Untitled Document
Brown Columbia Cornell Dartmouth Harvard Penn Princeton Yale



Username Post: KenPom 2018-19 Rankings        (Topic#21972)
PennFan10 
Postdoc
Posts: 3580

Reg: 02-15-15
10-21-18 06:49 PM - Post#263170    

KenPom released ore season Ivy Ranks:

1. Harvard 84
2. Yale 116
3. Princeton 148
4. Penn 152
5. Brown 199
6. Columbia 213
7. Cornell 253
8. Dartmouth 265

 
GoBigGreenBasketball 
Masters Student
Posts: 805

Age: 51
Reg: 05-19-16
Re: KenPom 2018-19 Rankings
10-21-18 10:54 PM - Post#263178    
    In response to PennFan10

I don't like where my team is but the overall league is respectably mid-major. #2BidIvy
"...no excuses - only results!”


 
Mike Porter 
Postdoc
Posts: 3615
Mike Porter
Loc: Los Angeles, CA
Reg: 11-21-04
10-22-18 01:09 AM - Post#263181    
    In response to GoBigGreenBasketball

If the key guys get and stay healthy (beyond losing Williams), I’m confident Penn will outperform. In general, I like Penn getting overlooked in many of these preseason rankings. Good motivation and will enjoy underdog roll this year. Now, here’s to hoping the guys prove them wrong again.

 
Silver Maple 
Postdoc
Posts: 3765

Loc: Westfield, New Jersey
Reg: 11-23-04
10-22-18 01:36 PM - Post#263224    
    In response to Mike Porter

I love it when the pundits gush about Harvard, and how this is finally going to be the year they do great things.

 
palestra38 
Professor
Posts: 32685

Reg: 11-21-04
10-22-18 01:45 PM - Post#263231    
    In response to Silver Maple

Except that I understand why they would do so. With Princeton, I am at a complete loss how they possibly could be ranked ahead of us. They lost their best defensive player off a defensively challenged team with no established frontcourt stars. AJ should feast on those guys.

 
mrjames 
Professor
Posts: 6062

Loc: Montclair, NJ
Reg: 11-21-04
10-22-18 07:02 PM - Post#263277    
    In response to palestra38

I wouldn’t put as much weight into KP’s system as into other systems. KP doesn’t do a lineup build - just mins continuity plus team-specific ratings baseline. So, because Princeton’s past few year baseline is a lot higher than Penn’s, KP’s system will regress them up toward their mean (minus its losses) and KP’s system will regress Penn down a bit (and a little more minus for its losses).

The lineup-based builds to preseason rankings (my preferred methodology, when I used to do this) that Bart Torvik and Dan Hanner do are far more likely to have Penn ahead of Princeton (Bart already does, and I suspect Dan will as well) due to the fact that the recent baseline may be more noise than signal in Princeton’s case.

Granted, though KP’s system is simple, it’s still quite powerful. In specific cases, though, your mileage may vary.

 
UPIA1968 
PhD Student
Posts: 1117
UPIA1968
Loc: Cornwall, PA
Reg: 11-20-06
Re: Ken pom
10-22-18 09:00 PM - Post#263280    
    In response to mrjames

These rankings, especially Harvard's, reminds us about what has to happen for Penn to realistically contend. We need two or three players to provide the SIGNIFICANT new contributions that Wang seems ready to do. The idea of depth, as talked about after the scrimmage is nice, but stars win championships not depth players. Silpe, for instance, would have to get waaaaay better to earn time on an Ivy Champ. Same for Goodman, although I think that is somewhat possible. I guess one could point to Foreman as an example of a depth player making a big jump. Still, he was clearly an outlier, who got hot at just the right time. Love him for it, but having four-year stars, ala Betley and AJ is the winning ticket.

All this said, isn't it great to be arguing Penn's chances at contention, with new talent at hand and more on the way, rather than hoping against hope as we did for ten years? I can't wait for the opening tipoff.




 
penn nation 
Professor
Posts: 21086

Reg: 12-02-04
KenPom 2018-19 Rankings
10-22-18 09:36 PM - Post#263283    
    In response to mrjames

Mike:

So help me understand, because I have noted this weighting as well:

Penn was significantly stronger than Princeton last year. Princeton was significantly stronger than Penn the season before.

Shouldn't last year be weighted more strongly than 2 years ago in terms of assessing the current team (let alone 3+ years ago, which presumably accounts for the 'baseline' you mention.....how is that at all relevant here)?

In fact, I think it was last year where Princeton got weighted much higher, not just in preseason rankings but even during the early portion of the actual season, simply because of its high ranking the previous year. We were all mystified by this until Princeton's poor play (plus presumably an eventual reduction of this early weighting) finally sent it down below Penn.

  • mrjames Said:
I wouldn’t put as much weight into KP’s system as into other systems. KP doesn’t do a lineup build - just mins continuity plus team-specific ratings baseline. So, because Princeton’s past few year baseline is a lot higher than Penn’s, KP’s system will regress them up toward their mean (minus its losses) and KP’s system will regress Penn down a bit (and a little more minus for its losses).

The lineup-based builds to preseason rankings (my preferred methodology, when I used to do this) that Bart Torvik and Dan Hanner do are far more likely to have Penn ahead of Princeton (Bart already does, and I suspect Dan will as well) due to the fact that the recent baseline may be more noise than signal in Princeton’s case.

Granted, though KP’s system is simple, it’s still quite powerful. In specific cases, though, your mileage may vary.




Edited by penn nation on 10-22-18 09:37 PM. Reason for edit: No reason given.

 
PennFan10 
Postdoc
Posts: 3580

Reg: 02-15-15
10-22-18 10:00 PM - Post#263284    
    In response to penn nation

If I am remembering correctly, not to speak for Mike (who is supposed to be rarely speaking) but I think KenPom gets more accurate as games are played as he weights current data more heavily and doesn’t adjust his models early like Torvik, et al (Which Mike already indicated). I think KP does a better job after the first third of the games are played.

 
mrjames 
Professor
Posts: 6062

Loc: Montclair, NJ
Reg: 11-21-04
Re: KenPom 2018-19 Rankings
10-23-18 08:19 AM - Post#263299    
    In response to penn nation

KP’s model is something (roughly) like this:

This Year’s Prediction = Last Year’s Finish + Difference From Team Baseline + Diff From Avg Mins Continuity Weighted For Producrion

So, let’s say the regression says you get 40% of the diff from baseline and 20 ranking spots for every 10pp you are diff from the avg mins continuity weighted for production (~2/3rds).

Princeton’s rank last year was 190. Its 5-10 year rankings baseline is ~110. So it gets 32 spots back. Then, even though it lost almost an average amount of mins, it lost no one using over 20% poss, so it’s actually going to be a little positive there, say another 10 spots.

Penn’s rank last year was 125. It’s 5-10 year rankings baseline is in the 210 range. So it loses 34 spots. Now it loses slightly fewer mins than Princeton but its mins are weighted more toward high usage, high efficiency. So, it’s going to barely get anything back from its returning value over average.

That’s a rough approximation of the simple model. I have no clue what exactly his coefficients are (and it seems like he does a separate regression for off and def efficiency versus the simple rank view I just laid out anyway), but that’s generally how you’d get there. I wouldn’t wholly dismiss the model, because it has a decent predictive track record, but I get more out of a player-by-player build. None of these have been developed to predict end of year finish, rather they exist so that early-season games can be judged in proper context to make the ratings converge faster (e.g. we know a win over UNC says you’re a much better team than a win over UNH, why treat both opponents on day one like they’re the same).

Hope that helps!

 
penn nation 
Professor
Posts: 21086

Reg: 12-02-04
Re: KenPom 2018-19 Rankings
10-23-18 08:32 AM - Post#263302    
    In response to mrjames

Thanks.

Still don’t understand why a 5-10 year baseline is included, though.

 
Penndemonium 
PhD Student
Posts: 1878

Reg: 11-29-04
Re: KenPom 2018-19 Rankings
10-23-18 09:22 AM - Post#263309    
    In response to mrjames

I was just thinking that the kenpom questions need a mrjames reply. Lo and behold. Thanks!

  • mrjames Said:
KP’s model is something (roughly) like this:

This Year’s Prediction = Last Year’s Finish + Difference From Team Baseline + Diff From Avg Mins Continuity Weighted For Producrion

So, let’s say the regression says you get 40% of the diff from baseline and 20 ranking spots for every 10pp you are diff from the avg mins continuity weighted for production (~2/3rds).

Princeton’s rank last year was 190. Its 5-10 year rankings baseline is ~110. So it gets 32 spots back. Then, even though it lost almost an average amount of mins, it lost no one using over 20% poss, so it’s actually going to be a little positive there, say another 10 spots.

Penn’s rank last year was 125. It’s 5-10 year rankings baseline is in the 210 range. So it loses 34 spots. Now it loses slightly fewer mins than Princeton but its mins are weighted more toward high usage, high efficiency. So, it’s going to barely get anything back from its returning value over average.

That’s a rough approximation of the simple model. I have no clue what exactly his coefficients are (and it seems like he does a separate regression for off and def efficiency versus the simple rank view I just laid out anyway), but that’s generally how you’d get there. I wouldn’t wholly dismiss the model, because it has a decent predictive track record, but I get more out of a player-by-player build. None of these have been developed to predict end of year finish, rather they exist so that early-season games can be judged in proper context to make the ratings converge faster (e.g. we know a win over UNC says you’re a much better team than a win over UNH, why treat both opponents on day one like they’re the same).

Hope that helps!




 
TheLine 
Professor
Posts: 5597

Age: 60
Reg: 07-07-09
Re: KenPom 2018-19 Rankings
10-23-18 10:39 AM - Post#263317    
    In response to penn nation

  • penn nation Said:
Thanks.

Still don’t understand why a 5-10 year baseline is included, though.


Most teams perform within a certain historical range year after year so regression to the mean typically makes sense.

It's possible (probable?) Penn is an outlier because Donahue is a better coach than the previous two. That's likely to cause KenPom' model to underrate Penn somewhat until a new baseline is established. That's not to say KenPom is 'wrong', just that his model doesn't evaluate the difference in coaching because it's not a scenario that comes up often enough and it's difficult to accurately predict.


 
mrjames 
Professor
Posts: 6062

Loc: Montclair, NJ
Reg: 11-21-04
Re: KenPom 2018-19 Rankings
10-23-18 05:53 PM - Post#263377    
    In response to TheLine

Yep - that’s why I like the lineup builds *especially* for a team like Penn.

 
Icon Legend Permissions Topic Options
Report Post

Quote Post

Quick Reply

Print Topic

Email Topic

1456 Views





Copyright © 2004-2012 Basketball U. Terms of Use for our Site and Privacy Policy are applicable to you. All rights reserved.
Basketball U. and its subsidiaries are not affiliated in any way with any NCAA athletic conference or member institution.
FusionBB™ Version 2.1 | ©2003-2007 InteractivePHP, Inc.
Execution time: 0.503 seconds.   Total Queries: 16   Zlib Compression is on.
All times are (GMT -0500) Eastern. Current time is 03:05 AM
Top