Moved all the way up to 19 on Kenpom
- Thread starter silentsaregolden2.0
- Start date
Continuing to move up is the idea until the data from last year drops off entirely. I’m not sure if Bart Torvik has updated yet, but I believe his stats discount garbage time, so the late run by Lipscomb would maybe not be included in his rankings.
Amy metric that uses games from last year to measure this years team is useless. With the transfer portal it never will again.
There’s inertia in the output because there’s a ton of total input and we can only update a little bit of that per week. Basically this is the scientific method and we’re still in the earliest stages of the experimentation phase. Our working hypothesis at this point is still more induction and deduction than anything else. Well, Ken’s is.
Ken’s model starts with an educated guess and sets the bar rather high for new developments to prove the data wrong. That can serve to eliminate noise (such as early-season jitters, the process of gelling teams together, freshmen boneheadedness, pending coaching decisions about how to utilize rosters, and blowouts of people who were just too short for you when you actually played like hot garbage yourself) from the signal of what a basketball team really is and will be.
But it lacks the advantages of other systems which use much less frontloading. Systems which can fluctuate wildly in their predictions, but at least can claim to a much larger extent that their predictions are based on “real” data. In other words, at least as they try to represent the whole league top to bottom, systems weighted like that have more analytical power as opposed to KenPom but less predictive power. They’re less blind but more fickle.
There is immense value in both approaches.
And also in Jeff Sagarin’s approach, which is to play dead every year until he finally feels he has enough “real” data to make real predictions and then his system pops out of a bowl of rice.
At some point they’ll all tend to converge anyway. But we’re a long way from that at this point.
One more thing about KenPom in this context. Ken does love to tweak it, both to hunt for untapped predictive markers and to adjust for trends he sees in the game. But those tweaks tend to be pretty small. In general from a bird’s-eye view it’s the same system every year, with the same inherent strengths and weaknesses, the same early shakiness, which gradually attenuates at about the same rate. People are used to using it, and to a large extent they have its downsides figured in so they can still use it intuitively, and still have an intuitive idea what they’re getting from it at a given point in a season.
Ken’s model starts with an educated guess and sets the bar rather high for new developments to prove the data wrong. That can serve to eliminate noise (such as early-season jitters, the process of gelling teams together, freshmen boneheadedness, pending coaching decisions about how to utilize rosters, and blowouts of people who were just too short for you when you actually played like hot garbage yourself) from the signal of what a basketball team really is and will be.
But it lacks the advantages of other systems which use much less frontloading. Systems which can fluctuate wildly in their predictions, but at least can claim to a much larger extent that their predictions are based on “real” data. In other words, at least as they try to represent the whole league top to bottom, systems weighted like that have more analytical power as opposed to KenPom but less predictive power. They’re less blind but more fickle.
There is immense value in both approaches.
And also in Jeff Sagarin’s approach, which is to play dead every year until he finally feels he has enough “real” data to make real predictions and then his system pops out of a bowl of rice.
At some point they’ll all tend to converge anyway. But we’re a long way from that at this point.
One more thing about KenPom in this context. Ken does love to tweak it, both to hunt for untapped predictive markers and to adjust for trends he sees in the game. But those tweaks tend to be pretty small. In general from a bird’s-eye view it’s the same system every year, with the same inherent strengths and weaknesses, the same early shakiness, which gradually attenuates at about the same rate. People are used to using it, and to a large extent they have its downsides figured in so they can still use it intuitively, and still have an intuitive idea what they’re getting from it at a given point in a season.
Why is it then seemingly impossible to insert the stats and see where it all stacks up? Pretty basic really but spot on with no assumptions necessary?
You realize Kenpom this early in the season is using a ton of last seasons data right? Give it 12-15 games and it will be much more accurate.
People keep saying you need to give these things time and for a completely new team like UK that might very well be the case.
But overall his average error in predicting results of games in November is extremely similar to the average error in predicting games in March/April.
Obviously when you return 0% of the output from the previous year, that's going to be an outlier. But it doesn't completely invalidate the system currently. Because even in this world of the transfer portal, there's still a sizable return on most teams.
Given that the score predictions even in the UK game have tended to mirror Vegas, either it's calibrated ok or everyone is behind. Could very well be that everyone is behind. We've outperformed the Vegas line I believe in all of our games (or have come close).
Also instead of just looking at 19th, we should be looking at the efficiency margin number. The difference between 19th UK (21.55) and say 9th Kansas (24.54) is like 2 points in a 70 possession game. The margins are not as large between these teams.
But it's just a myth these things are completely meaningless. Look at the top of the standings. Most of those teams will still be there towards the end.
But overall his average error in predicting results of games in November is extremely similar to the average error in predicting games in March/April.
Obviously when you return 0% of the output from the previous year, that's going to be an outlier. But it doesn't completely invalidate the system currently. Because even in this world of the transfer portal, there's still a sizable return on most teams.
Given that the score predictions even in the UK game have tended to mirror Vegas, either it's calibrated ok or everyone is behind. Could very well be that everyone is behind. We've outperformed the Vegas line I believe in all of our games (or have come close).
Also instead of just looking at 19th, we should be looking at the efficiency margin number. The difference between 19th UK (21.55) and say 9th Kansas (24.54) is like 2 points in a 70 possession game. The margins are not as large between these teams.
But it's just a myth these things are completely meaningless. Look at the top of the standings. Most of those teams will still be there towards the end.
Last thing is these things don't take nearly as long.......
We started the season 43rd. After four games (three of which against meh competition) we've moved to 19th. It quickly corrects itself.
We started the season 43rd. After four games (three of which against meh competition) we've moved to 19th. It quickly corrects itself.
Yeah I think at this point it should be crystal clear to anyone paying attention, we got a very good team here.
This is basically what we had on offense last season..........but with a better defense. We brought in shooters to replace guys like Reed and Reeves and we brought in big time defensive players which is something we did not have last year.
Kenpom is the hoops equivalent of Ann Selzer . TRASH . Why would anyone , NCAA tourney selection members included , listen to this nincompoop ?
Similar threads
