Pregame: Illinois vs Houston, Thursday, March 26th, 9:05pm CT, TBS

Status
Not open for further replies.
#533      
LouHenson
Greenville was a blast, if we can match the intensity I saw there, we can for sure win this game. Tomi played really well in both games, but against VCU he was very good. If we can keep Tomi now with the X factor of Andrej we are really tough to game plan for.
 
#534      
AirTran
wc_illini said:
Source?

I agree that sample size and styles of play are important considerations when looking at metrics, but the above does not agree with what I've seen about Torvik and KenPom. It is somewhat true with EvanMiya (because he performs an additional step to evaluate over/under performance as a function of opponent quality), but he has some methods in place to reduce the impact of sample size in that

Also, the fact that the line is relatively close to the metrics (after accounting for a small location advantage) on this kind of makes this argument moot w.r.t. this game
Click to expand...
Well, I like your answer very much. Professionally I use only quantitative and stochastic data and probability. The problem I (and most college basketball teams) have is that Kenpom anyway is too subjective on one side and materially statistically biased on the other.

The number 1 NCAA tourney money bet is betting against the Kenpom spread. Has been for 5 or 6 years. So your observations are (mathematically speaking) 100% accurate.
 
#535      
Thaillini22
Flemings hasn't hit a three in the last three games, only attempting three of them (per Sleepers). Also, we have more weapons offensively. I like the stats that seem to indicate if we keep UH to a sub-par offensive game, we win. Bring the defensive intensity, take care of the ball, get hot Keaton... we'd be in real good shape.
 
#536      
illinifan1987
Illini4Christ said:
View attachment 48592
Click to expand...
This is just another staple in what will become our greatest season in school history. Led by the all-time Freshman 🐐 and another top 5 Freshman in the Underwood era. After all of the bull:poop: we've been through, it's long overdue (y) Mirk as a Sophomore or even Junior?
Homer Drool GIF
 
#538      
aaeismacgychel
wc_illini said:
Source?

I agree that sample size and styles of play are important considerations when looking at metrics, but the above does not agree with what I've seen about Torvik and KenPom. It is somewhat true with EvanMiya (because he performs an additional step to evaluate over/under performance as a function of opponent quality), but he has some methods in place to reduce the impact of sample size in that

Also, the fact that the line is relatively close to the metrics (after accounting for a small location advantage) on this kind of makes this argument moot w.r.t. this game
Click to expand...
So this is more a statistical concept than talking about a specific ranking system though it does pertain to them intrinsically. The idea is when you're trying to fit a curve to data your data points closest to the ends are your more extreme points and have a higher intrinsic weight value. One simple way of seeing this is looking at a regression curve for say a linear data set and raising a single point much higher or lower on the line. If you raise a data point in the middle of the set, you'll see the whole line move up slightly but slope won't move much. If you put that on the end however, the line will tilt. Data points at your boundaries are simply intrinsically unstable.

Now both Kenpom and Torvik have ways built in to mitigate score effects such that in Kenpom for example a 60pt victory means only negligibly better than say a 35pt victory, and I'm not saying otherwise. In fact there is some extremely good discussion on what those score effects should be and how they should be calculated in an ideal world. But that's a separate issue than the one I was stating which is simply that the mere consideration of these points en masse in it of itself causes instability and that it does skew the data more towardsthe end points. Hence this is a systemic analytical issue regarding the sample data you're making your population assessments on, not an isolated individual system one. More plainly, if you think about it instead from an error bar or confidence interval perspective, teams having a lot of these points bakes in larger error and more noise. There's just no way around it. So a 1pt difference between teams metrics wise is fairly insignificant. Looking at weight adjusted metrics might be of more significance here.

As for the how this would show up say in metrics differential vs betting line, they're not 1 to 1 but they are fairly ballpark. For example it's rare to see a game where say Kenpom vs opening line is off by >2pts. And in that respect what I'm saying is indeed moot for that argument. What I'm saying instead is that there's enough uncertainty that a 1pt differential in metrics is fairly insignificant. That doesn't mean Vegas will open as a pick 'em or that they'll be greatly off from the metrics. They utilize their own collection of systems and formulas. Just that noise is noise and your output is only as good as your input.
 
#542      
stevej64
Chuck Nuggets said:
Really great quotes from Sampson about Illinois: "There's certain teams that you scout them for the first time, I haven't seen Illinois play much this year at all, but scouting 'em and just watching 'em for the first time I found myself being a fan. Just enjoyed watching them."

https://twitter.com/x/status/2036134319480037820
Click to expand...

Fun fact for you young folks, Ron Gunther was ready to hire Sampson but when he showed up at the airport in Oklahoma there was a media contingent waiting for him and he turned around and left without interviewing him. Gunther always acted in stealth mode and was pissed off. Can’t remember if that was when we ended up hiring Self or Weber though.
 
#544      
Iloveorangekoolaid
Fly Illini Fly! said:
Apparently the “eye test” of our Penn game says “Illinois not good.” Because, you know, all we did was rebound which is totally out of “luck” and where the ball is bouncing.
Click to expand...
I rewatched both of our games and I actually don’t think we played great in either of them. That’s just how great we can be. When everything is clicking, this team is ELITE. But I loved the intensity and tenacity we played with. That’s what we need to bring against Houston.
 
#546      
AirTran
aaeismacgychel said:
So this is more a statistical concept than talking about a specific ranking system though it does pertain to them intrinsically. The idea is when you're trying to fit a curve to data your data points closest to the ends are your more extreme points and have a higher intrinsic weight value. One simple way of seeing this is looking at a regression curve for say a linear data set and raising a single point much higher or lower on the line. If you raise a data point in the middle of the set, you'll see the whole line move up slightly but slope won't move much. If you put that on the end however, the line will tilt. Data points at your boundaries are simply intrinsically unstable.

Now both Kenpom and Torvik have ways built in to mitigate score effects such that in Kenpom for example a 60pt victory means only negligibly better than say a 35pt victory, and I'm not saying otherwise. In fact there is some extremely good discussion on what those score effects should be and how they should be calculated in an ideal world. But that's a separate issue than the one I was stating which is simply that the mere consideration of these points en masse in it of itself causes instability and that it does skew the data more towardsthe end points. Hence this is a systemic analytical issue regarding the sample data you're making your population assessments on, not an isolated individual system one. More plainly, if you think about it instead from an error bar or confidence interval perspective, teams having a lot of these points bakes in larger error and more noise. There's just no way around it. So a 1pt difference between teams metrics wise is fairly insignificant. Looking at weight adjusted metrics might be of more significance here.

As for the how this would show up say in metrics differential vs betting line, they're not 1 to 1 but they are fairly ballpark. For example it's rare to see a game where say Kenpom vs opening line is off by >2pts. And in that respect what I'm saying is indeed moot for that argument. What I'm saying instead is that there's enough uncertainty that a 1pt differential in metrics is fairly insignificant. That doesn't mean Vegas will open as a pick 'em or that they'll be greatly off from the metrics. They utilize their own collection of systems and formulas. Just that noise is noise and your output is only as good as your input.
Click to expand...
All accurate observations... at the same time its (1) a human correction, (2) Theses teams do not play the same schedule and (3) based on the adjustment factor, the more times you play highly ranked opponents (where there is human adjustment to several factors), the more statistically biased the output.
 
#549      
wc_illini
aaeismacgychel said:
The idea is when you're trying to fit a curve to data your data points closest to the ends are your more extreme points and have a higher intrinsic weight value. One simple way of seeing this is looking at a regression curve for say a linear data set and raising a single point much higher or lower on the line. If you raise a data point in the middle of the set, you'll see the whole line move up slightly but slope won't move much. If you put that on the end however, the line will tilt. Data points at your boundaries are simply intrinsically unstable.
Click to expand...
Sure, but KenPom and Tovik ratings don't try to fit a curve vs opponent quality- they're both just calculating season-long averages for adjusted off, def, and tempo, so there's no line/curve to tilt (well, you could say there is a curve depending on your frame of reference, but its slope/shape is constrained). Whether a team's games against Q4 opponents tell you anything about their future results against Q1 opponents is a good question, and I like that Torvik allows you to re-rank after performing filters on that. Sample sizes get pretty small though.

Limited inter-connectedness of teams can create a similar-sounding issue. If only one big ten team played one big 12 team, and all the other games were in-conference, that one game would determine how the teams from the two conferences are aligned in the rankings. While B1G and B12 teams have faced each other, those games were mostly a long time ago, so this is a genuine concern.

aaeismacgychel said:
So a 1pt difference between teams metrics wise is fairly insignificant. Looking at weight adjusted metrics might be of more significance here.
Click to expand...
I agree that 1pt is fairly insignificant. Not sure what you mean by weight-adjusted metrics (edit: but KenPom does employ game weights since he's determined that certain types of games are better indicators of team quality).
 
Last edited:
Status
Not open for further replies.
Back