TrueTorontoFan wrote:
View Post
Announcement
Collapse
No announcement yet.
Everything VanVleet
Collapse
X
-
DanH wrote: View Post
The sample sizes required to stabilize for different things are... different. Three point shooting is one of the slowest developing stabilizations. Raw points scored over a variety of possession types normalizes far more quickly - more variance in play type and opponent brings down the impact of high variation in any single play type and the impact of quality of opposition. You can still get outlier data but it's far less prominent and you can far more quickly have some confidence in projecting results. Obviously not 100% confidence, but still, you can feel good about using those sorts of trends as evidence. This isn't me saying this, this is understood about these measures and which ones normalize fastest. 3 point percentage is about the slowest stabilizing data point in the sport (outside of those where there is literally no correlation/prediction possible, like most defensive FG% stats - making defensive 3 point % especially useless).
The Birch sample is very small with each of those lineups but not bad when combined, and the point is more about us gathering a pretty decent sample of the small look NOT working very well, which would suggest trying the alternative would be a good idea. The alternatives with the biggest small samples with positive outcomes are the ones where Birch slides into that starting spot. What exactly is the harm in trying?
And more to the point, how can we possibly say with confidence that Birch starting would not help the starters, when all the evidence points entirely the other way? If the argument is the evidence is not strong enough, then we should definitely not be able to say anything about the benefits or costs of Birch starting with confidence. Right?
Comment
-
The Claw Reborn wrote: View Post
Maybe somehow at least watch some games
Is your position that I haven't watched every game this year? Is that honestly your position or are you just trolling? Either seems kind of dumb.
- 1 like
Comment
-
DanH wrote: View Post
Ah, the refrain of the most intelligent basketball observers. Congrats on grouping yourself with THAT crowd. Quite an accomplishment.
Is your position that I haven't watched every game this year? Is that honestly your position or are you just trolling? Either seems kind of dumb.DanH wrote: View PostThe small lineup with all 5 non-C starters has played nearly 100 minutes this season and is still a not quite break-even lineup (-0.6 net rating).
The looks with Birch in Trent's place (+7 net rating, 22 minutes) or Barnes' place (+13.7 net rating, 39 minutes) are both smaller samples but at least they are winning their minutes (Precious' results are even smaller samples and are all over the place in terms of success or lack thereof, as you'd expect from smaller samples).
There's not much evidence that the starters are better going small than going with Birch. Trent and Barnes both have games that can shine with the bench and are very valuable with the bench. They've been making the small starting group work by changing to the big one as quickly as they can in the 1st and 3rd Qs with an early sub for Birch, but I'm sure that is just locker room politics, easing into the obvious decision they want to make (starting Birch, Trent or Barnes to the bench), though it is also obvious how they would have a hard time removing either player from the starting group considering both have played great, but the lineup just doesn't work very well. It's not a disaster but I don't think "non-disaster" is their goal here.DanH wrote: View PostThe sample sizes required to stabilize for different things are... different. Three point shooting is one of the slowest developing stabilizations.
You have different standards especially if it helps you prove your crooked point.
This was the same ad nauseam you regurgitated when you defended the minutes of Gasol.
Comment
-
The Claw Reborn wrote: View Post
I am pointing out the inconsistencies of your stance.
You have different standards especially if it helps you prove your crooked point.
This was the same ad nauseam you regurgitated when you defended the minutes of Gasol.
My entire point was that there is NOT sufficient evidence that the small look is better. So when someone says there is no benefit to trying Birch over Trent, my response is there's no evidence for that, even the small samples of evidence we do have point in the opposite direction. How is any of that inconsistent?
Comment
-
DanH wrote: View Post
None of that is inconsistent. I clearly point out that the decent results from Birch starting are in a small sample, so to take with a grain of salt, but the driving point is we have a pretty big sample of the small look not really working very well. So we should try other things - leaving aside that they should obviously have been trying Birch as a starter right from the beginning, him being the closest thing they have to a starting quality C.
My entire point was that there is NOT sufficient evidence that the small look is better. So when someone says there is no benefit to trying Birch over Trent, my response is there's no evidence for that, even the small samples of evidence we do have point in the opposite direction. How is any of that inconsistent?
Yet them defending the three is an unstable sample size
ok let’s move on
Comment
-
The Claw Reborn wrote: View Post
So you are deducing a conclusion that the small look is not really looking well based on a big sample but I have to take it with a grain of salt the decent results from Birch starting even though in small sample. So they should try other things and blame Nurse because he could have tried starting Birch as a Center from the very beginning of the season because he is the closest thing to a starting quality Center LOL.
Yet them defending the three is an unstable sample size
ok let’s move on
Sample sizes needed to stabilize are different for different things. This is not a complex concept, I am sure you get it.
If you roll 100 dice together all at once, how many rolls before the sum of that stabilizes to an expected outcome?
If you roll one die, how many rolls before the average of its rolls stabilize to the expected outcome?
The raw net rating data for players and lineups is actually a summation of a huge amount of random variables - as such, they stabilize relatively quickly.
3 point percentage is the opposite. It's one die being rolled over and over again (except with more variance than a 6 sided die example can demonstrate), and it takes a lot of attempts to normalize. Defensive FG% is more variables, but also has been shown not to correlate over big samples.
So you are taking a data point that takes ages to stabilize in the best case and literally never does in the worst case and asking why the sample should be treated any differently from the net rating concepts that have been shown to be more predictive of final records than actual win loss records tend to be during the season - that are specifically of value in smaller samples as they inherently wash out a lot of random variance?
Yes. Different data should be treated differently, based on what we know about the behaviour and predictive value of those data sets. Not sure why this is construed as me mixing messages.
- 1 like
Comment
-
DanH wrote: View Post
All of that is correct.
Sample sizes needed to stabilize are different for different things. This is not a complex concept, I am sure you get it.
If you roll 100 dice together all at once, how many rolls before the sum of that stabilizes to an expected outcome?
If you roll one die, how many rolls before the average of its rolls stabilize to the expected outcome?
The raw net rating data for players and lineups is actually a summation of a huge amount of random variables - as such, they stabilize relatively quickly.
3 point percentage is the opposite. It's one die being rolled over and over again (except with more variance than a 6 sided die example can demonstrate), and it takes a lot of attempts to normalize. Defensive FG% is more variables, but also has been shown not to correlate over big samples.
So you are taking a data point that takes ages to stabilize in the best case and literally never does in the worst case and asking why the sample should be treated any differently from the net rating concepts that have been shown to be more predictive of final records than actual win loss records tend to be during the season - that are specifically of value in smaller samples as they inherently wash out a lot of random variance?
Yes. Different data should be treated differently, based on what we know about the behaviour and predictive value of those data sets. Not sure why this is construed as me mixing messages.
yet here you are, stating that the small line up is already not a break even line up with a minus 0.6 net rating but could have been better with Birch as a starter with a +7 net rating based on ……wait for it ….small sample size…
And now … you are accusing me to be taking a data point that would take ages to stabilize in the best case and never does in the worse case.
Comment
-
The Claw Reborn wrote: View Post
From the very beginning of our discussion about my take on the Raptors’s lackluster data in defending the three (and that was 20 games if I’m correct), you already rain on that parade and called it as noise….still a small sample size.
yet here you are, stating that the small line up is already not a break even line up with a minus 0.6 net rating but could have been better with Birch as a starter with a +7 net rating based on ……wait for it ….small sample size…
And now … you are accusing me to be taking a data point that would take ages to stabilize in the best case and never does in the worse case.
Lineup and team net rating data stabilizes much faster. The Birch sample is too small to rely on as yet, the small lineup sample is pretty decent. The small lineup is losing its minutes, which is bad, so we should explore other options. The Birch option doesn't have sufficient data to form a solid opinion on, but at least they've won those minutes, so they should try it more to increase the amount of data they have on it to find out if they can translate that small sample success into a more reliable sample while still having success. Once he is back from his broken nose, of course. Ideally you generate a solid sample of all your best options well before the playoffs so you know what works and what doesn't by the time it matters. Maybe the Birch lineup proves to be worse than the small look - won't know until we try!
Comment
-
DanH wrote: View Post
Right. 3 point shooting takes ages to stabilize for an individual player (or even for a team) and for a defence literally never stabilizes. It's never predictive.
Lineup and team net rating data stabilizes much faster. The Birch sample is too small to rely on as yet, the small lineup sample is pretty decent. The small lineup is losing its minutes, which is bad, so we should explore other options. The Birch option doesn't have sufficient data to form a solid opinion on, but at least they've won those minutes, so they should try it more to increase the amount of data they have on it to find out if they can translate that small sample success into a more reliable sample while still having success. Once he is back from his broken nose, of course. Ideally you generate a solid sample of all your best options well before the playoffs so you know what works and what doesn't by the time it matters. Maybe the Birch lineup proves to be worse than the small look - won't know until we try!
lets move on
Comment
-
golden wrote: View Post
This list makes a mockery of the mock drafts and so-called draft experts.
Comment
Comment