you're going to be using the Coaches poll, I think you are also going to have to include the Illinois State fraternity poll, and the poll of the employees of the Alaska Department of Transportation. All three pay about the same amount of attention, so if you're going to include one, you're going to need to include them all.

This ^ is exactly why more weight should be given to the computer polls. The Coaches Poll is a joke.
At least teh computers are paying attention. And yes, most of them rank Lehigh low - along with Gardner Webb, Charleston Southern and a few other teams that have snuck into the human polls.

[OhioHen]

Yeah, I like good and factual FCS information. RichH2, inspect your spellchecker :D


Chuck, how often will you continue to display your lack of understanding in this matter? All of the experts state there is enough points to determine a conclusion but YOU say there is not! What are your credentials to dispute what experts state? Sagarin is well known as an inaccurate measure of the FCS due to his formula mis-measure of non-D-I opponents. Chat about Lehigh and light-heartedly about the FCS in general but do not bite off more than you can chew. Just friendly advise.

Just have to :D xlolx xsmiley_wixat these two consecutive posts in the thread. First you tell RichH2 to inspect his spellchecker, then you forget to use your grammar check at least twice in the very next post (I will concede that the second grammatical error in a single sentence may or may not count given that the usage is consistent within the sentence).

[Wallace]

Just have to :D xlolx xsmiley_wixat these two consecutive posts in the thread. First you tell RichH2 to inspect his spellchecker, then you forget to use your grammar check...
GUILTY! hahaha