Re: Winston-Sagarin 2003-04 Ratings
> One thing that I'd like to see is how stable their ratings are.I think DanR's stuff is close enough to say with a reasonable degree
of certainty, "not very".
> It's possible that Turkoglu's high 2003-04 rating is due more toAll very true, but if Sagarin and Winston make this distinction,
> luck than to his actual talent level. But over a 3-year period,
> does he still show up as superior to Duncan? Or, a related
> question: how often does Duncan rank less than #1 on the
> Spurs, according to Winston-Sagarin? One could imagine
> that sheer luck might make some lucky teammate often
> pop up as #1 while Duncan comes in at #2 or #3.
> That wouldn't mean that the Winston-Sagarin ratings should
> be consigned to the laugh test dumpster, but would mean
> that any one season's ratings should be taken with Shaq-
> sized grains of salt.
they haven't made it particularly clear to the media. All these
articles report one-year results, not multi-year results. It's
certainly possible that Sagarin and Winston are trying to hide the
good stuff, they're saving that for Cuban, but why expose themselves
to this kind of "laughter" then? Why not say nothing of the one-year
And this is something that's more important than whether we think
the results pass the laugh test. This is what the public sees of NBA
statistics, and if it doesn't pass their laugh test, neither will
other statistical analysis. I can't count how many blogs
had "Moneyball for the NBA?" posts about Winval after the Washington
Times story last spring.
- According to SI, Brian Cardinal was #11 by winval last year.
--- In APBR_analysis@yahoogroups.com, "thedawgsareout"
>Yes, Dan does very clearly point out that these are unstable from year
> > One thing that I'd like to see is how stable their ratings are.
> I think DanR's stuff is close enough to say with a reasonable degree
> of certainty, "not very".
>Following the theory that any publicity is good publicity, even if
> All very true, but if Sagarin and Winston make this distinction,
> they haven't made it particularly clear to the media. All these
> articles report one-year results, not multi-year results. It's
> certainly possible that Sagarin and Winston are trying to hide the
> good stuff, they're saving that for Cuban, but why expose themselves
> to this kind of "laughter" then? Why not say nothing of the one-year
it's people laughing at you. Note that Dallas has done worse since
starting to use winval. (I think they started using it in 2002,
though I think W-S developed it in 1999 and they might have given it
to Dallas before 2002.)
Author, Basketball on Paper
"Dean Oliver looks at basketball with a fresh perspective. If you
want a new way to analyze the game, this book is for you. You'll
never watch a game the same way again. We use his stuff and it helps
us." Yvan Kelly, Scout, Seattle Sonics
> Following the theory that any publicity is good publicity, even if
> it's people laughing at you. Note that Dallas has done worse since
> starting to use winval. (I think they started using it in 2002,
> though I think W-S developed it in 1999 and they might have given
> to Dallas before 2002.)If Dallas took it at all seriously, Turkoglu would be starting for
the Mavs this season, no?
> Following the theory that any publicity is good publicity, even ifAsk Mike Gimbel about that theory for me. :)
> it's people laughing at you.
> Note that Dallas has done worse since starting to use winval. (IThis old article from the _Indianapolis Star_ reprinted at APBR
> think they started using it in 2002, though I think W-S developed
> it in 1999 and they might have given it
> to Dallas before 2002.)
implies summer of 2001, at latest:
"Cuban, for one, took WINVAL's ratings into account when he adjusted
his roster last summer. He acquired Evan Eschmeyer, Tim Hardaway and
Danny Manning, all of whom had rated highly the previous season, and
released Howard Eisley, who rated poorly."
- I have mentioned this to a few folks, but I feel that the NBA
statistics community takes two steps backwards every time Winston &
Sagarin (W&S) open their mouths.
Interestingly, Mitchell Butler - W&S best player on the Wizards last
season - got cut from the Nuggets this week. Guess the Nuggets
aren't going to be buying W&S product anytime soon.
More and more, I have moved to preferring the statistical rating that
I have derived from the adjusted plus/minus statistics. It captures
practically everything except the noise in the adjusted plus/minus
My top 5 for last season was the following.
1. Kevin Garnett
2. Tim Duncan
3. Shaquille O'Neal
4. Andrei Kirilenko
5. Tracy McGrady
W&S do a horrible job of handling the noise that comes with the pure
adjusted plus/minus ratings. Here is my top five in my adjusted
plus/minus ratings (among those playing 2,000 or more minutes).
2. Kevin Garnett
3. Andrei Kirilenko
4. Jason Williams
5. Vince Carter
Note that the differences between my pure adjusted plus/minus ratings
and those from W&S is likely due mostly to differences in how we
handle garbage and crunch time weighting.
- The differences between these adjusted plus/minus ratings and a
really good statistics-based rating are not stable over time. In
fact, using 175 observations from players who played 1,000 or more
minutes in both 2002-03 and 2003-04, the correlation between this
difference in the two years is 0.08146 with a p-value of 0.2839. In
other words, there is no strong evidence to suggest that the
differences are anything other than random error, i.e. luck.
But remember my statistical index is calibrated to predict adjusted
plus/minus ratings. It does a good job of rating players like Bruce
Bowen and Trent Hassell, players who contribute mostly on the
defensive end. The differences between adjusted plus/minus ratings
and other statistical ratings probably would have a positive and
In my adjusted plus/minus ratings, Turkoglu and Ginobili are rated
lower than Duncan in both 2002-03 and 2003-04. But Turkoglu was
much, much better in 2003-04 than in 2002-03. Ginobili also improved
and Duncan was about the same in both seasons.
Over time I think the great power of these adjusted plus/minus
ratings will be to tell us how to weigh the various statistics, in
particular new statistics that we may get from the charting projects
that Roland at www.82games.com is undertaking.
- --- In APBR_analysis@yahoogroups.com, "dan_t_rosenbaum"
the statistical rating that
> I have derived from the adjusted plus/minus statistics. Itcaptures
> practically everything except the noise in the adjusted plus/minusLooks pretty good -- especially relative to some that we have seen.
> My top 5 for last season was the following.
> 1. Kevin Garnett
> 2. Tim Duncan
> 3. Shaquille O'Neal
> 4. Andrei Kirilenko
> 5. Tracy McGrady
>.. Here is my top five in my adjustedYou say you prefer the results of the list above, and I have to
> plus/minus ratings (among those playing 2,000 or more minutes).
> 1. Nene
> 2. Kevin Garnett
> 3. Andrei Kirilenko
> 4. Jason Williams
> 5. Vince Carter
agree. So how exactly do you account for the extreme difference
here? This (2nd list) is basically a random mix of elite and
Here's a short list of players you have ranked at least 33% higher
than I did (mine used only last year's stats, players with 1500+
Most of these guys "need" some defensive recognition that I don't
grant them. Others (Cardinal, Bradley) receive less credit in my
system, which shrinks production for low-minutes players.
Now here are players I've apparently "overvalued" by some 40%,
relative to you.
Some one-dimensional guys, scorers, and journeymen. No defensive
I think you are on the right track.