76 replies to this topic

[TC-GOneBLUE]

i think alot of the the blow back is from "the stats say" which implies a conclusion when infact we should be watching where the stats are trending... even then, whether four games is enough to note a trend is debatable to me.

 

when i see an incomplete graph, i think.. i wonder where that point will be as t increases. and make a prediction or decide on possible outcomes. but not reach a conclusion.

[j_wehr]

How many of you would be comfortable betting that Dwight will be playing at Orlando level by the All Star break? We could define that as at least two of the following three things being true...

PER at or above 24
WS/48 at or above 0.22
RAPM at or above +8

Personally, based on the Simmons/Winston big-man theory, the eyeball test, and the stats to date, I would be very comfortable betting against that. But I'm curious how many of you are optimistic enough to take that bet.

[Rahat Huq]

I'd side with you on that - I don't see his PER ever climbing back to that level.  But to that end, I'm just not a fan of PER at all as I don't think it does much to capture his impact.  I think the best metric we have available is RAPM...and if we're betting on that, I also don't think he'll get back to his Lebron-level measures from his Orlando days.  But he'll look better than PER is showing him to be.  I think. 

[Mason Khamvilay]

I'd side with you on that - I don't see his PER ever climbing back to that level.  But to that end, I'm just not a fan of PER at all as I don't think it does much to capture his impact.  I think the best metric we have available is RAPM...and if we're betting on that, I also don't think he'll get back to his Lebron-level measures from his Orlando days.  But he'll look better than PER is showing him to be.  I think. 

I agree completely. I'm kinda hoping Dwight proves me wrong though and does go back to his Lebron-level measures. 

[Chichos]

How many of you would be comfortable betting that Dwight will be playing at Orlando level by the All Star break? We could define that as at least two of the following three things being true...

PER at or above 24
WS/48 at or above 0.22
RAPM at or above +8

Personally, based on the Simmons/Winston big-man theory, the eyeball test, and the stats to date, I would be very comfortable betting against that. But I'm curious how many of you are optimistic enough to take that bet.

It's too safe to bet the under on his Orlando numbers as they were so insanely high.  But what if you took 10% off them...

 

You would be betting he can get to

PER 21.6

WS/48 .198

RAPM 7.2

 

That is a much more interesting bet.  I still think he falls just short of those numbers.  But I believe he will get very close. 

[Johnny Rocket]

Justin--I really appreciate the article and your follow up, as it is a classic case that sometimes the eye test (lots of observers say that Howard "looks" better this year) differs a lot from the statistical test (especially the PER numbers).  I'm stuck on the sample size issue, though.  Are you really saying that superstars like James and Paul don't have five-game slumps in which their numbers are significantly below season and career averages?  I'm willing to believe it, but my guess would be that even these elite performers have slumps, and it is a possibility that Howard is going through one such slump right now (though it is nice to have a PER of almost 20 and be in a "slump").  It would be really interesting to examine the consistency of elite players, and how much their numbers vary over a season.

[thejohnnygold]

It's a good point.  Wasn't it last February that Lebron basically said, "Eff you" to the laws of physics and averages by posting the single most efficient stretch in basketball history?  Here is the LINK to Lebron's splits last season.  Scroll down to the monthly splits and let that all soak in.  I don't think there is any question he dominated the league last season.  To relate it back to this topic, we can look at his first and last month's stats--a total of 5 games.  Clearly, those stats deviated from his average, but the thing is we can only see that in hindsight.  Alas, the same is true here.  I say the odds are Dwight's final averages will be very good, but we will just have to wait and see how it all plays out.

[NorEastern]

Here I am again arguing against small sample size. Statistics does have some tools that can be used to handle small sample size. But they are generally reserved for experimental data where collecting more data is extremely difficult or expensive. Examples would be Stage 2 drug studies and MRI images. Also the confidence interval of these statistical analysis methods generally are 20-30% or more.

How I hate PER. Let me count the ways. Never mind.



How do you account for the 6th, 7th, 9th, 10th, 11th ... ranked players according to PER? Well at this point we all just point a finger to "small sample size". That cuts both ways, and certainly will apply to Harden and Howard at this point of the season.

Will Howard's advanced stats return to their glory days like in Orlando? Probably not. But for a different reason that most believe. In Orlando Howard was "the MAN". That is just not going to happen Houston. The likeliest stat to approach the 2008-2009 is RAPM because of the way it is calculated.

Edited by NorEastern, 07 November 2013 - 11:51 PM.

[j_wehr]

Chichos, I thought I was already being kind of conservative by putting the numbers where I was. I was using the minimum of his last few years in Orlando rather than the average. He had about 4 years where he performed above those stats, so I don't know why people wouldn't want to take that bet if they truly thought he was healthy. It's not like he's old. This should be the prime of his career.

 

Johnny Rocket, great reply, and I'm glad you pushed back on that. Here's how I'd put it: The best four or five players are so much better than ordinary players that they are rarely going to have a five-game stretch where they look statistically ordinary. They will regulary have five- or even twenty-game stretches where they perform well below average FOR THEM, but rarely will they go so low to cross into the "ordinary" realm. You're right that it would be very interesting to look at the data to see just how rare that is, but I'm confident that it is in fact rare. It's the nature of outliers. (Unfortunately my database is season-level rather than game-level and it would take a long time to collect, clean, and analyze game-level data.) What this all means for the Dwight discussion is that it's probably pretty unlikely that Dwight is still one of the best 4-5 players and that he just happens to be in an early season slump. I think it's more likely that he's just not going to be as dominant as he was.

 

Rahat, I know your comment on PER was kind of an aside, but I wanted to respond anyway. I can understand the distrust of PER since it only uses box score data and so pretty much ignores defense. I think this oversight is particularly salient for us Rockets fans since Asik looks like just an average player according to PER. But it's important to note that Asik is one of the biggest statistical enigmas in NBA history. He and Ben Wallace are the only players in my database who have HOF-level career RAPM while looking barely above average in career PER. Historically, the correlation between RAPM and PER is something like 80%. So this is all just a long-winded way of saying that I think PER is a fine metric if you just keep in mind that Asik is really weird.

 

NorEastern, being "the Man" is not going to affect these metrics much. They are measures of efficiency more so than total production. As one example, LeBron's Usg% dropped after he joined Wade and Bosh in Miami, but his PER and WS/48 did not. 

 

Also, it looks like you might have missed my earlier comment on sample size, so I'll paste it again:

 

Re: small sample size, all of the "elite" players that Dwight gets mentioned with -- Lebron, CP, KD, Love, Curry -- have a PER above 26 already, and they all have an even smaller sample size than Dwight because they played fewer games. With small sample sizes, the issue is that not-so-elite players temporarily look elite (e.g., CJ Miles, Zaza Pachulia), but there isn't so much of an issue with the reverse: in other words, you don't usually see that elite players temporarily look mediocre. The LeBrons of the world statistically float to the top pretty quickly. Dwight isn't there yet, and there's a good chance that means he won't be getting there anytime soon.

[NorEastern]

Harden is not there. Is there a "good chance that means he won't be getting there anytime soon"?

[NorEastern]

I have considered this for hours, but I have to say Mr. Wehr that you do need to examine your understanding of statistics. I pointed out that almost half of the top 20 PER leaders have no business on that list. That does not trigger "insignificant data" alarm bells in your head? At that point most statisticians just throw up their hands and walk away from the data. Nothing to be seen there. It is vitally important that statistical examinations of data err on the side of caution. Small sample size is just the most basic and trivial red flag.

[Mason Khamvilay]

I have considered this for hours, but I have to say Mr. Wehr that you do need to examine your understanding of statistics. I pointed out that almost half of the top 20 PER leaders have no business on that list. That does not trigger "insignificant data" alarm bells in your head? At that point most statisticians just throw up their hands and walk away from the data. Nothing to be seen there. It is vitally important that statistical examinations of data err on the side of caution. Small sample size is just the most basic and trivial red flag.

I've always felt the same way. It should be brutally obvious that small sample sizes exaggerate effects, so obvious it should go completely without saying. 

[NorEastern]

I did not post that lightly. It was unfortunate that I felt that it was necessary. Mr. Wehr could be a valuable addition here, if only he would acquire a bit of perspective.

[j_wehr]

Small sample size = higher variability = less precision = less trustworthy results. Of course. No where did I disagree with that. 

 

But then there's the radical view, which some of you seem to hold, ...

 

Small sample size = completely meaningless = run away.

 

People make the same mistake with "correlation does not equal causation." They think that means that correlations are meaningless. Fun way to intimidate a grad student during an interview: Ask them, "Doesn't the absence of correlation imply the absence of causation?"

 

The more important point, which no one other than Johnny Rocket has yet latched on to, is this: "With small sample sizes, the issue is that not-so-elite players temporarily look elite (e.g., CJ Miles, Zaza Pachulia), but there isn't so much of an issue with the reverse: in other words, you don't usually see that elite players temporarily look mediocre." It's a hypothesis and it may be wrong, but I'm pretty confident in it for reasons I gave in my reply to Johnny Rocket.

 

---

 

NorEastern, Harden's PER is right in line with the last two years (despite his unusually poor 3-point shooting thus far), so I don't see what you're trying to get at.

 

P.S. - This goes to everyone but especially NorEastern: When you disagree with me, please try to disagree with my statements rather than criticizing my "knowledge" or "understanding." It's really irritating when you make it personal. I do not want the discussion to devolve into citing credentials in some sort of power play.

[NorEastern]

"Throws hands up in the air". But I disagree with the basic precepts of your analysis. A regression to the mean analysis would help your argument immensely using the last 3 years of data (like RAPM) but that would hurt your Howard case. I am holding off from any analysis of this young season because I just cannot accept PER data that lists Jordan Hill as having the 9th best PER in the league as of early yesterday evening. "Alarm bells go off". When a data set has demonstrable flaws any analysis of that data set will by extension be flawed. When approximately half the data set used is obviously flawed is there anything of value to be garnered from an analysis of that data? Perhaps, but the confidence interval would be atrocious.

Edited by NorEastern, 08 November 2013 - 02:20 PM.

[j_wehr]

You aren't disagreeing with any of my statements. You're just continuing to cite the doctrine of small sample size.

I also find it amusing how you are so high on RAPM and so critical of PER. Like I said, career PER and career RAPM have something like an 80% correlation. How do you respond to that? And how do you explain Michael Jordan's RAPM? Doesn't that put you into a bind if you believe that "When a data set has demonstrable flaws any analysis of that data set will by extension be flawed."

[timetodienow1234567]

"Throws hands up in the air". But I disagree with the basic precepts of your analysis. A regression to the mean analysis would help your argument immensely using the last 3 years of data (like RAPM) but that would hurt your Howard case. I am holding off from any analysis of this young season because I just cannot accept PER data that lists Jordan Hill as having the 9th best PER in the league as of early yesterday evening. "Alarm bells go off". When a data set has demonstrable flaws any analysis of that data set will by extension be flawed. When approximately half the data set used is obviously flawed is there anything of value to be garnered from an analysis of that data? Perhaps, but the confidence interval would be atrocious.

Are you saying stats are worthless? because ALL advanced stats are flawed. Garbage In Garbage Out. Let's just stop using stats altogether and go back to the eye test then.

[thejohnnygold]

I'm not going to say we're off topic.....but are we still talking about basketball here?

 

NorEastern and Wehr, if you guys wish to debate the merits of each others' statistical approach and ability I think it would be better if it was done by PM.  This is especially true if your intention is to get a bit personal.

 

Now, back to this statistical stuff.  I am in agreement with both parties here.  Wehr, I agree that the outliers exist on both sides--this is why I have not dropped certain players from my fantasy roster for their poor performance and also why I haven't added others for their good performance.  I believe that when the dust settles everything will be in its right place.

 

As far as advanced metrics go, I dislike PER less than RAPM.  This is because PER does not attempt, nor claim, to measure what it cannot.  I understand that it ignores defense.  I can use my own knowledge to fill those gaps.  Those who have been around here a while know how I feel about RAPM.

 

As far as this small sample size thing goes.  It would never need to be said if people weren't constantly trying to force their opinions down others throats with them.  We all know the effects of small samples; yet, we don't all exercise discretion when using that data.  I believe Mr. Wehr is fully aware of this, but loves numbers and is using what he's got to give us something to think about and talk about--I appreciate that.  Just take it with a grain of salt and let's all try to be more constructive with these by contributing and building on ideas rather than just tearing everything to shreds.

 

Here's something I'd love to see some info on if anyone wants to take on the task:  I would like a measure of how much time Harden spends ball-watching on defense and not paying attention to his man paired with a direct correlation to shots taken by his man once he loses him and the TS% on those shots.  You're going to need a stopwatch....

[j_wehr]

JohnnyGold, thanks for the thoughtful reply. Just to add to that: NorEastern, I'd request that you either PM me or refrain from commenting on my posts at all if your intention is only to make vague claims about the integrity of my analysis or my personal credentials.

To bring it back, I think this might be where people are misinterpreting what I'm saying, or I might not be explaining myself clearly. I am not suggesting that we should expect Dwight's PER (or other stats) to be where it is now by the end of the season or by the All-Star break. It could easily move 3 points in either direction. But what seems unlikely is that it will jump all the way up to Orlando levels, which provides an EARLY but strong indication that Dwight is not back to his normal self, which is contrary to what media reports have been saying. If anyone would like to provide evidence against that argument, please do!

[Mason Khamvilay]

As far as advanced metrics go, I dislike PER less than RAPM.  This is because PER does not attempt, nor claim, to measure what it cannot.

 

I could be wrong here, but I think what RAPM actually claims to measure and what YOU think RAPM claims to measure are two different things.