Professional and College Basketball Forums banner

1 - 18 of 18 Posts

·
Premium Member
Joined
·
1,000 Posts
Discussion Starter #1
As noted by others on Twitter, 12 of the 16 A-10 teams are in the top 100 (well, 101) in the KenPom rankings. Almost as importantly, no teams in the bottom 100 (or 101).
 

·
Registered
Joined
·
6,915 Posts
As noted by others on Twitter, 12 of the 16 A-10 teams are in the top 100 (well, 101) in the KenPom rankings. Almost as importantly, no teams in the bottom 100 (or 101).
I've never been one to look at Kenpom much. How does he factor in luck? And his SOS is way different then an RPI SOS.

Sent from my Galaxy Nexus using VerticalSports.Com App
 

·
Premium Member
Joined
·
1,000 Posts
Discussion Starter #3
He has a pretty good explanation of his system here. As for Luck, he actually disaggregates it and you can see that UMass, for example, has been one of the "luckiest" teams this year while VCU is guilty of leaving the rabbit's foot at home.
 

·
Premium Member
Joined
·
1,000 Posts
Discussion Starter #5
How is it that teams with such crappy rankings are considered lucky?
I think he calculates it as a difference between predicted and actual result. i.e., if the formula can't explain it, it must be luck.
 

·
Registered
Joined
·
267 Posts
Dayton is behind Lasalle?

Dayton will be a top 10 team by the end of the season.
 

·
Registered
Joined
·
3,864 Posts
I've never been one to look at Kenpom much. How does he factor in luck? And his SOS is way different then an RPI SOS.

Sent from my Galaxy Nexus using VerticalSports.Com App
As BM explained, he does have a "luck" factor, meaning how the team has actually performed vs. how it's stats suggest it should have performed.

But the reason it's interesting is that it is all based on tempo-neutral stats and efficiency. That is, it helps you compare the defense of an uptempo team (like UMASS) vs. a slow-paced team like Richmond, and accounts for the quality (and style) of the opponent. Most people would just look at the fact that UMASS gives up 79 points a game and Richmond gives up 60 and assume that Richmond is the better defensive team. But the truth is more complicated (at least so far this year).

UMASS' plays at a faster tempo, so on a points per possession basis, which is Pomeroy's main metric, UMASS has actually played better defense so far this year.

Obviously the RPI is the one that the NCAA committee uses (sort of), but Pomeroy's stats are much more instructive about the quality and style of a team based on its efficiency on each possession.

Of course there are drawbacks (like turnovers by the walk-ons in a 30 point win get as much weight as any play by the starters with the game on the line), but it's a very informative way to better understand the game.
 

·
Registered
Joined
·
1,002 Posts
The other clear drawback and why I'm not the huge fan of KenPom many are:

Offense and defense are treated as mutually exclusive. Defense can lead to offense (steals to layups of jams), and that creates sometimes inflated offensive values for transition teams (see Duquesne under RE for example), particularly in blowouts like FQ alluded to earlier.
 

·
Registered
Joined
·
776 Posts
I suppose "luck" sounds better than "error".
It's the same as flipping a coin 100 times and having tails come up 60 times. The "best odds" are that tails would have only come up 50 times. Over 100 samples, you probably are not going to get a perfect 50/50 distribution. Over a very larger number of trials, the probability of the results getting to a perfect 50/50 split get better.

So what would you call the fact that tails came up +20 over the course of 100 trials? Luck? Expected variation? I guess it is lucky if you wanted tails to come up rather than heads. Has anyone here ever taken a course in stats?

What if you win money if tails comes up and you flip a coin three times. Three times heads comes up and you win. Aren't you lucky?
 

·
Registered
Joined
·
5,837 Posts
Has anyone here ever taken a course in stats?

What if you win money if tails comes up and you flip a coin three times. Three times heads comes up and you win. Aren't you lucky?
I guessing that if you win money on tails, and heads comes up...that you probably don't win.
 

·
Registered
Joined
·
1,478 Posts
It's the same as flipping a coin 100 times and having tails come up 60 times. The "best odds" are that tails would have only come up 50 times. Over 100 samples, you probably are not going to get a perfect 50/50 distribution. Over a very larger number of trials, the probability of the results getting to a perfect 50/50 split get better.

So what would you call the fact that tails came up +20 over the course of 100 trials? Luck? Expected variation? I guess it is lucky if you wanted tails to come up rather than heads. Has anyone here ever taken a course in stats?

What if you win money if tails comes up and you flip a coin three times. Three times heads comes up and you win. Aren't you lucky?
Yes, I've taken a course in stats before.

Pomeroy calling it luck is assuming that his model is perfectly predictive of what "should" happen, i.e. the underlying odds. Think of it another way. What if your coin is unbalanced, so that heads tends to come up slightly more than tails? Then the +20 is not +20 luck, it's actually + something random variation and + something error about the true likelihood of getting tails.

There's certainly going to be random variation around a true mean, but I'm dubious he's got the underlying odds that perfectly predicted for each game encounter, particularly this early in the season.
 

·
Registered
Joined
·
3,588 Posts
Think of it as a combination of an error term and one or more omitted variables.
 

·
Registered
Joined
·
1,478 Posts
Think of it as a combination of an error term and one or more omitted variables.
I was just picking at Pomeroy. I'm sure he recognizes that "luck" as most people define it isn't necessarily all that stat is showing, particularly early in the season. After a lot of games have been played, you could argue his model should have better predictive power due to reduced error in estimating variables, though it will still omit variables that could matter.
 
1 - 18 of 18 Posts
Top