Chris Broussard – of ESPN the Magazine and ESPN.com – offered a profile of Danny Granger this past week with the following provocative title: What’s The Difference Between Danny Granger and Kobe Bryant?: Their stats suggest not much. Then again, some will tell you stats lie. Particularly when your team isn’t all that.
Despite the title, Broussard’s story barely mentions Kobe. What it does do is tell us how the stats suggest Granger is a really good NBA player.
Unfortunately – as often happens when sports writers start discussing stats – the actual statistical arguments are quite weak. The primary stat Broussard is scoring totals. Yes, he does mention factors like shooting efficiency and blocked shots. But it’s clear when you read the article that Broussard’s primary focus – as the following paragraph from the article indicates — is Granger’s scoring.
Being able to reconcile the space between a player’s stats and his team’s record is a big part of what separates championship-level talent evaluators from the recycled masses. All sorts of x factors come into play: Who is the guy scoring against? How does he get his points? Is he effective in crunch time? And the consensus on Granger, not just in front offices, but on sidelines and in locker rooms as well, is that he’s legit.
When we look at this paragraph it becomes clear that for Broussard, the proper word is not “stats” but “stat”. In other words, he is primarily interested in scoring. And scoring is captured for Broussard by one stat, the number of points credited to a player. When we move past this one stat, though, a very different story is told about Granger and Kobe.
Granger vs. Kobe
A couple of weeks ago I offered an entire column detailing how all the statistics – not just scoring totals – indicate Granger is not very different from an average NBA player. What was missing from the earlier post was a comparison between Granger and Kobe (it was missing since I had no idea someone would make such a comparison).
That is offered in Table One below.
Table One: Comparing Danny Granger and Kobe Bryant
If we focus just on scoring we see some similarities between the two players. Both take more than 30 shots from the field per 48 minutes and both are efficient.
When we move past scoring, though, we see clear differences. Relative to an average small forward, Granger is below average with respect to everything except blocked shots and assists. In contrast, Kobe is above average – relative to an average shooting guard – with respect to almost everything. Consequently when we look at the big picture – via Win Score – was see that Kobe is much more productive.
Win Score is easy to calculate but not so easy to interpret. In other words, we can see Kobe has a higher Win Score, but what do the differences mean in terms of wins? To answer that question we turn to Wins Produced.
Here is what Kobe produced in 2007-08 and 2008-09 (as of Saturday night):
28.7 Wins Produced, 0.237 WP48 [Wins Produced per 48 minutes].
Granger plays both small forward and power forward. If he played strictly at small forward, though, this is what he would have produced the past two seasons:
13.0 Wins Produced, 0.126 WP48 [Wins Produced per 48 minutes].
If we consider the time Granger spent at power forward as well, these numbers decline to 9.7 Wins Produced and a 0.094 WP48.
When we look at these numbers we see that Kobe is more than twice as productive as Granger. In sum, Broussard’s title is more than a bit misleading. If we focus strictly on scoring – which is primarily the approach Broussard took – then Granger and Kobe are similar. But that conclusion could be thought of as “lying with statistics.” When we consider all the stats, the differences between these two players are substantial.
Using Stats
Often the story in sports is the stats. After all, every contest in basketball is decided by just two stats (points scored and surrendered). And these stats are also determined by a collection of other numbers. So the numbers in basketball matter.
Not to pick on Broussard (since this argument applies to most sportswriters), but one gets the sense in reading Broussard that he has never spent much time learning how to analyze statistics. Consequently his “analysis” of the “stat” has problems.
Before I get to the problem, let me emphasize again that what I am saying about Broussard applies to sports writers in general. Sports writers often tell me that they have no training in formal statistical analysis. Consequently they have trouble knowing what story the stats are actually telling.
And I would add, the lack of formal training also hampers the ability of sports writers to evaluate statistical models. Frequently the results of a model are simply matched to what the sports writer previously believed. If there is a match (as we see with something like PERs), then the model is “good.” If there isn’t a match (as often happens with Wins Produced, Adjusted Plus-Minus, or Dean Oliver’s work) then the sports writer calls for more research.
This approach to research, though, is incorrect. Research is not done to confirm what we already believe. Research is done to teach us something new. Of course that doesn’t mean that all “new research” is “good.” In fact, much of what I see on-line with respect to sports would never be considered “good” by a peer-reviewed academic journal.
But we don’t reach the conclusion of “good” or “bad” by checking the results of the research against our prior beliefs. No, “good” or “bad” is determined by looking at how the research was done and interpreted.
Applying that standard to the work of Broussard we can locate the problems with his analysis. Again, he essentially confined his story to one statistic. In other words, his informal model – if I can be formal for a moment – was mis-specified. Points scored are not the only factor to consider in evaluating player performance. And when you ignore the other factors – or weight these other factors incorrectly – your analysis will be flawed. In sum, despite Broussard’s analysis, Kobe is really much more productive than Danny Granger.
Now if we compared Kobe to Lebron, or Chris Paul, or Dwight Howard, or… okay, that story can be told again another day.
- DJ
The WoW Journal Comments Policy
Our research on the NBA was summarized HERE.
The Technical Notes at wagesofwins.com provides substantially more information on the published research behind Wins Produced and Win Score
Wins Produced, Win Score, and PAWSmin are also discussed in the following posts:
Simple Models of Player Performance
What Wins Produced Says and What It Does Not Say
Introducing PAWSmin — and a Defense of Box Score Statistics
Finally, A Guide to Evaluating Models contains useful hints on how to interpret and evaluate statistical models.
25 responses so far ↓
simon // March 29, 2009 at 12:43 pm
The link to the “Different Answers — Same Conclusions” article isn’t quite right.
But yes, the validity of “laugh test” or “eyeball test” to evaluate statistical models has always fascinated me. It’s also partially because the +/- system has produced some very, umm, “interesting” results – remember when Hedo Turkoglu was crowned the most productive player according to the Winval system? Now he definitely was a solid player but that result was so far away from other evaluation system as well being inconsistent with Hedo’s consequent performances that many stats bashers used that as an example of how useless using stats is.
dberri // March 29, 2009 at 12:53 pm
Thanks Simon. I think I fixed the link.
I forgot to mention the “laugh test” again. As far as I know, that is not a real statistical test. It is used by people who don’t like the results of statistical analysis and are either a) too lazy, or b) don’t know how to refute a statistical argument.
Lior // March 29, 2009 at 3:33 pm
“Most people use statistics the way a drunk uses a lamp-post: for support rather than illumination” (attributed to Andrew Lang and A.E. Housman, among others)
simon // March 29, 2009 at 4:41 pm
Lior, but I think we do need to differentiate between using “stats” and using a model. Using statistics in the manner you’ve quoted usually means someone cherry-picking a limited number of specific statistical data to support a preconceived notion without any evidence. Such practice is quite different from establishing a model and using that to evaluate what’s happening in the real world.
Evan // March 29, 2009 at 9:32 pm
Bring on the KOBE lovers!
Anon // March 29, 2009 at 10:46 pm
no one ever says anything when DJ says kobe is better than someone, especially someone like danny granger, who is obviously an inferior player no matter how you look at it.
Good post though.
Adam // March 29, 2009 at 10:49 pm
When the author suggested that “there is not much difference between Granger and Bryant”, he probably meant “…compared to the average NBA player”. If we change the basis of comparison to “Granger vs SF” and “Bryant vs SG”, we probably should expect a different result!
thepeaceblogger // March 30, 2009 at 6:36 am
“Most people use statistics the way a drunk uses a lamp-post: for support rather than illumination” (attributed to Andrew Lang and A.E. Housman, among others)
@Lior
Like this quote which say it all….
jamerchant // March 30, 2009 at 9:37 am
There’s a bit of statistical legerdemain going on in this article, too, unfortunately. As Adam said, the thing that really makes Kobe Bryant and Danny Granger’s WP differ is that Bryant is designated as a SG and Granger as a SF. That distinction makes a huge difference in how WP is calculated. As I understand it, WP is calculated in a raw form, and then the average WP for the player’s position is subtracted from the raw number. SGs tend to have lower raw WP than SFs, so even if Granger and Bryant had identical raw scores, the adjustment would make Bryant’s score look better.
I think it’s an open question whether the traditional position designations are at all meaningful. Granger may play the position “small forward” and Bryant may play “shooting guard”, but if you watch them play, they both occupy a fairly similar role on their respective teams: perimeter player who takes a lot of jump shots. The most common Pacer lineup this year (according to 82games) was Granger, Marquis Daniels, Troy Murphy, TJ Ford, and Rasho Nesterovic. Who was the small forward there? Granger? No. Marquis Daniels played the more prototypical small forward role. Who was the power forward? Granger? No — Murphy. So… why should we compare Granger to small forwards or power forwards?
By a common standard, Bryant and Granger _are_ pretty similar. Bryant is probably still better (it would be nice to see their raw scores), due to his higher assist, steal, and shooting efficiency numbers, but Granger has the edge in rebounding and blocks (his block numbers are pretty outstanding for someone on the perimeter). Also, crucially, Granger missed something like 11 games with an injury this year, while Bryant has missed 0 (zero!) games so far. Of course Granger’s total WP will be lower.
For an article that criticizes others for misusing statistics, this one does plenty of obfuscation of its own.
Jason E // March 30, 2009 at 10:01 am
jamerchant, Dave’s table presents Granger and Bryant’s raw per 48 ‘win score’ without position adjustment. In terms of win score, Bryant is performing better before any position correction. And the correction difference between SF and SG is not all that large. Far more pronounced is the difference between PFs and SFs.
I agree though that the position correction can lead to some rather subjective decision making on classifications. Most teams spend a considerable amount of time with players playing ‘out of position’ when they eschew a big man in favor of a lineup with 4 g/sf types. I’m still struggling on how this should be dealt with, but it’s a question for research rather than an attempt to rationalize after the fact.
mrparker // March 30, 2009 at 10:41 am
jamerchant,
My argument has always been that wp48 has a concept of versatility “coded” into it. Usually the high wp48 guys(sg,sf,pf) are guys who can play more than one position well if their team needs it(Garnett, Duncan, Lebron, Kobe, etc). Whomever said sf and sg are about the same position is exactly right. Neither position is loaded with legendary players like the other 3 spots. They are both wing players, but a great sf would be able to play some pf as well while a great sg would be able to play some point guard as well. Sean Marion is one of the rare backwards guys who is built like a sf but is a much more effective pf though he is an above average sf. Someone like Lebron who can conceivably play all 4 well is truly exceptional. This also makes Magic stand out in my eyes because its conceivable that he could have done the same.
My point is that Granger has a limited skill set while Kobe does everything well. Granger is not as useful a player to his team because he can only help(be above average) in one position.
jamerchant // March 30, 2009 at 10:52 am
Thanks for the clarification Jason E — I was confusing WinScore with Wins Produced, and didn’t catch the “these numbers are per 48 minutes” caveat on his chart. Forget what I said about Granger playing fewer games/minutes (all apologies,Prof. Berri). It’s interesting, though, that by WinScore, Bryant and Granger are fairly close — 10.9 to 9.3.
Oren // March 30, 2009 at 11:15 am
“The most common Pacer lineup this year (according to 82games) was Granger, Marquis Daniels, Troy Murphy, TJ Ford, and Rasho Nesterovic. Who was the small forward there? Granger? No. Marquis Daniels played the more prototypical small forward role. ”
According to 82games, Daniels mostly played the Shooting Guard and not the Small Forward role. If this is incorrect, then that’s an error in the data and not an error in the statistics.
brgulker // March 30, 2009 at 1:07 pm
Dr. Berri,
I’ve been eating up your posts since I found your blog, and I’ve noticed a trend: you don’t agree with sports writers.
I wonder, if you were to develop something like a college curriculum for a sports journalist, what would it look like?
Which courses in statistics would you require and recommend?
At the end of the day, it might be a losing battle, because so much of sports journalism is purely hype, shock value, and entertainment, but I’d be interested to hear what you have to say nonetheless.
Moreover, it might give those of us without statistical training but who do enjoy your work a bit of a baseline understanding of the analysis going on in your work.
Jason E // March 30, 2009 at 2:00 pm
Oren, I’ve looked at the 82games ‘position assessment’ and cannot find explicitly how they determine it, save that it appears that they assume a 5-position lineup at all times. In reality, when a team “goes small” (e.g. plays a center and 4 wings) it’s not really a case of the PF spot being occupied by anyone. And whether someone is playing SG or SF is somewhat artificial in itself. What duties are clearly separate such that this classification need always be made?
I’d love a better explanation of how they assess it, but for now I’m skeptical but interested.
dberri // March 30, 2009 at 2:25 pm
brgulker,
I teach sports economics and spend about two weeks explaining the basics of regression analysis. In these two weeks they learn about standard errors, r-squared, residuals, and other basic stuff. After these two weeks I think most of my students could spot where mistakes are made by on-line analysts. Certainly my students can spot when people mis-interpret standard errors or use residuals incorrectly to forecast and evaluate models.
brgulker // March 31, 2009 at 7:08 am
Dr. Berri,
I wish I could take your class! By chance, does your university offer it online?
Oren // March 31, 2009 at 8:44 am
Jason,
It seemed to me that jamerchant argued that Prof Berri misused statistics since Granger shouldn’t really be considered a SF/PF but rather a SG/SF. I was noting that even if this is the case, the fact remains that the data used claims that Granger is a SF/PF. If there is an error, it is in the data and not in the analysis.
I would think that assuming a five position lineup at all times is the way to go for a model that attempts to indicate how good a team is. After all, a team of all average players that plays two SG and no SF will be worse then average. This is assuming that the average SG statistics don’t improve when there are two SGs and no SF is in the lineup.
I would think it is Indiana’s fault for not playing Granger at SG(provided he truly is most productive at SG) and playing him “out of position” at SF/PF. If Granger is most productive at SG, then from a model’s theoretical perspective, I would think that’s where he should play.
As for comparing individual players, I think you have a good point. After all, it isn’t Granger’s fault he’s being played out of position(presuming that he can maintain the same stats while being considered a shooting guard).
It seems to me that you would need different criteria to analyze individual players(perhaps getting rid of positions altogether and using descriptions such as point guards, wing players and bigs) then you would to analyze players as part of a team.
brgulker // March 31, 2009 at 9:01 am
Concerning the debate about traditional positions:
If these positions are conceived of as prescriptive of how a player must play in order to be successful at that position, then I think it’s highly problematic.
Speaking of Granger specifically, I think that is how I would describe my concern. When compared to other PF’s, he’s not a good rebounder — at least if PF is a prescriptive indicator of how a PF should play in order to be successful.
However, if they are conceived of as descriptive, then they still might have some value in the conversation. In other words, PG’s tend to get the most assists, SG’s tend to take the most outside jumpshots, PF’s and C’s tend to get the most rebounds, etc.
I think people have been right to argue that Granger’s productivity as a SF is quite good; it’s only when he plays at PF do we see a notable decline.
This is where I think interpretation is critical. One could argue that Granger himself is to blame; in other words, the decline in productivity is a direct result of him not being a productive player.
By contrast, I tend to think that Granger’s game is better suited to the traditional, descriptive SF position, rather than a PF position. When he plays at SF, his statistics show that. Consequently, I would attribute his overall decline in productivity not on him directly but rather on the coaching staff who insists on “going small” consistently.
In order to win games, which is what this column is all about, teams need to rebound, and history shows that big men are the best rebounders.
You can’t throw a guy like Granger at PF night in and night out and expect to win a bunch of games as a result. That’s on the coaching staff, not Granger.
Evan // March 31, 2009 at 11:57 am
brgulker wrote:
Moreover, it might give those of us without statistical training but who do enjoy your work a bit of a baseline understanding of the analysis going on in your work.
There are about 6 links at the bottom of every NBA post that explain Prof Berri’s model. If you haven’t read them, especially the technical notes and the guide to evaluating models, then you should.
Evan // March 31, 2009 at 11:58 am
Ooh, I guess the bold tag didn’t close. Sorry.
uhhh // March 31, 2009 at 2:57 pm
Putting down a flawed statistical analysis using another flawed statistical analysis is pretty stupid.
Evan // March 31, 2009 at 7:43 pm
Trolling on a blog isn’t all that productive either, uhhh.
Tom Mandel // April 2, 2009 at 9:10 am
“Bryant and Granger are fairly close — 10.9 to 9.3″ — that’s not so close!
Danny Granger is the Most Improved? « The Wages of Wins Journal // May 12, 2009 at 6:47 pm
[...] There is Not Much Difference Between Danny Granger and Kobe Bryant? [...]