Iverson's career stats

[xkonk]

@xkonk

Iverson's +/- playoff numbers suffer hugely because especially in Philly he played almost 95% of available minutes in playoffs. He usually came out for a min or two in a game and that small sample size taken from a game skews +/- numbers. The only playoff game 76ers played without him in a series that they eventually won, was a lose. But that doesn't mean anything at all like his playoff +/- numbers.

That's a valid point. Although it hasn't really hurt LeBron, who never played fewer than 86% of his team's available playoff minutes until last year, or Durant, who's been right around 90% every year except for his first trip to the playoffs.

We can also argue all day about who had worse back-ups and teammates, but it makes you wonder why the 76ers were so unsuccessful when they had the second-best player of all time playing 95% of their minutes. People complained about LeBron's supporting cast every year he was in Cleveland, but they never won fewer than 35 games while he was there, and once they made the playoffs they never lost in the first round. The few apologists LeBron had after the Decision said that he had to leave because his teammates would never be good enough to get a title, but he still got them to back-to-back 60 win seasons. Iverson's 76ers were never that successful; even when they made it to the Finals they only won 56 games.

[permaximum]

@xkonk

Iverson's +/- playoff numbers suffer hugely because especially in Philly he played almost 95% of available minutes in playoffs. He usually came out for a min or two in a game and that small sample size taken from a game skews +/- numbers. The only playoff game 76ers played without him in a series that they eventually won, was a lose. But that doesn't mean anything at all like his playoff +/- numbers.

That's a valid point. Although it hasn't really hurt LeBron, who never played fewer than 86% of his team's available playoff minutes until last year, or Durant, who's been right around 90% every year except for his first trip to the playoffs.

We can also argue all day about who had worse back-ups and teammates, but it makes you wonder why the 76ers were so unsuccessful when they had the second-best player of all time playing 95% of their minutes. People complained about LeBron's supporting cast every year he was in Cleveland, but they never won fewer than 35 games while he was there, and once they made the playoffs they never lost in the first round. The few apologists LeBron had after the Decision said that he had to leave because his teammates would never be good enough to get a title, but he still got them to back-to-back 60 win seasons. Iverson's 76ers were never that successful; even when they made it to the Finals they only won 56 games.

Well, LeBron is better if we're talking about their whole careers. Iverson in his prime is better than LeBron, especially in playoffs but evaluating whole careers is something else. I would rank Iverson's whole career somewhere between 8-15.

If we're looking at whole careers instead of players' absoIute primes, I like both the methodology and the outcome of VORP. The best metric out there atm imo (RAPM has too much noise and it needs to be stabilized) and I generally agree with it more or less. You can take a look at it here. Interestingly, I seem to like anything that come out of this site http://godismyjudgeok.com/DStats/aspm-a ... -nba-aspm/

[xkonk]

Going by that very link, Iverson's best year was at a 6.5 VORP, good for 67th best season all time. LeBron has only had one season lower than that. So I guess I'm a little confused. Do you not like VORP at the single season level?

[permaximum]

Minor thing, his best is 64th not 67th. As for your question, I like it at single season level too. And I can agree that his best regular season may rank 64th all time because MJ has lots of great regular seasons. In that single season list. he's the 21st player. Not that different from his 18 all-time (tie with Clyde) VORP rank according to his regular-season career.

Unfortunately that data doesn't include playoffs. Iverson's playoff VORP should rank even upper. I don't think Lebron's playoff VORP surpasses Iverson's. If it's, it will fall into the error margin of the metric lol

Edit: Does anyone remember, someone from this forum reviewed all these metrics (RAPM, PER, WS, WP, ASPM etc.) and eventually DSmok1's model was found as the best at explaining what happened before? I hope I'm not imagining things.

[Mike G]

... there's a lack of ability in today's players to take over in a playoff game where it matters. .. Yes, I rank prime Iverson second behind prime Jordan.

Between the years of 1999 and 2005 (Philly playoff years), Iverson played 62 playoff games and scored 40+ in 10 games (9-1 record...
What I'm trying to say is this is what "greatness" is about. Winning the games that matter without getting much help.

Iverson in his prime is better than LeBron, especially in playoffs...

Are playoff games "the games that matter"?
Iverson teams were 30-41 in playoff games. Just 1-8 in Den, after 29-33 in Phl.
His Sixers went 16-12 in first-round games, 13-21 thereafter.

Iverson's shooting% and per-game averages in each playoff round in Philly:

Code: Select all

Rd  G   2fg%   3fg%  3fgRt  eFG%    FT%    TS%       Min    TSA    Pts   Reb   Ast   Stl   TO
1  28   .444   .358   .22   .465   .753   .518      44.7   29.5   30.6   3.6   6.3   2.4   3.4
2  23   .407   .315   .18   .419   .784   .478      46.0   30.9   29.6   4.3   6.2   2.1   3.2
3   6   .347   .333   .19   .376   .782   .435      45.7   35.0   30.5   4.8   6.8   2.2   3.2
4   5   .447   .282   .24   .441   .729   .486      47.4   36.6   35.6   5.6   3.8   1.8   2.4

We see odd things vs specific opponents. His one conference final was vs the Bucks, and he had a string of terrible shooting nights.
Vs the Lakers in the Finals, he shot more and passed less, didn't turn it over much. Steals also were down.
In each round, he rebounds more. He was a 3.9 Reb/G guy in his Phl career (42 mpg).

[permaximum]

Well his teams were bad first of all...

Just like most of the players he had bad nights too but the difference was Iverson was one of those rhythm guys. He kept shooting to got into the rhythm since he was the focal point of offense in 76ers. When he didn't got into, his FG% suffered. Just like he scored 52 with 79% TS, he had 6/25 nights. The difference is 76ers wouldn't win those games at all if Iverson went 11/25 instead of 6/25. Just the margin would be lower. Also, shooting depends on many factors and a player that shot 40% may very well played great.

Found those articles about comparing different metrics on explaining the past and predicting the future. VORP(ASPM) looks like the clear winner. Most importantly it's methodology and reasoning suit my view on the matter perfectly. No surprise it's the only metric Iverson lead the league in 2001 where he had 93 first-place MVP votes out of a possible 124.

http://sportskeptic.wordpress.com/2012/ ... -happened/
http://sportskeptic.wordpress.com/2012/ ... the-goods/

[schtevie]

Found those articles about comparing different metrics on explaining the past and predicting the future. VORP(ASPM) looks like the clear winner. Most importantly it's methodology and reasoning suit my view on the matter perfectly. No surprise it's the only metric Iverson lead the league in 2001 where he had 93 first-place MVP votes out of a possible 124.

I think you should think a bit about confirmation bias here. As I read the linked blog posts, the "clear winner" was making the finest of distinctions. Over the four available, common years, I calculate an average difference of PPG error of 0.1, in ASPM's favor over "new" RAPM. And then what also ought to be recognized, as I understand the presentation, is that this is a summation of the contributions of five players on the court, so on a per player basis there is no difference.

If this interpretation is correct, it follows that this comparison offers no comment on the relative value of the competing metrics on the accuracy of the ratings of individual players. And for this, we need to dig a bit deeper. In particular, as for the specific value of Allen Iverson, we might ask whether one method or the other is apt to offer a more solid estimate.

On the one hand, Daniel's ASPM (and Daniel, please feel free to advise) offers a pared down regression ("Over the years, I have composed the regression a number of ways, and have gradually simplified the structure, minimizing terms to minimize over-fitting issues." http://godismyjudgeok.com/DStats/aspm-and-vorp/) what yields positive returns for both MPG and USG%. Fair enough. But note that there are theoretical reasons to believe that this such returns are not, in fact, positive on the extreme high end, and, sure enough, who is the extreme outlier in these categories? None other than AI. (From 1981 onward, he ranking 1st and 4th in these career statistics, for players playing more than 24 mpg in the latter case.)

Conversely, "new" RAPM, though it has a prior (the specifics of which I cannot recall - what might similarly exalt AI or not, on MPG and USG% grounds) also asks the contemporaneous +/- contributions to speak, and these offer a very different result, in terms of his rank. Again, as noted, his yearly best on J.E.'s site was 18th - what doesn't quite square, I think, with a 19th rank in notional HOF value.

So, my point is this: If one wishes to believe that AI is all that, and I'm just talkin' practice, so be it. But I don't think one should go to this specific rendition of ASPM to offer definitive opinion on a very outlying player.

[permaximum]

I tried to stay away from getting into the details but...

Over the four available, common years, I calculate an average difference of PPG error of 0.1, in ASPM's favor over "new" RAPM.

I calculate the average difference of R square error between them as 0.394. It's not that low. It matters quite a lot when you deal with 0.27 and 0.67.

And then what also ought to be recognized, as I understand the presentation, is that this is a summation of the contributions of five players on the court, so on a per player basis there is no difference. If this interpretation is correct, it follows that this comparison offers no comment on the relative value of the competing metrics on the accuracy of the ratings of individual players. And for this, we need to dig a bit deeper.

I don't really know what do you mean with that. If I guessed right, you mean he used point differential of teams as the response and sum of player contributions as the predictor. Do you know a better method to test for the accuracy of these metrics? There's nothing wrong with it. It actually confirms these metrics as reliable because individual sum of players more or less equal the margin. He could have done it per game basis, but that wouldn't be much of an improvement.

On the one hand, Daniel's ASPM (and Daniel, please feel free to advise) offers a pared down regression ("Over the years, I have composed the regression a number of ways, and have gradually simplified the structure, minimizing terms to minimize over-fitting issues." http://godismyjudgeok.com/DStats/aspm-and-vorp/) what yields positive returns for both MPG and USG%. Fair enough. But note that there are theoretical reasons to believe that this such returns are not, in fact, positive on the extreme high end, and, sure enough, who is the extreme outlier in these categories? None other than AI. (From 1981 onward, he ranking 1st and 4th in these career statistics, for players playing more than 24 mpg in the latter case.)

Conversely, "new" RAPM, though it has a prior (the specifics of which I cannot recall - what might similarly exalt AI or not, on MPG and USG% grounds) also asks the contemporaneous +/- contributions to speak, and these offer a very different result, in terms of his rank. Again, as noted, his yearly best on J.E.'s site was 18th - what doesn't quite square, I think, with a 19th rank in notional HOF value.

First of all there's a threshold in ASPM. You have to pass it first to get a positive value from usage. It's not that simple as you made it to be.

Usage*(points produced per possession – 1.47832) is the formula.

Second, Usage coefficient value is a very small number (0.01160). I don't think any player can get even 1 point of ASPM from increasing his usage to 100%. Don't worry, TS% is also in the euqation.

MPG is no different with a value of 0.08033.

So, my point is this: If one wishes to believe that AI is all that, and I'm just talkin' practice, so be it. But I don't think one should go to this specific rendition of ASPM to offer definitive opinion on a very outlying player.

It's not about AI. It's about using a metric that's more accurate if we don't value our own eyes because of a potential bias. Don't worry about "specific rendition of ASPM". ASPM ranks AI #20, VORP ranks him #19, Hall Rtg ranks him #19. All renditions of ASPM then I pressume . ASPM's #20 says a lot about how AI got his value. Surely not much from minutes played since VORP ranking is almost the same.

I once created a better and purely empirical version of PER using a very huge data I scraped from basketball-reference. Although it was definetely better than PER, defense and assists have been the problem. Decided simple-box scores are not enough. I quickly dismissed anything advanced box-score stat related because there were too much estimations and I hate estimations. Then I have calculated seasonal NPI RAPMs 2001-2008 myself (compared it with J.E's) and even used a very demanding LOOCV to get the best lambda to see if it helps the results. No luck. Then I calculated 2001-08 with one big regression (no age adjustment) and the results got only a bit better. Eventually I decided that RAPM noise cannot be reduced enough for an average NBA player's career time and there were other problems with +/- itself.. In the end I found myself thinking about something like DSMok1's ASPM and giving advanced stats an another go. His ASPM can be improved upon on certain points but I'm 100% convinced that it's the best metric out there atm. I even like it's variations VORP and Hall Rating more.

So, referring to ASPM is not about Iverson. It's because it's the best metric out there.

You didn't have any problem with an invalid (but reliable) metric such as WS or too noisy RAPM aka "Nick Collison metric" "Amir Johnson metric" goes on an on.

[schtevie]

A couple of points.

First, apologies. I was about to amend my remarks about USG% being the concerning factor in the AI rating, when it is, in fact, only the MPG margin - where, again, he ranks #1, post 1981. To contextualize the regression results (please check) they suggest that were AI to have played every minute of every game (assume 48.4) then he could have missed every shot and free throw (TS% = 0) and improved his rating by +0.4 points per possession. The point being: AI is an extreme outlier in this dimension (and in USG%) and his relative rank is accordingly sensitive to the specification of the regression.

Second, there is perhaps an interesting conversation to be had contrasting the fact that ASPM and "new" RAPM between 2008 and 2011 yielded identical PPG error though the former had higher R2, but the question remains: why should one think that the latter method - what has no particular bone to pick with AI (or any other player) and what directly incorporates on-court performance - would rank him decisively lower than the former? And might it be because the former gives positive weight to MPG, beyond what theory might suggest is justified?

[permaximum]

First, apologies. I was about to amend my remarks about USG% being the concerning factor in the AI rating, when it is, in fact, only the MPG margin - where, again, he ranks #1, post 1981. To contextualize the regression results (please check) they suggest that were AI to have played every minute of every game (assume 48.4) then he could have missed every shot and free throw (TS% = 0) and improved his rating by +0.4 points per possession. The point being: AI is an extreme outlier in this dimension (and in USG%) and his relative rank is accordingly sensitive to the specification of the regression.

DSMok1 published the excel spreadsheet where all calculations are done. Here: https://docs.google.com/open?id=0Bx1NfC ... g1NGZjNmQz

Instead of discussing MP's effect theoretically, I decided to see what happens practically. According to data in that sheet Chris Paul played 60% of available minutes and got an ASPM rating of 7.07. I increased his MP% enormously by 33% to 80% without touching anything else like TS% and in the end his ASPM rating was increased by 0.66 and settled at 7.73. Try it yourself and you'll have the same result. Conclusion of the matter is, MP cannot be a real issue for not just top players but anyone...

Second, there is perhaps an interesting conversation to be had contrasting the fact that ASPM and "new" RAPM between 2008 and 2011 yielded identical PPG error though the former had higher R2, but the question remains: why should one think that the latter method - what has no particular bone to pick with AI (or any other player) and what directly incorporates on-court performance - would rank him decisively lower than the former? And might it be because the former gives positive weight to MPG, beyond what theory might suggest is justified?

If I remember right new RAPM or xRAPM uses box-score too and gives more positive weight to minutes total. Never mind. I found it.

MP -0.04221383 0.1703430321

The point is ASPM was developed to reduce the noise of RAPM and stabilize it. That's why there are some big differences with RAPM and in fact there should be. It's good that there aren't very surprising names at the top of ASPM rankings.

[schtevie]

I take (and was aware of) the point that plausible alterations of the MPG variable value in the ASPM regression (or "new" RAPM prior) wouldn't make a tremendous difference in the resulting estimates for any given player (though I have the distinct impression that AI is relatively favored by the specification of the regression in Daniel's work, given the nature of how he filled up the stat sheet) and a notional adjustment of a fraction of a point per 100 possessions isn't going to significantly reshuffle any given player's HOF rank.

The underlying issue (relating to the topic of this thread) though is what approach to rating individual players, in particular elite players (because, let's be honest, by our conversations, it is revealed that nobody cares about notionally non-elite players) is more likely to offer "true" results. Is it necessarily estimates based on a system that best explains overall variation? I don't think such a claim can be supported.

I neglected to reply to this remark:

You didn't have any problem with an invalid (but reliable) metric such as WS or too noisy RAPM aka "Nick Collison metric" "Amir Johnson metric" goes on an on.

but will now, as it is on point, and I should have said that this couldn't be a more complete misrepresentation of my views (as expressed in this forum).

A case in point. I believe I was the first person to express dissatisfaction with one particular and conspicuous aspect of initial RAPM results. The tremendous improvement in explanatory power? Great! You want to use this instead of APM to retro/pre-dict team results? But of course. The implication, however, that the greatest player seasons were no longer about +10 and a bit above (as suggested by multi-year APM) but only contributed, as I recall, less than half that? Not so great!

Then came the first round (I think I'm getting the history right) of xRAPM. Tremendous improvement! The yearly (+/-) data could speak for itself with only relatively minor intervention (the box-score prior) and for the topic at hand (the relative contributions of the elite as a whole) a bit of sanity was restored.

And then came successive amendments of xRAPM; I think I count three. Modifications for: the observed fact that players player harder when behind and slack off when ahead, aging effects, and coaching effects. In terms of generally improving the regression results, there is no doubt. In terms of improving the valuation of the NBA elite, a mixed bag, but I suspect an overall step for the better.

Might the generally observed "countercyclical effort" not describe the league's best players? Perhaps that is one factor that tends to make the great great? Whatever the case, as I recall, the effect wasn't that large anyway. Adding the aging curve? Again, kind of revealed to be small potatoes, but unquestionably a step in the right direction, as no player can escape time. Modifying player estimates to take into account notional effects of coaching impact? In theory, a very good idea, but, in practice, I don't think the +/- data are able to speak correctly. Do the coaching estimates provided improve the explanatory power of the regression? Yes. Do they improve our perception of the relative value of the NBA elite? Poor Kevin Durant.

So, to summarize, the question is this: If one is interested in the performance of a particular player, an elite player, for whom there are years and years of data of thousands and thousands of possessions, and one is generally persuaded as to the superiority of +/- approaches (as any right-thinking person should be) why would one not prioritize estimates that are directly (and serially) informed by these actual on-court contributions rather than a synthetic approximation, based on box-score regression, no matter if "on average" it proves better league-wide?

[permaximum]

If one is interested in the performance of a particular player, an elite player, for whom there are years and years of data of thousands and thousands of possessions, and one is generally persuaded as to the superiority of +/- approaches (as any right-thinking person should be) why would one not prioritize estimates that are directly (and serially) informed by these actual on-court contributions rather than a synthetic approximation, based on box-score regression, no matter if "on average" it proves better league-wide?

Because no data is enough to reduce RAPM's noise. Not even 1996-2014 RAPM. That's why J.E. used multiple years with different weights at first, then introduced previous years' RAPM as priors, then introduced box-score as prior, then introduced box-score into the equation itself, then aging adjustments, then coach adjustments. There will be home-court adjustments if there aren't already, periodic minute-possession adjustments and even more. To make it more stable and increse the predictive power he'll do anything. In the end RAPM's going to be noisy still but the metric would include too much bias and lose it's validity if it hasn't lost it already. Simply 15-year play span of an average NBA player is not enough for RAPM..

If he can beat Vegas odds +52% of the time with all that predictive power, he can be a rich man eventually. But the question is can he?

Until the time that someone can prove the current iteration of RAPM or WS or any metric has better explanatory power than ASPM without losing a great deal of predictive power I'll continue to use it to rank players if people don't care about my eyes and brain's opinion.

[schtevie]

Well, if one's belief is that the sole criterion for appraising individuals, Allen Iverson and his ilk, is to select a ranking system that is expected to lose money in Vegas, betting on game outcomes, slightly more slowly than another, there isn't really much to discuss. In for a penny, in for a pound.

But as a perhaps final comment, it might be instructive to introduce some summary statistics of the actual Allen Iverson comparison, between xRAPM and ASPM, for both Offense and Defense. If for no other reason it turns out the focus of the conversation has been partially misleading.

As it turns out, the higher valuation of AI that ASPM provides is approximately half owing to offense and half to defense. Looking at the years 2001 to 2010 (his last), the (straight) average yearly premium that ASPM offers on offense and defense is 0.73 and 0.84, respectively. And if you ignore the last two years of his career, when he was in decline, the defensive contribution is relatively more important: 0.75 and 1.16.

So, given it's relative importance, the first question is about defense: Is DASPM, with the paucity of relevant defensive box score stats, considered a more reliable metric than DxRAPM? And second question: Does the implied story of the the DASPM premium ring true? And for this we need to look at the levels, for context.

The DxRAPM post-2000 career average is -.64 (or -.625 ignoring the final two years). By contrast, DASPM suggests that AI's defensive contributions were actually positive, with the average being 0.20 (or 0.54). So, which of these estimates rings more true, knowing what else we know about his career?

Well, if Eli Witus' positional APM averages are a guide, and assuming that AI was a point guard on defense (given his stature), the baseline is -2.0. So, in both schemes, AI is represented as an above-average defender - what his high STL% is consistent with. But in ASPM-land, he must be truly elite. And that is a bit surprising, (much) less because he was never recognized with Defensive awards (to my knowledge) but more because, given his central role in offense and how many minutes he played, that he would have the energy to achieve such defensive heights. Oh, and in only two years, 2008 and 2010, does the DxRAPM estimate exceed that of DASPM.

On to offense. The OxRAPM post-2000 career average is 2.53 (or 3.31 without the final decline) vs. 3.26 (or 4.06). But what is especially notable is that the OASPM premium appears every year. No exceptions. So there appears to be something systematic going on. But what could that be? Well, we know that AI was among the highest of high usage guys if not the highest; 5 of the top 20 seasons, post 1981, are his. And we "know" that theoretically there is reason to believe that the usage vs. efficiency curve slopes downward at the extreme.

So, given this larger perspective, which set of estimates would it be more prudent to believe?

[Crow]

Permaximim cited the first two parts of the sport skeptic series but not the third http://sportskeptic.wordpress.com/2012/ ... ect-blend/
A metric blend beat any single one at explanation or prediction. As I have said before, that is my main takeaway. I have yet to hear a team executive provide this answer publicly to the debated question or even acknowledge that it is an available option (though one now inside analyst agreed with me that blending was useful / powerful just before he went inside). The non-perfect can be a useful element.

I asked somewhere if Alex or someone else might produce player level data of these blends but this step remains undone at the precise mixes, to my knowledge. Perhaps will get around to it later. ( I used my own guessed blend as major part of my approach to winning the team win projection here twice in past.)

As I have noted before the concept of a blend of distinct metrics goes all the way back to Dan Rosenbaum's early articles. That concept was not followed in the public writings published at 82games.com. I previously wondered if blends were used by the Cavs privately. I assume they were to some extent. It took years for JE to bring us to hybrid metrics. Is a true blend still the better path? I'd say probably given the issues that pure box score, spm and RAPM each have. A blend still has inherent errors in construction but apparently less from the sport skeptic results on average as a blend than any one standing alone. Jon Nichols advocated metric blending just before he went inside, but I recall it was a simple, equal weight blending whereas I thought some metrics deserved more. My understanding is that he is now with Cavs, so perhaps that organization still believes in this approach even as personnel changes.

Pure box score metrics assigns a full amount of credit but only to box score elements and from a rigid per play value perspective. SPM shifts those weights around but again within a limited or very limited set of stat elements, leaving non box score actions uncredited and their "fair shares" hung onto the spm elements. RAPM measures all, though imperfectly. RAPM factors assigns out that credit and is useful to the extent you accept that 4 factors capture the essentials of the game (it is still a simplification).

It would corrupt Alex's best fit metric, but if you removed the old and new wins produced and used a blend of say around 45% ASPM, 30% RAPM and 25% winshares you might have a blended metric result worth looking at and refining to adjust for the wins produced removal (assuming you wanted to do that). What would Iverson's blended number be? Does it put him close to#100 best instead of "#20" or 180th? (need to be clear whether we are measuring for all-time or just for the period where rapm and conceivably aspm is available. should be the later given data availability issues) How does that feel to people, the two proponents of choosing one metric over another and everyone else? Who are his closest neighbors in this blended metric data set? Does seeing that make it look / feel / smell alright? In the big picture 30% for RAPM and the rest from spm and a version of pure box score is not that different from the share pure apm got in Dan's original blend 12 years ago (bottom of article http://www.82games.com/comm30.htm )

At least has posted a blended metric (3 part I think) but I don't immediately recall the location or the blend mix. I'll look but does anyone else recall? (update: there is a plan to use a blend here viewtopic.php?f=2&t=8633&start=30 but I might have seen another blend somewhere) (update I did see an article by nbacouchside with a similar blend to the one just noted and different than the blend I was suggesting. Might be same person? Might be yet another blend out there.)
Yep. http://hoopdon.weebly.com/articles/a-bl ... -nba-teams. http://hoopdon.weebly.com/articles/the- ... nd-ratings xrapm got a 22% share here producing these top results http://hoopdon.weebly.com/blend-player-ratings.html Then a move back to hybrid? http://hoopdon.weebly.com/articles/2014 ... s-rapm-spm
James Brocato of shut up and ham and now a consultant with the Mavs also produced a box score informed RAPM and his own box score metric. They could be compared or combined. As much as teams in general seem to dismiss one metrics, it appears some teams have the resources and presumably at least some interest in them.

[Crow]

Still no playoff only rapm but there is this source for playoff only spm.

http://shutupandjam.net/nba-ncaa-stats/ ... ed-impact/

Iverson had 2-3 +3 playoffs, according to this metric.