View Full Version : CalPreps Computer Projections
Plaindriver
10-22-07, 05:01 PM
I was looking at the CalPreps computer predictions for last weekends games among the top 40 or so teams in the USA. Its downright uncanny just how close they came on the vast majority of games! Ck it out:
www.calpreps.com/2007/polls_view_scoreboard.htm
Perhaps they should use their match-up program to help their rankings a bit....
Plaindriver
10-23-07, 01:08 AM
Perhaps they should use their match-up program to help their rankings a bit....
Whats your beef with their rankings? Here is mine: Numbers 4, 8, 11, 12, 14, 16 and 23 (out of the top 25) are ranked too high. And the following are ranked too low: 24, 31, 32, 33, 36, 37, 42, and 45.
Pied, whats the skinny on Periman? Are they for real this yr, or have they had a steady diet of pup-cakes? And who might they see come playoff time?
#31 Northwestern is silly. I don't really care too much after that to be honest. It appears that the in state predictions seem to be close though.
Who knows.
Permian, the only time I saw them was against Trinity. They looked pretty salty in that game. Trinity traveled game day and it was early.
I think it's a good thing when traditional teams like Permian are good. I hope they make a long run.
skyway28
10-23-07, 09:49 AM
From a previous thread:
To consolidate and summarize, here is what we have right now(from another board):
More evidence is surfacing which illustrates how something in the programming of the calpreps computer arbitrarily diminishes Florida teams.
1) The three point margin SLC was favored over MNW has expanded to 6. (In reality, MNW won in Dallas 29-21)
2) Miami BTW is now projected to beat Summerville 28-19. (In reality, BTW won AT Summerville 36-3).
3) Byrnes is favored to win 31-14 over Dr. Phillips (In reality, Byrnes scratched out an 18-14 win).
4) Berkeley is favored by 26-14 over First Coast(In reality, FC won at Berkeley by a FG)
5) St. Xavier 2006 is favored to beat Lakeland 2006 by 10. (In reality, Lakeland beat X)
Something is terribly wrong when factors outside of head-to-head competition carry significantly more wieght than said head-to-head competetion. I don't know what the flaw in the programming is or how it happened to be in the program, i.e. programmer bias or just random flaws.
The only argument I see is that the predictions change throughout the year. Teams get better, teams get worse, some teams improve at different rates. Not impossible for me to see a team that loses early in the season be favored in a rematch later on.
That being said the facts listed above should at least provide an eyebrow to be raised, if not a little mocking....
skyway28
10-23-07, 09:59 AM
Well, the last week's results changed some things more:
1) SLC is a four point pick over MNW
2) BTW is now only a five-point pick over Summerville
3) Byrnes is now a 20-point pick over DP
4) Berkeley is still a 12-point pick over FC
It appears as if, as the season goes on, Florida's numbers go down. Maybe you'd put stock into this....if the aforementioned games hadn't actualy happenned!!!
From a previous thread:
To consolidate and summarize, here is what we have right now(from another board):
More evidence is surfacing which illustrates how something in the programming of the calpreps computer arbitrarily diminishes Florida teams.
1) The three point margin SLC was favored over MNW has expanded to 6. (In reality, MNW won in Dallas 29-21)
2) Miami BTW is now projected to beat Summerville 28-19. (In reality, BTW won AT Summerville 36-3).
3) Byrnes is favored to win 31-14 over Dr. Phillips (In reality, Byrnes scratched out an 18-14 win).
4) Berkeley is favored by 26-14 over First Coast(In reality, FC won at Berkeley by a FG)
5) St. Xavier 2006 is favored to beat Lakeland 2006 by 10. (In reality, Lakeland beat X)
Something is terribly wrong when factors outside of head-to-head competition carry significantly more wieght than said head-to-head competetion. I don't know what the flaw in the programming is or how it happened to be in the program, i.e. programmer bias or just random flaws.
You have the benefit of 20-20 hindsight. I'm wondering how often Calpreps gets it right? My guess is that they get it right most of the time.
Also, upsets in football happen all the time. Sometimes the better team loses. Even heavily favored teams aren't going to win every time but that doesn't mean they shouldn't still be favored (and even remain favored after they lose the game).
In addition, many teams improve significantly during the year. I believe the outcome of games played in week 1 or week 2 could be much different if the games were played in week 13 or week 14.
BTW, us SLC homers would LOVE to get another shot at MNW. That very young defense (most starting their 2nd varsity game) has progressed nicely and there is no way SLC has 5 TOs again. I know I'm a homer, but I'd pick SLC in a rematch. In my homer opinion, Calpreps has it right in this case.
You have the benefit of 20-20 hindsight. I'm wondering how often Calpreps gets it right? My guess is that they get it right most of the time.
Also, upsets in football happen all the time. Sometimes the better team loses. Even heavily favored teams aren't going to win every time but that doesn't mean they shouldn't still be favored (and even remain favored after they lose the game).
BTW, us SLC homers would LOVE to get another shot at MNW. That very young defense (most starting their 2nd varsity game) has progressed nicely and there is no way SLC has 5 TOs again. I know I'm a homer, but I'd pick SLC in a rematch. In my homer opinion, Calpreps has it right in this case.
OK, so what do they have right?
The prediction of the game, but the rankings should not be used yet as MNW is higher than SLC?
Gilmer > SLC?
Stephenville > Katy?
Please let us know what is right and what is wrong about the program. It seems obvious to you.
*Wait until later in the season for the rankings. SLC will probably shoot up, hmmmm.
*Use the predict a game because they are accurate. They predict SLC to beat MNW whom they lost to earlier this year.
Upon further consideration I don't need a computer to predict your response.....
OK, so what do they have right?
The prediction of the game, but the rankings should not be used yet as MNW is higher than SLC?
Gilmer > SLC?
Stephenville > Katy?
Please let us know what is right and what is wrong about the program. It seems obvious to you.
*Wait until later in the season for the rankings. SLC will probably shoot up, hmmmm.
*Use the predict a game because they are accurate. They predict SLC to beat MNW whom they lost to earlier this year.
Upon further consideration I don't need a computer to predict your response.....
Read my posts. Here are the points:
1. Any computer model works better with the most data possible. That's a fact. The human polls are better equipped to deal with less data. That's why the computer says MNW is #31 but the humans know they are arguably the best team in the nation. Simple.
2. Just because a losing team was favored before (or after) the game was played doesn't mean they shouldn't have been favored before (or after) the game. Sometimes the better team loses. Rather than finding 5 examples of where the computer got it wrong, wouldn't it be much better to look at a much larger sample and see how the computer did? Read Plaindriver's post that started this thread. He posted how accurate the Calpreps model was in predicting games.
Skyway stated a thread with this comment: "please don't cite any ratings-be it SOS or otherwise-from the calpreps computer as evidence of ANYTHING." The fact is, the Calpreps model does a good job overall but sometimes produces goofy results early in the season. That's no reason to trash the entire model.
-
skyway28
10-23-07, 11:32 AM
You have the benefit of 20-20 hindsight. I'm wondering how often Calpreps gets it right? My guess is that they get it right most of the time.
Also, upsets in football happen all the time. Sometimes the better team loses. Even heavily favored teams aren't going to win every time but that doesn't mean they shouldn't still be favored (and even remain favored after they lose the game).
In addition, many teams improve significantly during the year. I believe the outcome of games played in week 1 or week 2 could be much different if the games were played in week 13 or week 14.
BTW, us SLC homers would LOVE to get another shot at MNW. That very young defense (most starting their 2nd varsity game) has progressed nicely and there is no way SLC has 5 TOs again. I know I'm a homer, but I'd pick SLC in a rematch. In my homer opinion, Calpreps has it right in this case.
No, no no. In the case of ALL games listed by me above, calpreps also had the benefit of hindsight, at least in the sense that the games have already happened, yet calpreps is PRESENTLY projecting what they are. They have that same "hindsight".
Many teams do improve. But, that improvement, as it would pertain to these out-of-state matchups is a entirely subjective judgement, which we've been told does not enter into the computer. There is no way we can OBJECTIVELY measure SLC's rate of improvement relative to MNW's, for instance.
As you said, you are an SLC homer. MNW's defense and RB's were young as well.
skyway28
10-23-07, 11:42 AM
Read my posts. Here are the points:
1. Any computer model works better with the most data possible. That's a fact. The human polls are better equipped to deal with less data. That's why the computer says MNW is #31 but the humans know they are arguably the best team in the nation. Simple.
2. Just because a losing team was favored before (or after) the game was played doesn't mean they shouldn't have been favored before (or after) the game. Sometimes the better team loses. Rather than finding 5 examples of where the computer got it wrong, wouldn't it be much better to look at a much larger sample and see how the computer did? Read Plaindriver's post that started this thread. He posted how accurate the Calpreps model was in predicting games.
Skyway stated a thread with this comment: "please don't cite any ratings-be it SOS or otherwise-from the calpreps computer as evidence of ANYTHING." The fact is, the Calpreps model does a good job overall but sometimes produces goofy results early in the season. That's no reason to trash the entire model.
-
I think the way in which, as you say, human rankings are more equipped to deal with less "data" is telling. How are they more equipped?
Your second point is entirely subjective in nature. What in the game, or in the results since, would indicate the better team lost in MNW-SLC? Since we're dealing with a supposedly objective computer model, please include only objective observations in your reply.
In computer models, all of the data, by nature, is interrelated. That means the data used to generate this week's ratings is part of the whole that will generate the final ratings. You can't completely dismiss this week's ratings but then lend credence to the final product which includes the same type of data that led to this week's #s.
Read my posts. Here are the points:
1. Any computer model works better with the most data possible. That's a fact. The human polls are better equipped to deal with less data. That's why the computer says MNW is #31 but the humans know they are arguably the best team in the nation. Simple.-As do human polls, correct? Do you think the preseason or current AP college poll is more accurate? If you want to state that they are more accurate at the end of the season, what is the criteria for saying so? Is it that they more closely resemble the humna polls? If not, how can it be determined they are more less accurate?
2. Just because a losing team was favored before (or after) the game was played doesn't mean they shouldn't have been favored before (or after) the game. Sometimes the better team loses. Rather than finding 5 examples of where the computer got it wrong, wouldn't it be much better to look at a much larger sample and see how the computer did? Read Plaindriver's post that started this thread. He posted how accurate the Calpreps model was in predicting games.-I agree. The method seems to be fairly accurate, but the rankings associated with it does not seem to be in line. SLC are predicted to beat two of the TX teams ranked in the top 50. I agree teams may get better, but what has SLC done on defense to show the computers they have improved since MNW. The computers, I do not believe know when the starters are removed, they only know that Justin Northwest scored 24 on the Dragons. How does that help?
Skyway stated a thread with this comment: "please don't cite any ratings-be it SOS or otherwise-from the calpreps computer as evidence of ANYTHING." The fact is, the Calpreps model does a good job overall but sometimes produces goofy results early in the season. That's no reason to trash the entire model.
-
When do the results cease to be goofy?
RockinL
10-23-07, 12:44 PM
OK, so what do they have right?
The prediction of the game, but the rankings should not be used yet as MNW is higher than SLC?
Gilmer > SLC?
Stephenville > Katy?
Ive seen Gilmer and the only thing that would stop them from playing 4 quarters with SLC is depth off the bench. For a 3A team, they are very, very good.
skyway28
10-23-07, 12:55 PM
Ive seen Gilmer and the only thing that would stop them from playing 4 quarters with SLC is depth off the bench. For a 3A team, they are very, very good.
Nobosy has disputed your last sentence there. They are disputing the notion that Gilmer is flat-out better than SLC.
RockinL
10-23-07, 01:08 PM
ust out of curiosity....why isnt a team like Longview, Tx. ranked ahead of these teams that the computer says loses to Longview?
I dont know much about highschool football in Ne., or Mt., but I truly believe Longview would pretty much have their way with them.
I saw 2 teams from the Denver area the other day (Elways son was the QB for one), and I wouldnt see much of a problem beating the Denver area teams either.
I could be wrong.
at Longview (TX)
[2007] Longview (TX) 28, [2007] Kearney (NE) 14
at Kearney (NE)
[2007] Longview (TX) 28, [2007] Kearney (NE) 17
at neutral site
[2007] Longview (TX) 28, [2007] Kearney (NE) 14
at Longview (TX)
[2007] Longview (TX) 31, [2007] Helena (MT) 13
at Helena (MT)
[2007] Longview (TX) 31, [2007] Helena (MT) 14
at neutral site
[2007] Longview (TX) 31, [2007] Helena (MT) 14
at Longview (TX)
[2007] Longview (TX) 21, [2007] Columbine (Littleton, CO) 20
at Columbine (Littleton, CO)
[2007] Longview (TX) 21, [2007] Columbine (Littleton, CO) 20
at neutral site
[2007] Longview (TX) 21, [2007] Columbine (Littleton, CO) 20
No, no no. In the case of ALL games listed by me above, calpreps also had the benefit of hindsight, at least in the sense that the games have already happened, yet calpreps is PRESENTLY projecting what they are. They have that same "hindsight".
Not exactly. You can look at the exact game and you know the score. Your conclusion is based 100% on that individual game. Not only that, you are essentially assuming the score would be the same if the game was played again.
The Calpreps model works differently. It looks at the entire schedule of a team. It treats all games pretty much the same. It's prediction is not based on the fact that the game has already been played but is based on the current evauation of the teams. It is a prediction of the score if the game was played NOW, not in the past.
RockinL, here is some interesting info to me. Two teams whose records are identical 7-0. Simple numbers, who should be ranked higher?
Team 1
Opp ranking PTs for Pts against
780 71 28
1113 35 12
2545 39 0
6825 42 14
8851 63 6
10259 63 7
12680 59 7
Avg 6150 53 11
Team 2
Opp ranking Pts for Pts against
723 45 0
1553 41 14
1709 24 13
2129 48 7
3714 38 0
3872 48 0
7274 52 12
avg 2996 42 7
RockinL
10-23-07, 01:22 PM
RockinL, here is some interesting info to me. Two teams whose records are identical 7-0. Simple numbers, who should be ranked higher?
Team 1
Opp ranking PTs for Pts against
780 71 28
1113 35 12
2545 39 0
6825 42 14
8851 63 6
10259 63 7
12680 59 7
Avg 6150 53 11
Team 2
Opp ranking Pts for Pts against
723 45 0
1553 41 14
1709 24 13
2129 48 7
3714 38 0
3872 48 0
7274 52 12
avg 2996 42 7
I understand the point your making, and I think calpreps does an ok. job based ONLY on numbers....but thats also my point, and why I dont put much stock in it. Its intresting to discuss though.
How well do you think Helena, Mt. would do in District 12-5A?
RockinL
10-23-07, 01:25 PM
Nobosy has disputed your last sentence there. They are disputing the notion that Gilmer is flat-out better than SLC.
I think man for man they are as good. I understand what you are saying though.
I understand the point your making, and I think calpreps does an ok. job based ONLY on numbers....but thats also my point, and why I dont put much stock in it. Its intresting to discuss though.
How well do you think Helena, Mt. would do in District 12-5A?
Are they playing in MT or TX? Just kidding.
The two teams I listed were Gilmer and Katy. Gilmer gets 10 more points of offense on a schedule that is twice as bad and gives up four more points but is raked higher.
I agree w/SLCDad on this one, it appears kind of goofy.
I agree. The method seems to be fairly accurate, but the rankings associated with it does not seem to be in line. SLC are predicted to beat two of the TX teams ranked in the top 50. I agree teams may get better, but what has SLC done on defense to show the computers they have improved since MNW. The computers, I do not believe know when the starters are removed, they only know that Justin Northwest scored 24 on the Dragons. How does that help?
Your comment illustrates my point about computer rankings. The computers don't know that the SLC/Northwest game score was 55-10 at halftime and 65-10 in the 4th quarter. They don't know that Riley Dodge and Tre Newton were pulled in the 2nd quarter. They don't know that Northwest scored 2 TDs against the SLC 3rd teamers late in the 4th.
Obivously, the humans don't follow every quarter of every game (for the top teams) BUT the humans are capable of evaluationg much more than the small number of inputs used in the computer models. That's one reason why I say the humans can do a better job when fewer games have been played.
skyway28
10-23-07, 02:04 PM
Not exactly. You can look at the exact game and you know the score. Your conclusion is based 100% on that individual game. Not only that, you are essentially assuming the score would be the same if the game was played again.
The Calpreps model works differently. It looks at the entire schedule of a team. It treats all games pretty much the same. It's prediction is not based on the fact that the game has already been played but is based on the current evauation of the teams. It is a prediction of the score if the game was played NOW, not in the past.
The head-head results of a matchup should supersede other factors. This is especially true when you are talking about evalauating two teams who play so vastly differenet sets of opponents that there are no common opponents and no common opponents of opponents and on down the line.
What information could the computer objectively use to project an outcome different of what actually happenned a couple weeks earlier?
skyway28
10-23-07, 02:07 PM
Your comment illustrates my point about computer rankings. The computers don't know that the SLC/Northwest game score was 55-10 at halftime and 65-10 in the 4th quarter. They don't know that Riley Dodge and Tre Newton were pulled in the 2nd quarter. They don't know that Northwest scored 2 TDs against the SLC 3rd teamers late in the 4th.
Obivously, the humans don't follow every quarter of every game (for the top teams) BUT the humans are capable of evaluationg much more than the small number of inputs used in the computer models. That's one reason why I say the humans can do a better job when fewer games have been played.
Now, you seem to be getting it somewhat. But, what happens in future games won't chaneg the fact that things such as those in the Northwest game happenned. The results are still skewed.
Now, you seem to be getting it somewhat. But, what happens in future games won't chaneg the fact that things such as those in the Northwest game happenned. The results are still skewed.
Now I'm getting it? LOL Perhaps you are finally understanding what I've been saying all along.
Based on what you just posted I don't think you understand the head to head match-up predictions either. However, I'm not going attempt to explain it further.
The head-head results of a matchup should supersede other factors. This is especially true when you are talking about evalauating two teams who play so vastly differenet sets of opponents that there are no common opponents and no common opponents of opponents and on down the line.
What information could the computer objectively use to project an outcome different of what actually happenned a couple weeks earlier?
You really don't get it. As I said already, the computer uses the same criteria/method to predict all of the head to head match-ups. Even if the teams have already played each other, that game wouldn't carry any more weight than any other. By the way, that is the right way to do it in my opinion.
skyway28
10-24-07, 10:04 AM
Let's see. Two teams are evaluated, Team A (5A Texas) and Team B (5A Florida)-classification in parentheses. They have no common opponents, no common opponents of opponents, no common opponents of opponents of opponents. In short, there is no link between the two teams at all, until you get to to something oddball like Team A (5A )played Tony, TX (4A) who played Walla Walla, Tx (2A) which played LaLa, LA (3A) who in turn played Wohoo, LA (A) who played Smith, AL (A) who played Johnson, AL (A) who played Pensacola, FL (A)who played Lincan, FL (A) who played Edgeville, FL (A) who played Bartow, FL (A) who played Hardee, FL (A) who played Broward, FL (2A) who played Jackson, FL (4A) who played Team B (5A).
As such, the computer deems Tony better than Jackson due to some combination of results. so, Team A's win over Tony boosts them above Team B, which beat Jackson...which was not as good a win. So, Team A is rated better than Team B. In short, the results of predominantely Class A and 2A matchups in Louisiana, Alabama and the Florida panhandle are the overriding factor in how Team A and Team B are ultimately rated because that is where the bulk of the "link" exists. Never mind that Team B actually beat Team A head-to-head in a game both teams spent all off-season obsessing about. That result is merely one game in the mix. LOL.
vBulletin® v3.6.4, Copyright ©2000-2013, Jelsoft Enterprises Ltd.