Kawhi is a slightly above average defender this season...

[dabom]

I was saying I was wrong to the "if" part. Not your stance.

[Chinook]

It never has. It's why players like Kobe, Iverson, et al rarely ranked among the league's best (Top 5) in PER despite being "MVP Level" players.

I don't mean chuckers like guys who take a bunch of horrible shots. I mean it like guys who dominate the ball at best, it's a rate stat measure quan y of production with no comment on quality. That's why along with guys like Harden and Russ who dominate the ball, we see guys like Boban be up there as well.

You don't need to know the math to know it's flawed.

You kinda do need to know what it's trying to say before you can say that it's wrong. As I said, it's possible that RPM is being marketed as being stronger than it is, but that's not the fault of the stat itself, only the people misusing it.

I'm okay with it punishing players like Korver. He's a low-volume scorer who was assisted on .84 percent of his shots through his career.

Oh, so NOW assists are important. Rewarding or punishing are both wrong-headed words. It's not capturing how much Korver actually affected his team's offense, which means that PER can't really be used to measure offensive effectiveness.

Link to 82games methodology in this case?

No idea why they do it. But you can see it in their tables.

It infers an outcome from something that hasn't even happened.

That sounds more like your assumption that the model's. It could well be controlled for role, meaning that it's not arguing who is better but only who plays better in their role. Like maybe it suggests that Korver was a more elite spot-up guy than Batum was a lead perimeter creator. Or that Boban was a better garbage-time big than LMA was a front-court anchor. That's a perfectly reasonable interpretation of that data -- it's just weaker to the point that it would have to be part of a larger argument.

That said, individual defense will probably forever remain unmeasurable because of how team/lineup dependent defensive effectiveness is.

Eh, plus-minus with huge sample sizes will always be the best way to measure defensive effectiveness. There's a level a nuance it won't be able to catch, but you are what your on/offs say you are eventually, no matter how you try to cut the data to argue otherwise.

[dabom]

I rate a players worth on several stats. Not just one. RPM is one of those stats I use. When using multiple stats I can make an educated assumption on a players worth. What is it you want to debate OP.

[midnightpulp]

I rate a players worth on several stats. Not just one. RPM is one of those stats I use. When using multiple stats I can make an educated assumption on a players worth. What is it you want to debate OP.

RPM underrates Kawhi.

Whole point of this thread.

[dabom]

RPM underrates Kawhi.

Whole point of this thread.

You said Kawhis defense has slipped. Explain more.

[midnightpulp]

I don't mean chuckers like guys who take a bunch of horrible shots. I mean it like guys who dominate the ball at best, it's a rate stat measure quan y of production with no comment on quality. That's why along with guys like Harden and Russ who dominate the ball, we see guys like Boban be up there as well.

Guys who can create for themselves or for their teammates over multiple possessions are very valuable now, especially in the playoffs where offenses tighten up due to pressure. Counterintuitively, less ball movement/less plays can be a good thing. Relatively efficient ball dominant players have never really been a detriment: Magic, Jordan, Lebron, and even the much maligned Westbrook (he may shoot 8-23, but he'll likely go 14-16 from the stripe).

You kinda do need to know what it's trying to say before you can say that it's wrong. As I said, it's possible that RPM is being marketed as being stronger than it is, but that's not the fault of the stat itself, only the people misusing it.

I just don't see how you can extrapolate garbage time/limited minutes performances over 48 minutes with any sort of accuracy when the sample sizes are marginal at best.

Oh, so NOW assists are important. Rewarding or punishing are both wrong-headed words. It's not capturing how much Korver actually affected his team's offense, which means that PER can't really be used to measure offensive effectiveness.

I never said assists weren't important, just that they're an overrated stat. A guy shoveling the ball off 2 feet to an open player who makes a 24 foot shot is ranked "equally" as a guy who actually creates the shot for his teammate. Assists are an "eye test" stat. And Korver needed to be created for, which was my point. I don't think PER underrates him all that much. He's an average/slightly above average player who fit a specific role for his team. He's not a plug-and-play player who could fit into multiple systems. You see it all the time with spot up shooters. They have a great year, sign a contract to another team, and then under perform.

That sounds more like your assumption that the model's. It could well be controlled for role, meaning that it's not arguing who is better but only who plays better in their role. Like maybe it suggests that Korver was a more elite spot-up guy than Batum was a lead perimeter creator. Or that Boban was a better garbage-time big than LMA was a front-court anchor. That's a perfectly reasonable interpretation of that data -- it's just weaker to the point that it would have to be part of a larger argument.

I have no problem if that's what it is doing.

Eh, plus-minus with huge sample sizes will always be the best way to measure defensive effectiveness. There's a level a nuance it won't be able to catch, but you are what your on/offs say you are eventually, no matter how you try to cut the data to argue otherwise.

I disagree here. I agree if a team only had one bench player, but I don't see how you can isolate individual player +/- when players often check into the game with other players while the opponent is also shuffling their lineups around.

On/off champ and forum favorite Patty Mills has nice numbers here, but when he's subbed in straight up for Parker, the results aren't as good. But too many people read the stat in a vacuum and think a player would always be a net positive/negative regardless of lineup and other game conditions.

[Perry Mason]

I rate a players worth on several stats. Not just one. RPM is one of those stats I use. When using multiple stats I can make an educated assumption on a players worth. What is it you want to debate OP.

Garbage in, Garbage out. 3 inaccurate stats doesn't make the stats automatically accurate. I too am highly skeptical of the Bayesian and other inference stats that "adjust" for teammates, etc, to come up with a true impact or +/-. The more complicated the math, the more skewed results can occur, and then when even the formula (like ESPN's) is hidden, you cannot even deduce why the stat is having a bad result. It's bizzarre and faith-based only.

All of the impact stats, which can be interesting, are still circular because they are based on relative performance only. And no matter how hard they try, you can't just isolate the player from the lineup on the floor and infer the strength of that lineup.

[Perry Mason]

RPM underrates Kawhi.

Whole point of this thread.

Lol. Confirmation that Dabom doesn't read. He spent this whole thread thinking you were criticizing Kawhi.

[midnightpulp]

You said Kawhis defense has slipped. Explain more.

His counterparts are slightly more productive this season than last season. Eye test: He doesn't crash the boards as much and doesn't contest with as much "ferocity" on every play like he usually does, probably out of energy concerns. But his offense has surged, so it's all moot.

RPM, however, says his defense has tanked.

[midnightpulp]

Garbage in, Garbage out. 3 inaccurate stats doesn't make the stats automatically accurate. I too am highly skeptical of the Bayesian and other inference stats that "adjust" for teammates, etc, to come up with a true impact or +/-. The more complicated the math, the more skewed results can occur, and then when even the formula (like ESPN's) is hidden, you cannot even deduce why the stat is having a bad result. It's bizzarre and faith-based only.

All of the impact stats, which can be interesting, are still circular because they are based on relative performance only. And no matter how hard they try, you can't just isolate the player from the lineup on the floor and infer the strength of that lineup.

This guy gets it.

[Cry Havoc]

I don't mean chuckers like guys who take a bunch of horrible shots. I mean it like guys who dominate the ball at best, it's a rate stat measure quan y of production with no comment on quality. That's why along with guys like Harden and Russ who dominate the ball, we see guys like Boban be up there as well.

PER has a known bias against high usage rate guys.

But I don't think this is an error calculating guys like Boban high. That's basic statistical variance when you have low sample sizes. It is practically impossible to account for extremely low minute players, because it's nearly impossible to extrapolate how their play would translate to a full 48 from a purely mathematical standpoint. Of course if a guy comes in to play garbage time and scores 8 points in 4 minutes against scrubs, no one would say, "hey he should be a starter" (well maybe a few would ) but there's no way to translate that to a predictive model of future play. Statistics struggle with low amounts of information (like Boban's limited play) but that's not a flaw in the algorithm, that's a simple lack of data.

You kinda do need to know what it's trying to say before you can say that it's wrong. As I said, it's possible that RPM is being marketed as being stronger than it is, but that's not the fault of the stat itself, only the people misusing it.[/quote]

You don't need math to know that it's wrong, but you need it to explain why it's wrong.

Oh, so NOW assists are important. Rewarding or punishing are both wrong-headed words. It's not capturing how much Korver actually affected his team's offense, which means that PER can't really be used to measure offensive effectiveness.

PER can and is used frequently, that's why we see it quoted all the time. But like everything else it's not a perfect stat, and Hollinger said as much from day one.

Even baseball, the king of sports information, does not have perfect statistics to measure player impact. They are certainly closer than hoops, but not perfect.

Eh, plus-minus with huge sample sizes will always be the best way to measure defensive effectiveness. There's a level a nuance it won't be able to catch, but you are what your on/offs say you are eventually, no matter how you try to cut the data to argue otherwise.

I wouldn't say that's a guarantee by any stretch. It's true that the envelope for inaccuracy gets narrower as you gain more information, but like everything else, it too is flawed. For instance, Marcus Camby is rated as #3 all time in DBPM. He was not a better defender than DRob (#4) or Tim (#6). FWIW, I know Bo Outlaw was no slouch, but he checks in at #8, above Hakeem at #11. All of these players have more than enough sample size to account for error.

[TheGreatYacht]

Microwave Paddy is, in fact, the worst defender on the team. Anyone with eyes can see this

[Cry Havoc]

Guys who can create for themselves or for their teammates over multiple possessions are very valuable now, especially in the playoffs where offenses tighten up due to pressure. Counterintuitively, less ball movement/less plays can be a good thing. Relatively efficient ball dominant players have never really been a detriment: Magic, Jordan, Lebron, and even the much maligned Westbrook (he may shoot 8-23, but he'll likely go 14-16 from the stripe).

WB had 10 ing turnovers last night.

I just don't see how you can extrapolate garbage time/limited minutes performances over 48 minutes with any sort of accuracy when the sample sizes are marginal at best.

Statistically it's not possible by definition, at least with the stats that we currently use in the NBA. If there WERE a stat that would tell you which bench guys would be monsters if given more minutes that passed a p-value muster, there would be an NBA job waiting for you tomorrow.

I never said assists weren't important, just that they're an overrated stat. A guy shoveling the ball off 2 feet to an open player who makes a 24 foot shot is ranked "equally" as a guy who actually creates the shot for his teammate. Assists are an "eye test" stat. And Korver needed to be created for, which was my point. I don't think PER underrates him all that much. He's an average/slightly above average player who fit a specific role for his team. He's not a plug-and-play player who could fit into multiple systems. You see it all the time with spot up shooters. They have a great year, sign a contract to another team, and then under perform.

It's sort of unfortunate that the playoffs are so relatively short, because I think as defenses tighten you see some of this noise get sorted. But then, we don't have enough data in the playoffs to really have a certainty of anything, so we go back to the eye-test.

On/off champ and forum favorite Patty Mills has nice numbers here, but when he's subbed in straight up for Parker, the results aren't as good. But too many people read the stat in a vacuum and think a player would always be a net positive/negative regardless of lineup and other game conditions.

People also literally think they know basketball better than NBA head coaches on this forum.

[midnightpulp]

WB had 10 ing turnovers last night.



Statistically it's not possible by definition, at least with the stats that we currently use in the NBA. If there WERE a stat that would tell you which bench guys would be monsters if given more minutes that passed a p-value muster, there would be an NBA job waiting for you tomorrow.




It's sort of unfortunate that the playoffs are so relatively short, because I think as defenses tighten you see some of this noise get sorted. But then, we don't have enough data in the playoffs to really have a certainty of anything, so we go back to the eye-test.



People also literally think they know basketball better than NBA head coaches on this forum.

[dabom]

Try evenings OP if you make threads directed at me. Can't be replying during work or lunch. Got better to do.

[spurraider21]

I'm actually Jewish.

when did you renounce jehova's witness?

[Chinook]

Guys who can create for themselves or for their teammates over multiple possessions are very valuable now, especially in the playoffs where offenses tighten up due to pressure. Counterintuitively, less ball movement/less plays can be a good thing. Relatively efficient ball dominant players have never really been a detriment: Magic, Jordan, Lebron, and even the much maligned Westbrook (he may shoot 8-23, but he'll likely go 14-16 from the stripe).

And things like gravity count even more in the playoffs than they do in the regular season. The reason why most elite spot-up guys see their numbers dip in the post season is because teams make stopping them a greater emphasis. So when you have guys like Korver sprinting around the arc, you better believe that the Teague's of the world have easier driving lanes, if you want to give Jeff credit for "creating" when he passes out to Korver when the defense breaks down.

I just don't see how you can extrapolate garbage time/limited minutes performances over 48 minutes with any sort of accuracy when the sample sizes are marginal at best.

You go into this a bit more later, but RPM doesn't extrapolate , as far as we know -- people who use it do or might.

I never said assists weren't important, just that they're an overrated stat. A guy shoveling the ball off 2 feet to an open player who makes a 24 foot shot is ranked "equally" as a guy who actually creates the shot for his teammate. Assists are an "eye test" stat. And Korver needed to be created for, which was my point. I don't think PER underrates him all that much. He's an average/slightly above average player who fit a specific role for his team. He's not a plug-and-play player who could fit into multiple systems. You see it all the time with spot up shooters. They have a great year, sign a contract to another team, and then under perform.

The ability for a team to leverage a player's skill-set to help their team goes far beyond a player's ability to play a very specific style that happens to fit on most teams. These are supposed to be stats that measure impact and performance, even in your mind. So if Korver helps his team more than Teague, why would he not have higher stats. If anything PER just shows who are replaceable, since most guys with high PERs play the same style.

I have no problem if that's what it is doing.

Well, that's what the math is for, and that's why marketing and application matter.

I disagree here. I agree if a team only had one bench player, but I don't see how you can isolate individual player +/- when players often check into the game with other players while the opponent is also shuffling their lineups around.

This is exactly why sample size is important. Once you get enough data in those situations, you can parse out the effect that lineup has. RAPM has a similar flaw if taken to the extreme, but it's a different stat in greater samples because most of those correlations get reduced due to injuries, shuffling rotations and the number of different lineups faced. The smaller the rotations, the worse the bias would be.

On/off champ and forum favorite Patty Mills has nice numbers here, but when he's subbed in straight up for Parker, the results aren't as good. But too many people read the stat in a vacuum and think a player would always be a net positive/negative regardless of lineup and other game conditions.

Yet Parker actually has better on/offs than Mills does with the starters. So at best it seems like a data-cutting issue.

[Horse]

Kawhi is the leagues # 1 defender until further notice.

[SpurPadre]

Are you this re ed or just being Mexican?

Really dude? You had some kind of respectability trying to have an interesting basketball discussion and then you go this route? smh

[GSH]

I actually just read your entire post, fair analysis..although I still don't see why this needed a thread.

My first thought was, "Haven't we had this exact same thread under different les?" Different flavor of the same ice cream.

[GSH]

PER has a known bias against high usage rate guys.

Yep. Stats without context can be as bad as no stats at all.

It is practically impossible to account for extremely low minute players, because it's nearly impossible to extrapolate how their play would translate to a full 48 from a purely mathematical standpoint.

Not to mention the fact that there are "energy guys" who come off the bench and change the momentum. They benefit from the abrupt change of pace, and the fact that they are usually playing against other bench players. The idea that they would continue to pile up numbers at the same rate if they were playing starter minutes, against starters? Too ridiculous to even take seriously. Once again, stats without context can be worse than no stats at all.

[HarlemHeat37]

1. PER is not a ty stat, but it's antiquated and virtually useless for determining the contribution that a player made to the success of his team. It's a summary stat that describes the contributions of a player as an individual, completely ignoring the actual impact a player has on team success.

It's inherently biased against low-usage perimeter players and generally rewards high-usage guys. It's literally meaningless when assessing role players, particularly role players. Somebody like Bruce Bowen is categorized as a D-league caliber player by PER due to his role and usage rate, for example(whereas RPM consistently rated Bowen as an above average player). The reason PER gained recognition and credibility is because it became popular at a time where we didn't have other advanced metrics, nor did we have tracking websites and the massive amount of data we have today.

2. I don't believe 82games actually tracks assigned position data. They simply decide what position a player is playing and evaluate the opposition at the same position IMO. They have been around since 2001(same formula), way before the public ever had access to tracking data or mass access to games, I find it very difficult to believe that they had somebody watching every single game, every single night and assessing assignments(the website has always been updated nightly, so ya, it would be virtually impossible to watch entire games every night when they only had 2 employees in 2002). Their numbers are also much closer to hoopsstats than they are to Synergy, which makes sense.

3. RPM, BPM and other on/off numbers are far more accurate than the other advanced metrics IMO. Their accuracy is based on the person using them correctly.

First of all, I don't believe any of them should be used to separate offense and defense. As proven, the results are very skewed when isolating the numbers. In basketball, there is too much action that occurs between a half court offensive possession and a half court defensive possession to separate the numbers between offense and defense. On/off numbers are far more accurate and logical when used as a whole, rather than separating offense vs. defense.

Regarding the actual evaluation of the numbers, I've never interpreted them as X player is more important than Y player because his RPM is higher. Even when adjusted to factor teammates and minutes, it should still be interpreted as X player has been more important at playing his role compared to Y player in his role(Patty Mills is not a better player than Parker, nor does it mean he would fit better with the starters. But Mills has been more valuable in his role at backup PG than TP has been in his role as starter/lead facilitator).

The numbers are falsely interpreted when somebody looks at a list and says "Kyle Anderson is higher than Carmelo Anthony, he must be a better player and more valuable in his role". That isn't how it works.

4. Defense is far easier to measure than in previous eras. Adjusted on/off numbers are flawed, but they're clearly indicative of impact over a large sample size, especially if you're using a multi-season approach. More importantly, it's 2017, NBA.com and Synergy numbers provide us with mass amounts of thoroughly collected, descriptive data. Certain categories like "spotting up" will always be flawed since judging rotations and strategy is subjective, but "isolation", "pick and roll", "post up", etc are pretty black and white IMO.

5. A player can have relatively poor advanced metrics and still be a championship-caliber player in the correct role. Kyrie Irving is an example of this. Most on/off numbers had him as a net negative player for obvious reasons. He's a PG that doesn't make anybody better, horrible playmaker and defender. However, you put him on a team with Lebron, and he can excel at his specialty of 1-on-1 scoring when required, without needing to worry about the rest of the team(although this only works when the chucker is shooting efficiently, as Irving did in the playoffs).

Aldridge is another example, he's widely hated by the metrics community. His numbers are relatively poor for obvious reasons. He doesn't make teammates better or have any notable effect on the other players on the floor.

Both these players have high PERs, though, because they are great individual players.

6. Damn, Patty Mills and Parker have caused a crusade against numbers on ST the past month

[DPG21920]

I got her numba. Howdoyou like dem apples.

[Chinook]

Well Harlem just turned Mid's anus into a nuclear wasteland.

[GSH]

Well Harlem just turned Mid's anus into a nuclear wasteland.

LOL. I can't wait for the twisted-logic response. By the time he's done, he'll be arguing against Harlem never said, and claiming he was right all along.

Maybe I'll add him to my sig.