Northwestern falls out of calpreps top 25

[skyway28]

Checking the calpreps power ratings, which will be released on the maxpreps site tommorow, MNW has dropped out of the top 25 down to #34. The Bulls are now behind the likes of Gates-Chili Of Rochester; Eden Prairie (MN);Marion Local of Maria Stein, OH; Arrowhead of Hartland, WI and many other gems.

They say any publicity is good publicity. Hopefully, if the publicity so clearly illustrates the flaw in something, it is not good publicity in the end. Once again, please don't cite any ratings-be it SOS or otherwise-from the calpreps computer as evidence of ANYTHING.

 

[SLCDad]

Checking the calpreps power ratings, which will be released on the maxpreps site tommorow, MNW has dropped out of the top 25 down to #34. The Bulls are now behind the likes of Gates-Chili Of Rochester; Eden Prairie (MN);Marion Local of Maria Stein, OH; Arrowhead of Hartland, WI and many other gems.

They say any publicity is good publicity. Hopefully, if the publicity so clearly illustrates the flaw in something, it is not good publicity in the end. Once again, please don't cite any ratings-be it SOS or otherwise-from the calpreps computer as evidence of ANYTHING.

In my opinion, Calpreps is producing some really goofy results at this point this year. Historically it's output has been pretty good. I'm thinking the Calpreps model will get better as the season progresses and more games are played. In Texas we've only played 7 games of a 16 game (potential) season. Computer models are usually better when they have the most data possible.

 

[skyway28]

1) Why would they produce "better" results in previous years than this year? Did the program suddenly become of less quality? LOL

2) You shouldn't need more results than we have now to avoid a result where among other oddities, a Division V school (Marion Local) not considered even in the same of X and Colerain etc -see my thread on the Ohio football board on this site- is ranked ahead of the likes of SLC and MNW. That should NEVER happen.

 

[SLCDad]

1) Why would they produce "better" results in previous years than this year? Did the program suddenly become of less quality? LOL

2) You shouldn't need more results than we have now to avoid a result where among other oddities, a Division V school (Marion Local) not considered even in the same of X and Colerain etc -see my thread on the Ohio football board on this site- is ranked ahead of the likes of SLC and MNW. That should NEVER happen.

You laugh because you it appears you don't know how the model works.

Pretty much every computer model like this produces "better" results as teams play more games. For example, for the first 4-5 weeks of the season the Calpreps model relies on data from the prior year. After that, the prior year's results are dropped. As more games are input into the model the rankings get better.

Also, some team's haven't played their toughest opponents yet so their strength of schedule isn't high enough to give them a high ranking (even though the team should be ranked high). SLC is a good example of this. SLC's district opponents so far have been cream puffs. As a result, SLC hasn't had the opportunity to play a great team except for MNW. No matter how good SLC really is, the model won't respect them until they play some better teams. SLC will have plenty of games against great teams if they advance deep in the playoffs. That's why you need the entire season to get the best computer ranking output.

When we look at prior years we are looking at complete seasons. Those rankings are MUCH BETTER than the current 2007 ranking which has less that 1/2 of the season included. Most of the highly ranked teams will play 13, 14, 15 or 16 games. We now only have 6-8 of those games included.

Do you now understand how prior year's rankings can be "better" than the current 2007 ranking?

 

[pied]

Pretty much every computer model like this produces "better" results as teams play more games.

My predictions get better as the games are played as well. I am 100% now in predicting Super Bowls 1-42. Right now I'd say NE over Dallas. Mid Feb I'll have my exact score prediction....

 

[skyway28]

You laugh because you it appears you don't know how the model works.

Pretty much every computer model like this produces "better" results as teams play more games. For example, for the first 4-5 weeks of the season the Calpreps model relies on data from the prior year. After that, the prior year's results are dropped. As more games are input into the model the rankings get better.

Also, some team's haven't played their toughest opponents yet so their strength of schedule isn't high enough to give them a high ranking (even though the team should be ranked high). SLC is a good example of this. SLC's district opponents so far have been cream puffs. As a result, SLC hasn't had the opportunity to play a great team except for MNW. No matter how good SLC really is, the model won't respect them until they play some better teams. SLC will have plenty of games against great teams if they advance deep in the playoffs. That's why you need the entire season to get the best computer ranking output.

When we look at prior years we are looking at complete seasons. Those rankings are MUCH BETTER than the current 2007 ranking which has less that 1/2 of the season included. Most of the highly ranked teams will play 13, 14, 15 or 16 games. We now only have 6-8 of those games included.

Do you now understand how prior year's rankings can be "better" than the current 2007 ranking?

Exactly, pied. lol

Fact is, Marion Local may well blow through Division V and rack up big wins in the process. And they won't have common opponents with Iggy etc. So, they may well maintain their ranking. They are on a different level all together and should NEVER-at any point in the season-be ranked ahead of an Iggy.

And what about the project-a-matchup results I posted for seasons completed? Like 2006 X being projected a 10 point winner over 2006 Lakeland?

 

[SLCDad]

As the season progresses the computer models will become more accurate. The cream of the crop will rise to the top even though there are very few games between the top teams.

It's somewhat foolish to mock a computer model at this point of the season because the amount of data included is small.

 

[WoodyHayes]

Am I Reading It Wrong

Checking the calpreps power ratings, which will be released on the maxpreps site tommorow, MNW has dropped out of the top 25 down to #34. The Bulls are now behind the likes of Gates-Chili Of Rochester; Eden Prairie (MN);Marion Local of Maria Stein, OH; Arrowhead of Hartland, WI and many other gems.

They say any publicity is good publicity. Hopefully, if the publicity so clearly illustrates the flaw in something, it is not good publicity in the end. Once again, please don't cite any ratings-be it SOS or otherwise-from the calpreps computer as evidence of ANYTHING.

I pulled it up and I dont see how they could drop to 34. If they drop to 34 Carroll has to drop to at least 40.

 

[SLCDad]

I pulled it up and I dont see how they could drop to 34. If they drop to 34 Carroll has to drop to at least 40.

The Freeman model (Calpreps) says MNW is 31st. EVERY human ranking except one had MNW #1 last week.

The Freeman model says SLC is #73. EVERY human ranking puts SLC from 3rd to 7th in the nation.


If we only use the human rankings the Top 5 national teams are (last week):

1. Northwestern
2. St Xavier
3. Katy
4. Southlake Carroll (tie)
4. Mater Dei (tie)

I'm not saying we should reject the computer rankings but they are producing some goofy results at the present time.

 

[pied]

As the season progresses the computer models will become more accurate. The cream of the crop will rise to the top even though there are very few games between the top teams.

It's somewhat foolish to mock a computer model at this point of the season because the amount of data included is small.

Just curious, is it mocking to call the results thus far "really goofy"? I think so.

You made these statements:

Pretty much every computer model like this produces "better" results as teams play more games.

How is that significantly different that human polls? SLC was #1 or #2 until they lost. They are lower now. Which poll was more accurate?

When we look at prior years we are looking at complete seasons. Those rankings are MUCH BETTER than the current 2007 ranking which has less that 1/2 of the season included.

Then why bother publishing the results weekly? How many authors release their rough drafts, right?

If you don't publish them weekly and wait until they are accurate at the end of the year, then what's the point?

Also how do we know they are more accurate at the end of the year? Is it because they more closely resemble what the human polls have been saying all year?

 

[skyway28]

Again, pied is right on. If the current ratings are nothing more than the roughest of rough drafts, why put them online for public view? By making the ratings available to the public, the folks who run the program are endorsing and approving of the product.

Fact is this: The very computer model which says MNW is #34, SLC #73 and every other cr*p rating cited here so far is the EXACT SAME model which generated results in previous seasons like in 2005 rating a Wayzata, MN team that was not undefeated way ahead of 15-0 Lakeland, FL. The same basis used to generate this week's ratings will be used later on. We can assume that Marion Local will go ahead and continue to blow through the teams in Division V as they have thus far and will maintain a rating similar to what they have now. The fact that they are a DV school will never be factored into the equation.

 

[SLCDad]

Again, pied is right on. If the current ratings are nothing more than the roughest of rough drafts, why put them online for public view? By making the ratings available to the public, the folks who run the program are endorsing and approving of the product.

Fact is this: The very computer model which says MNW is #34, SLC #73 and every other cr*p rating cited here so far is the EXACT SAME model which generated results in previous seasons like in 2005 rating a Wayzata, MN team that was not undefeated way ahead of 15-0 Lakeland, FL. The same basis used to generate this week's ratings will be used later on. We can assume that Marion Local will go ahead and continue to blow through the teams in Division V as they have thus far and will maintain a rating similar to what they have now. The fact that they are a DV school will never be factored into the equation.

Apparently you can't comprehend the importance of sample size in computer modeling. That's the answer to your question.

 

[pied]

Apparently you can't comprehend the importance of sample size in computer modeling. That's the answer to your question.

I understand sample size, but basically you are making the point that the rankings are incomplete and should only be viewed later in the season, apparently 7-8 weeks is not long enough.

They are the most accurate after the season is complete.

Fine, but how do you determine that they are more accurate? What makes it more accurate at the end than the beginning.

I am going out on a limb, but if Gilmer and Katy win out, Gilmer will drop like a rock and Katy will improve. Go figure. It makes them now pointless. The question is how could these ratings more accurately rate Katy vs. St. X let's say?

What would the determinig factors for us as fans to say that the computers ranked them appropriately?

 

[MA Fan]

Apparently you can't comprehend the importance of sample size in computer modeling.

But the sample size is never going to be big enough to give accurate ratings of every HS team in the country. It probably isn't even big enough to give good results for a large state.

At this point in the season, interstate play is pretty much done. Interdivisional play in a lot of states is done. If teams from a given state or division are overrated or underrated, that's locked in--all they can do is transfer their overratedness or underratedness to other teams in their state or division.

 

[pied]

Very good points.

 

[SLCDad]

But the sample size is never going to be big enough to give accurate ratings of every HS team in the country. It probably isn't even big enough to give good results for a large state.

At this point in the season, interstate play is pretty much done. Interdivisional play in a lot of states is done. If teams from a given state or division are overrated or underrated, that's locked in--all they can do is transfer their overratedness or underratedness to other teams in their state or division.

I'll give you an example that illustrates why the sample size is having a huge impact on the Calpreps rankings at the present time. (As everybody knows, I'm very familiar with SLC so that's where the example will come from.)

SLC has played four district opponents so far. Three of those are the three worst teams in the district. As a result, SLC's strength of schedule is skewed downward. SLC has been scoring over 50 points in the first half and pulling their starters in the 2nd quarter of these games. Since these district opponents have been weak, SLC doesn't have the SOS rating to earn a high rating. That impacts the ratings of all of SLC's opponents and is one of the reasons MNW has dropped in the computer rankings.

None of this has anything to do with how good SLC or MNW really are, but it's having an impact on the computer rankings. When SLC plays the "better" teams in the district over the next few weeks and as they enter the playoffs the computer ratings will change significantly. SLC will rise in the rankings and so will their opponents. The human pollsters can take this into account. The computers can't.

 

[pied]

The human pollsters can take this into account. The computers can't.

So which is a better ranking on 10/24?

 

[skyway28]

But the sample size is never going to be big enough to give accurate ratings of every HS team in the country. It probably isn't even big enough to give good results for a large state.

At this point in the season, interstate play is pretty much done. Interdivisional play in a lot of states is done. If teams from a given state or division are overrated or underrated, that's locked in--all they can do is transfer their overratedness or underratedness to other teams in their state or division.

This is exactly correct and the basis for my point. Sample size is NOWHERE near large enough. And, with interstate play complete, we have nothing to add to the sample size as it pertains to teams from one state mathcing up with those from another.

 

[skyway28]

I'll give you an example that illustrates why the sample size is having a huge impact on the Calpreps rankings at the present time. (As everybody knows, I'm very familiar with SLC so that's where the example will come from.)

SLC has played four district opponents so far. Three of those are the three worst teams in the district. As a result, SLC's strength of schedule is skewed downward. SLC has been scoring over 50 points in the first half and pulling their starters in the 2nd quarter of these games. Since these district opponents have been weak, SLC doesn't have the SOS rating to earn a high rating. That impacts the ratings of all of SLC's opponents and is one of the reasons MNW has dropped in the computer rankings.

None of this has anything to do with how good SLC or MNW really are, but it's having an impact on the computer rankings. When SLC plays the "better" teams in the district over the next few weeks and as they enter the playoffs the computer ratings will change significantly. SLC will rise in the rankings and so will their opponents. The human pollsters can take this into account. The computers can't.

Again, the key is the ratings for those in one state or area relative to those from far away where there are no common opponents or opponents of opponents. How do we assess the value of a win over Central as compared to a win over Grapevine, for instance? Not only did Central and Grapevine not play, neither played any out-of-state games. To somehow link Central and grapevine-other, of course, then the MNW-SLC game-would be so close to impossible that your results would be meaningless.

 

[SLCDad]

But the sample size is never going to be big enough to give accurate ratings of every HS team in the country. It probably isn't even big enough to give good results for a large state.

At this point in the season, interstate play is pretty much done. Interdivisional play in a lot of states is done. If teams from a given state or division are overrated or underrated, that's locked in--all they can do is transfer their overratedness or underratedness to other teams in their state or division.

Except, the computer ratings have been shown to be as accurate as the human rankings. How do you explain that?

They looked at games between ranked teams and found that the computers did just as good or perhaps better in predicting the winner (based on who was ranked higher) than did the human polls.

Overall, the computers do an good job. They aren't perfect but no ranking is.

 

[MA Fan]

They looked at games between ranked teams and found that the computers did just as good or perhaps better in predicting the winner (based on who was ranked higher) than did the human polls.

And how large is the sample size of games between teams from different states and ranked in national polls? Large enough to draw any conclusions?

 

[SLCDad]

And how large is the sample size of games between teams from different states and ranked in national polls? Large enough to draw any conclusions?

The sample size was the ranked teams (national and regional). The conclusion we can draw is that the computers do just as good of job (or better) of ranking national and regional teams as do the human polls.

 

[skyway28]

The sample size was the ranked teams (national and regional). The conclusion we can draw is that the computers do just as good of job (or better) of ranking national and regional teams as do the human polls.

The question, dad, is regarding the sample size of matchups between teams from locations far enough removed from each other-i.e. from other states-for which there are no common opponents and on and on and on.

The matchups you refer to are those among teams from within a state.

 

[concha]

The calpreps system needs to be modified such that it attaches weight to the highest ranked opponents AFTER the lowest ranked team to which a team has lost. Example:

Opponent Rank/Result:

#234 W
#321 L
#450 W
#780 W
#1179 W...

I would program such that team #450 is given weight as the highest ranking against which the team in question has proven it can win consistently. While they did beat a team as high as #234, they have not proven they can consistently perform at such a level.

Such a system would also help in situations like that faced by Colerain. Their average regular season opponents can be pretty mediocre (their best league opponent so far is ranked #1333). That obviously limits the rankings potential for what is an exceptional program. However, doing what I propose would a) focus more on the Hoover win (which is now diluted by mediocre opponent rankings) and, b) make a win over a quality tournament foe matter more (ex. if they beat St. X).


MNW's problem in the calprep's model is that most of their opponents are apparently godawful. The SLC win just gets thrown in the hopper with the rest. Using the system I propose, the SLC game would get weighted to reflect just how good of an opponent MNW has proven they can beat (without having lost to anyone lower-ranked than SLC).

 

[MA Fan]

The sample size was the ranked teams (national and regional). The conclusion we can draw is that the computers do just as good of job (or better) of ranking national and regional teams as do the human polls.

You didn't answer the question.

1. I'm looking for a number.

2. I'm asking specifically about games between teams from different states. The majority of games that actually do get played between ranked teams are going to be between teams from the same state, which are fundamentally unrepresentative of what is being measured in national rankings.

 

[skyway28]

So, I hereby defer to MA Fan for all replies in this debate. He/she is right on the money on everything related to these national computer rankings.

 

[pied]

As the season progresses the computer models will become more accurate. The cream of the crop will rise to the top even though there are very few games between the top teams.

It's somewhat foolish to mock a computer model at this point of the season because the amount of data included is small.

OK, so eight days ago you stated that posters here were foolish for mocking the model.

Now, you have this gem:

Anybody who believes the Calpreps model right now is a fool.

Why am I reminded of the George Bush quote??

 

[SLCDad]

OK, so eight days ago you stated that posters here were foolish for mocking the model.

Now, you have this gem:




Why am I reminded of the George Bush quote??

Pied, you really don't get it do you. I'll try to make it really simple so hopefully you can understand but I'm not sure that you will.

A poster mocked the Calpreps model. He essentially said that the model should not be used for anything. He implied the model should be trashed. He said the rankings and the SOS ratings should never be used for anything EVER. Quote: "Once again, please don't cite ANY ratings-be it SOS or otherwise-from the calpreps computer as evidence of ANYTHING."

Then we discussed that the model IS producing some goofy rankings right now in some cases. Part of the reason is there is not yet enough data. The results of the model WILL get better as more games are played.

YES, it is foolish to use the output of the Calpreps model right now (that's what I posted).

HOWEVER, it IS EQUALLY FOOLISH to reject the entire model for all years because of the output for a partial season is goofy for some teams. The Calpreps model does a very good job overall. My guess is that the rankings will get better as the season progresses.

Plaindriver posted: "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! "

Pied, if you try real hard and think a little deeper you will see that my posts are consistent. You need to read what I actually posted and NOT what you think I said.

 

[pied]

you posted this eight days ago.

As the season progresses the computer models will become more accurate. The cream of the crop will rise to the top even though there are very few games between the top teams.

It's somewhat foolish to mock a computer model at this point of the season because the amount of data included is small.

Seems to me you are being pretty specific about this years results.

One week after posting that you state:

Anybody who believes the Calpreps model right now is a fool.

I keep looking at what you are posting and it appears very very clear. Even if someone was using this years results to qualify last years rankings, it seem obvious to me that you felt that with more time that the model wqould produce more accurate results.

One week later you seem to be saying the exact opposite.

 

[skyway28]

given that the computer tallies data from ALL teams on the nation, and that MOST teams' seasons are close to over, there is n fact relatively little data to be added. That the top dozen or so teams will play another seven or so games is an extremely small part of the big picture. And, there are NO interstate games remaining. There is very little reason to believe the ratings will change drastically for the teams that continue to blow through their own states. As we know, that one team is from, say, Nebraska, does nothing to devalue it. \

As MA Fan said, there is smply an ncredible shortage of info available, period.