Posted: Fri Feb 17, 2006 1:18 pm Post subject: Roland Ratings
Let me start by saying I really appreciate the 82games website. Processing raw information, formatting it, commenting on it and making it publicly available at no charge. That's really the internet at its very best. Being a Raptor fan, I have e-mailed Toronto management numerous times suggesting they avail themselves of the services.
That said, I have a bone to pick.
Since 82games started up one of their best features has been the Roland Ratings (OnCourt/OffCourt differentials.) These were featured on the front page of each team and offered a great alternative to PER and other rating schemes based on individual statistics.
Obviously OnCourt/OffCourt numbers are subject to numerous weaknesses and I was hoping they would eventually include at least some rudimentary adjustments for "teammates on the court" and "opponents on the court" etc. The kind of thing Dan Rosenbaum has done so well.
Instead, 82games has begun combining the On/Off data with PER and Opponent Counterpart PER in the Roland Rating. I realize the raw OnCourt/OffCourt numbers are still available on the site but the new presentation greatly emphasizes the combined numbers and undermines the "alternative viewpoint" approach. I feel the Opponent Counterpart PER, in particular, really taints the ratings.
Opponent PER is basically worthless, unless the NBA was a game where everyone played man all the time, and always guarding the opposing player of the same position, however that's not close to how the real game works, there are many variations in defensive sets. I consider this an instance, which a lot of stats are guilty of, of trying to make the triangle piece fit in the square hole.
So yeah, I agree with you, and this is something that has stood out to me also.
Detroit: Opponent's Counterpart PER
11.5 Prince
13.5 Billups
14.1 Hamilton
14.1 B.Wallace
17.3 R.Wallace
Wow, According to this the Detroit big men are nothing special defensively. In fact, they are being carried by the perimeter defenders. The problem is, this flatly contradicts Dan Rosenbaum's rigorous three year study of individual defense which ranked Ben Wallace as the league's top defender. Rasheed fourth among power forwards and didn't mention Chauncey, Rip or Tayshaun at all.
Of course, Rosenbaum's study was based on data from previous seasons, but have things really have changed that much? Is Ben no longer the key to the Piston's defense? Let's use 82games "Player Pairs" to calculate the Points per 48 minutes allowed by the Piston's starters when operating without Big Ben on the court.
Detroit: Points Allowed per 48 Minutes
89.8 R.Wallace On, B.Wallace Off
89.8 Prince On, B.Wallace Off
95.0 Billups On, B.Wallace Off
95.3 Hamilton On, B.Wallace Off
BTW, all the combinations with Ben OnCourt and one of the other starters OffCourt are better than those above, ranging from 84.3 (Ben On, Rip Off) to 88.4 (Ben On, Rasheed Off)
So what's going on here? Why do the "Opponent Counterpart PERs" differ so sharply from the OnCourt/OffCourt data?
I would suggest that the problem is that the "Counterpart PER" works on the mistaken assumption that defense is primarily "man-to-man" whereas, in reality, the Piston's play a scheme where the perimeter defenders job is mainly to funnel their men into the 3 time DPOY and his pal 'sheed. Any semi-athletic guard who stays within the system should be able to do a reasonable job of that. Without the Wallace/Wallace defense behind them Billups and Rip are exposed as mediocre defenders.
I have no particular problem with PER or even Opponent Counterpart PER when presented on their own. They are just numbers which can be used, ignored or interpreted as the user pleases. However, I think it's unfortunate that 82games has decided to include them in their featured "Roland Ratings."
Posted: Fri Feb 17, 2006 2:22 pm Post subject: A different view
Opponent's Counterpart PER isnt simply saying "Wow, According to this the Detroit big men are nothing special defensively.", it is a statment about their play on both sides of the court, defense and offense, and locally rather than globally. Ben Wallace - in his role- doesnt score at anywhere near an average rate and that offsets a large part of his defensive prowess, and Rasheed can be scored on if you take right at him at a pretty average 19 pts per 48 minutes overall (he probably rates as a little below average at one on one center defense given the lower league offense quality there. That affects their opponent PERs and is fair. But you are correct that it fails to pick to help defense value (they prefer a lot of funneling of other guys to the inside second level of defense as you described because it leads to bad mid-range shots, turnovers, out of sync play) but that shows up on the team +/- side of the formula, so I can live fine with the local/global combination and see it is as an improvement over the original just team global data.
And instead of the other 50% of the Roland rating being team +/- on/off, I have suggested it be 25% team +/- just while on and then 25% on/off because this emphasizes their on court number more and lessens the mixed impact/meaning of how good their substitute is. Of course I also would prefer fully adjusted on/off scoring, but they have not chosen to release that.
"Without the Wallace/Wallace defense behind them Billups and Rip are exposed as mediocre defenders. " I tend to disagree based on observation and FG% allowed stats which are influenced by help defense but hard to be good unless you can defend yourself as well. I believe they are well above average as defenders themselves and that it is nearly half the reason they succeed so well. Unfortunately 82games data doesnt exist for their pre-Detroit days to inform on this.
"I think it's unfortunate that 82games has decided to include them in their featured "Roland Ratings. It’s a big step in the wrong direction."
I feel differently. This newer version of Roland rating can provide some useful general but not perfect information on overall player impact. Having it totally team-based always seemed a major weakness with the individual player's impact on team stats very dilute- only one of 10 players on the court when on and 0 of the 10 when off- and the case for exactly what they have contributed toward the team on/off +/- results can be uneven and not accurate in all cases- I think they are many free riders on good teams and guys failing to get a proper score based on their own abilities/performance because of lousy teammates, not identifiable by just team +/- on/off.
Having it now be half based on counterpart production seems more appropriate. 100% on counterpart production can also be criticized as several have done properly for being overly local given the way the game is actually played and won or lost, but 50% on counterpart and 50% overall team seems like a pretty good choice to try to match fairly closely with local/global reality to me. Even with switches and help, I think it is still close to accurate to think that at least 50% of plays come down to one on one play against their normal counterpart. I think the newer Roland Rating basis is a worthwhile step toward being more about the specific player in balance with the total team effect that can be further improved as I suggest I above.
Until they go fully adjusted, this a decent compromise. But you can still use the team +/- on, or on/off, or counterpart net points and/or PER as well, along with lots more, so there are a range of choices available to for folks to find the one they are interested in, or look at them all.
Last edited by Mark on Thu Mar 02, 2006 10:31 am; edited 1 time in total
One problem with your theory is opponent PER is not really a defensive stat. Its a production stat that equates both ends of the floor. 17.3 opponents PER doesn't really suggest that the Pistons are being carried by there perimeter players. Opponents PER is not like points per game. Its about alot more than just someone scoring on you. Rasheed Wallace is one of the worst rebounders at his position in the NBA. His rebounding numbers hurt his PER and help this opponents. Not only that Rasheed biggest weakness as a rebounder is on the offensive glass. Which means that Rasheed's opponents are grabbing alot of defensive rebounds.
The only guy on that list that defensive numbers are lacking is Rasheed Wallace. Ben Wallace has always been a guy that team defense is alot better than his individual defense. For PF/C around the league scoring on Ben Wallace in the paint has not been something you can do and do very well. Plus the Pistons defense has not been as good this year.
Detroit: Opponent's Counterpart PER
11.5 Prince
13.5 Billups
14.1 Hamilton
14.1 B.Wallace
17.3 R.Wallace
Wow, According to this the Detroit big men are nothing special defensively. In fact, they are being carried by the perimeter defenders. The problem is, this flatly contradicts Dan Rosenbaum's rigorous three year study of individual defense which ranked Ben Wallace as the league's top defender. Rasheed fourth among power forwards and didn't mention Chauncey, Rip or Tayshaun at all.
Of course, Rosenbaum's study was based on data from previous seasons, but have things really have changed that much? Is Ben no longer the key to the Piston's defense? Let's use 82games "Player Pairs" to calculate the Points per 48 minutes allowed by the Piston's starters when operating without Big Ben on the court.
Detroit: Points Allowed per 48 Minutes
89.8 R.Wallace On, B.Wallace Off
89.8 Prince On, B.Wallace Off
95.0 Billups On, B.Wallace Off
95.3 Hamilton On, B.Wallace Off
BTW, all the combinations with Ben OnCourt and one of the other starters OffCourt are better than those above, ranging from 84.3 (Ben On, Rip Off) to 88.4 (Ben On, Rasheed Off)
So what's going on here? Why do the "Opponent Counterpart PERs" differ so sharply from the OnCourt/OffCourt data?
I would suggest that the problem is that the "Counterpart PER" works on the mistaken assumption that defense is primarily "man-to-man" whereas, in reality, the Piston's play a scheme where the perimeter defenders job is mainly to funnel their men into the 3 time DPOY and his pal 'sheed. Any semi-athletic guard who stays within the system should be able to do a reasonable job of that. Without the Wallace/Wallace defense behind them Billups and Rip are exposed as mediocre defenders.
I have no particular problem with PER or even Opponent Counterpart PER when presented on their own. They are just numbers which can be used, ignored or interpreted as the user pleases. However, I think it's unfortunate that 82games has decided to include them in their featured "Roland Ratings."
It’s a big step in the wrong direction.
Let me say first that I also think that combining the stats isn't the right direction.
With that said, I think both stats are useful. It may seem absurd that the Wallaces have such numbers, but if you've been following the Pistons results you'll notice that lots of big men have been going off on the Pistons this year. Now, I'm fine with finding a reason why that's not the Wallaces' fault, but I think it's clear why those less than impressive numbers exist.
You should also realize that Detroit's defense isn't as strong as it's been in previous years, and that defensive reputations are often based off of previous years. If you look at Ben's opponent's PER in recent years, you'll notice in previous years (when he was closer to the time when he first earned his rep) he was significantly better:
05-06: 14.1
04-05: 12.9
03-04: 11.5
With that said, I wouldn't ever trust this stat alone to measure a defender's worth. I see to many possible corruptors (positional bias of PER, blame association with help defenders, putting stronger defenders on stronger players, etc)
If you look at Ben's opponent's PER in recent years, you'll notice in previous years (when he was closer to the time when he first earned his rep) he was significantly better:
05-06: 14.1
04-05: 12.9
03-04: 11.5
I seem to be lost.
I incorrectly assumed "Production / Opp" in the Roland Ratings table was the same as PER on the individual player's "by Position" page, under the heading "Opponent Counterpart 48-Minute Production." But clearly the numbers do not mach.
What is the difference between these numbers? Where do the numbers you quoted above come from?
I incorrectly assumed "Production / Opp" in the Roland Ratings table was the same as PER on the individual player's "by Position" page, under the heading "Opponent Counterpart 48-Minute Production." But clearly the numbers do not mach.
Roland is working on something new that will make up the Roland Ratings in the near future, but what exactly that is remains uncertain.
If you think about, I think you can sympathize with what Roland is thinking. Roland Ratings are the signature rating of the site, so when someone sees Greg Ostertag in the top ten, they might instantly write off the site. Including an individual-stat-based measure of player quality ensures the ratings pass the laugh test.
My issue is that I've branded the on-court/off-court differential as Roland Rating, and I'm not really sure what to call it now.
I understand wanting the featured statistic to seem reasonable to traditionalists. However, I am not sure incorporating something as deeply flawed as "Opponent Counterpart PER" is a good solution.
Isn't there a better way to deal with this. Take a look at Dan Rosenbaum's Defensive Player Ratings. His methodology is sound and I don't see a problem with those passing a "laugh test."
2. Across from Ben Wallace's name I see 19.2 under the "Production / Own" column and 14.1 under the "Production / Opp" column.
3. I click on Wallace's name and then click the "by Position" tab.
4. Looking under "Player 48-Minute Production by Position" I see a PER of 19.2. That matches the the Roland "Own" number. However, looking under "Opponent Counterpart 48-Minute Production" I see a PER of 16.7 -- which does not match the Roland "Opp" number.
This is the same for all the players I checked. What am I missing here?
If you look at Ben's opponent's PER in recent years, you'll notice in previous years (when he was closer to the time when he first earned his rep) he was significantly better:
05-06: 14.1
04-05: 12.9
03-04: 11.5
I seem to be lost.
I incorrectly assumed "Production / Opp" in the Roland Ratings table was the same as PER on the individual player's "by Position" page, under the heading "Opponent Counterpart 48-Minute Production." But clearly the numbers do not mach.
What is the difference between these numbers? Where do the numbers you quoted above come from?
It's my understanding that the PER stats used in the Roland ratings are based off of whoever the player happens to be guarding at any given time. Whereas the opponent counterpart is simply how the opposing player playing the same position as the player in question is doing.
What's the difference? Well Roland opponent's PER then is a far better judge of what opponent's production the player is actually responsible for since it doesn't make broad stroke assumption that a player is always playing man defense on his counterpart.
If your explanation is correct, I was unaware of it. It should be clearly explained. If true it allows new analysis on who helps well/bad, and possibly infer about receipt of help too and impact on counterpart defense numbers.
But for this to be true it would seem there would have to be visual confirmation of player matchups. The other thread seemed to be emphasizing 82games use of play by play ( I assume written data).
Which is used, one or both? I would think some further clarification on these matters would be appropriate.
Last edited by Mark on Wed Feb 22, 2006 8:03 pm; edited 1 time in total
If your explanation is correct, I was unaware of it. It should be clearly explained. If true it allows new analysis on who helps well/bad, and possibly infer about receipt of help too and impact on counterpart defense numbers.
But for this to be true it would seem there would have to be visual confirmation of player matchups. The other thread seemed to be emphasizing 82games use of play by play ( I assume written data).
Which is used one or both? How? I would think some clarification on these matters would be approproate, not unreasonable and helpful to 82games by increasing user understanding and confidence.
Well, maybe I'm wrong. Someone told me this who said they had talk with 82games, and I didn't doubt it because honestly I can't think of another explaination that makes any sense. Presuming that the Roland opponent's PER is not just a combination of the couterpart PERs (which seems obvious to me looking at the stats, but again I could be wrong), what other possibilities seem plausible?
Note that on the other thread, while play-by-play was emphasize, I don't think most people (I sure wasn't) were trying to claim that absolutely everything on 82games could be made from play-by-play. For my part, I was trying to say that most stats, including the +/- ones which have been the focal point of the site, could be gotten from play-by-play, and so while I could understand having issues with some stats, it doesn't make sense to throw everything on the site under the bus.
You're right of course that you can only do so much without visual confirmation. I would presume this is part of what thier game trackers do (you know they've in the past put out calls for game tracker, right?), and why also this stat of theirs only dates back to 03-04 whereas the +/- stats date back earlier.
Is anyone here a video tracker for them? I would think it likely somebody here would be. Maybe they could offer some insight.
I left a heads up with Roland at his site about this overall discussion.
I inquired twice about video game tracking work opportunity months ago and left an e-mail (not phone # because i wanted preliminary info on pay & hours available first) but was never contacted.
The appeal for video trackers is no longer on the front page (dropped sometime in last month I think) and I didnt see anywhere else either.
...PER is not really a defensive stat. Its a production stat that equates both ends of the floor.
I am not well versed in the underlying theories behind PER, but looking at the formula at it seems that the defensive component is quite minor.
A quick check shows the last five DPOYs average a modest PER of 17.8, while the last five scoring leaders all have much higher PERs, averaging 25.8. That suggest that - like most "all in one" formulas - PER reflects the heavy bias toward offense that is inherent in the collection of individual player statistics. That's not intended as a criticism of the PER "formula." I just don't see how PER "equates" offense and defense.
The concept of "Opponent Counerpart" PER seems like a logical attempt to deal with this offense/defense imbalance. Unfortunately, I think the implementation is too flawed to produce useful results.
All times are GMT - 5 Hours Goto page 1, 2, 3, 4Next
Page 1 of 4
You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum
FAQ
Search
Memberlist
Usergroups
Register
Profile
Log in to check your private messages
Log in
